Hoenstly this sounds like the best android SoC in years..everything just falls in place.
And shows just how much Samsung is in a bad timing place with their Note series..technically ther Exynos 9820 could've been ready by October but waiting for Snapdragon 855 would take too long..so they have to settle for the "old" 10nm SoC's that will get spanked by Huawei.
Huawei announcing this doesn't mean you can buy it tomorrow. It will probably be in phones at the end of this year or more likely start of next year. By that time, Note 9 will be half a year old, Qcom will have the 855 launch and Huawei won't have any headstart. All of the SoC providers are gated by 7nm process.
This indeed looks like a great performer, and I like how Huawei shares a lot of numbers. They seem too good to be true, so I am still a bit skeptical. Yet I don't necessarily see anything wrong with what they have shared from their testing methods and comparisons. Except for the GPU turbo that suddenly fixes their GPU performance, this really seems something fishy, so looking forward into the planned article for that feature. Huawei seems to mix up decent performance numbers and marketing nonsense in their announcements, but at least that makes up for an interesting exercise for the reader. Also, let's hope they have put a lot of effort into their scheduler. Using triple cluster systems makes this problem already a lot harder, and we see that Samsung has messed this part up too.
Very rude to a reasonable question, especially when the article doesn't mention Galileo support. It only mentions L1 and L5 support from the GPS satellites.
Sounds like scheduling will make or break this one. Their own WiFi solution adds a risk too- ofc best case scenario, it would be an upside. Does it support uMCP memory solutions?
Looks like TSMC 7nm density is looking good... Similar transistor count to GP104 for 1/3 the size. I can't wait to see what this means for 7nm amd and nvidia gpus.
Well, there's an assumption there about the die size being very close to 100mm^2, rather than, say, 90mm^2. Hopefully an accurate measurement can be made soon enough.
And this is real world transistor density, not perfect theoretical transistor density. The A76 cores, for example, will likely be made with bigger transistors to get extra performance, which lowers density.
What's the use when the Huawei locks down bootloader and pump bloat without any means to remove. Also no US availability even if its available, the Pro versions without jacks, QHD panels or SD slots are favored.
To add VLC banned them due to aggressive background killing except for their stupid bloat a.k.a iOS clone EMUI.
Better to use an Exynos with BL unlocked + Custom ROM community to get a full control and streamlined optimized UX.
Plus comparing a 7nm + New cortex chip to old is hilarious lol, Let's see how it stacks up with A11/12 GPU perf or even the Adreno.
To add a subjective opinion, these all benchmarks can gtfo vs real world usage, having used Sultan XDA kernel + ROM I can't think any developer who can match that UX in smoothness + perf with custom written drivers and tweaks with hardcoded underclocking. This stupid Kirin BS can barely get to that level of QComms CAF and it's large community.
Wow, so calling out the elephant in the room and their glaring flaws means troll. Thanks
The benchmark part that i said is not to dismiss your articles in anyway, perhaps I worded it wrong. I actually meant the way OEMs use them to manipulate like Samsung 9810 did.
What elephant? 99.99% of users DON'T about bootloader. SD card and 3.5mm is done and dusted, none of the popular has them besides Samsung. Their declining sales speaks volumes - most people don't care.
You are always gonna choose Samsung anyway, why bother ranting here?
I agree. Some people just like to pretend that “features” that are only good for an almost unmeasurable fraction, are of great importance. They’re not. The truth is that no manufacturer cares about that tiny fraction other than manufacturers who themselves only have tiny sales, and so have to appeal to that tiny fraction for a large part of theirs.
Other than who have tiny sales ? Go and speak about that with Xiaomi India and OnePlus in Indian market which is the world's fastest growing and highest market for Smartphones. And all of these OEMs phones have full unlockable bootloaders.
Just because you ceded the control of the devices you own and how Apple and Samsung rule the world doesn't mean that's generalized. There are still people who care to have control over what they own. If the 9810 was locked down, What would have Andrei done ? Dump the Samsung brand and move over. End of the story, he's an XDA developer and contributed much enough. What if the option didn't exist ?
A shame that people thinking this way. Soon you'll be using a thin client on your smartphone and run the OS over their servers and bend to them. Don't worry they got you covered, since 99.9% people are fine, that's what going to happen. GL
Other than who have tiny sales ? Go and speak about that with Xiaomi India and OnePlus in Indian market which is the world's fastest growing and highest market for Smartphones. And all of these OEMs phones have full unlockable bootloaders.
Just because you ceded the control of the devices you own and how Apple and Samsung rule the world doesn't mean that's generalized. There are still people who care to have control over what they own. If the 9810 was locked down, What would have Andrei done ? Dump the Samsung brand and move over. End of the story, he's an XDA developer and contributed much enough. What if the option didn't exist ?
A shame that people thinking this way. Soon you'll be using a thin client on your smartphone and run the OS over their servers and bend to them. Don't worry they got you covered, since 99.9% people are fine, that's what going to happen. GL
The phones are consumer devices, aimed at...consumers. Companies are right in locking down bootloaders when they know that an extreme minority will end up causing more headaches for them. OnePlus is still a niche player, aiming at a captive audience. Xiaomi is increasingly raising the wait time for their bootloader unlocks.
Developers and enthusiasts alike should be aiming at developer devices. The Android One program has made unlocked devices available at the lower-end. If all phones are shipped with locked bootloaders, then move your development and tinkering to development boards. You can choose the devices to have control over rather than expecting the OEM to cede control.
The sad thing is that, given his talents, Andrei is unlikely to stay at AnandTech forever, if much longer at all. I really love these SoC pieces. Thanks. I hope Apple's A12 makes it as well.
Kirin 980, and supposedly the next QC flagship, might just be what we hoped E9810'd be, and with a GPU that could give Adreno a run for its money. Fingers crossed.
The most likely reason he came back is that Imagination fired him (lucky for us) after Apple dropped them and they had to make staff reductions. I hope his talents go unnoticed for as long as possible, but when the time comes he will leave and not look back (most of what he receives in comments seems to be hate).
Their loss, our gain. Your SoC articles are a real highlight of this site, at least IMO. They are one of the reasons why I look at this site frequently.
yeeeeman - show some respect. I am sure customers don't come to your place of employment and tell you that you aren't any good at your job.....or maybe they do?
Yes the SoC articles are really fantastic. Most unique in the whole journalism. I'm also really thankful to Andrei for his top grade deep dive analysis. Not to forget the 9810 parts esp the tuning article he published and his tweaks to that SoC, that helped a lot with the s9 community. Plus his thoughts in comments also the A11 snips from the same. Its an honor tbh to be able to have access to them.
Coming to the second point I don't think Adreno can be beaten so easily. Samsung LSIs GPU division just scaled up recently I don't think we can see one from them next year.
And on a side note, this SoC can't be customized or tuned due to Huawei stance on Bootloader lock and no. CAF/Exynos level commitment.
Thanks for the article, Andrei! The Kirin 980 looks like a definite step up from the 970, the one fly in the spec-ointment is the somewhat puzzling inability to encode 4K at 60 frames/second; the 980 should have the raw horsepower to do so. Hopefully they have a high bitrate 1080p60 mode with both EIS and OIS enabled (and got rid of EIS artifacts).
@Andrei: I have a question is about availability: Given that Apple has pretty much a right of first refusal regarding TSMC's 7 nm fab capacities, has Huawei confirmed that the 980 is in volume production now or that a large production run has already completed? I am curious because now that GloFo is out at 7 nm, many companies out there state "we'll do 7 nm with TSMC", but the numbers just don't add up. Did Huawei mention expected sales numbers for their 980-bearing flagships for Q4 2018 and for 2019? Did they make sense to you in light of TSMC's obligation to Apple?
I wouldn't be concerned about volume as Huawei's product lineup is pretty staggered and vastly lower volume compared to Apple, the Mate 10 to date "only" shipped 10 million units. I don't expect the Mate 20 to suddenly be that more successful that they run into volume issues, but who knows.
Thanks Andrei! The next company where that question will be acute for is Qualcomm; AFAIK, they typically ship several times the volume of their snapdragon flagship SoCs compared to the top-dog Kirin, and QC wanted to go TSMC exclusievly for 7 nm. Might be interesting, given that Apple and QC have a serious beef with each other right now, so Apple could squeeze QC at the fab level.
Apple is estimated to be between 20-25% of TSMCs business, so likely, QC wouldn’t have a problem. And companies don’t order hundreds of million of dollars of parts out of schedule to hinder a rival.
Actually, the numbers are more the other way around. eetimes published some estimates (from multiple sources) that Apple is taking around 75% - 80% of TSMC's 7 nm capacity, with Qualcomm, Huawei, AMD and everybody else getting whatever is left over. Let's not forget that Apple has what is basically a first right of refusal for TSMC's 7 nm capacity, which was a key part of them giving TSMC the exclusive for their A12 and possible others going forward. As for not ordering large inventory just to hinder a rival: Why wouldn't they (Apple)? 1. Apple and QC are arguing over billions of dollars in license fees, and 2. SoCs are definitely ordered in bulk once they are past sampling stage (comes with how they are made). Of course, I am not privy to Apple's strategy vs. QC, but Apple is known to play hardball and even dirty if that helps them win. Although, in this case, I don't exactly feel sorry for QC, either.
Isn't that capacity for 2018 only? The only products that will ship in volume in 2018 will be the A12 and Kirin 980. TSMC will expand capacity in 2019 and will start volume production of their 7nm+ in Q2-Q3. Also Apple needs a lot of chips up front (tens of millions) huawei need 10-20 million a year and I believe qualcom needs a number similar to apple (even if there are more android sales than iphone only a tiny fraction of those have flagship socs and an significant part of those are Kirin and Exynos) so with the usual release of flagship android phones in Q2-Q3 2019 there should be no problem.
The 75-80% for Apple is for 2018, AFAIK. Huawei will likely be okay, while they ship a lot of phones, they don't/won't sell most of them with 980s inside, and had apparently reserved ahead of time. The situation is different for QC: a lot of "flagship" Android phones (almost all bound for the US) are equipped with the respective top snapdragon, so their volume requirements are bigger. The eetimes analysis suggests that QC will get about the same sliver of TSMC's 7 nm capacity as Huawei, but QC will want a lot more chips that that, and they didn't want to wait for Samsung to come online. They may not have much choice now, though. Also, while 2019 might well bring a more relaxed supply situation, if you're hoping to ship your new flagship phones in 2018, you'll need those SoCs to start coming in about now. The 7 nm capacity crunch has become a game of musical chairs, and now the music has stopped as GloFo out of it altogether, so there will be some fabless customers left standing. The main one is likely AMD, I believe it's unlikely we'll see 7 nm VEGAs on 7 nm anytime soon. With QC, I wouldn't be surprised if a number of their execs are flying to South Korea a lot these days.
Highly unlikely to near impossible. Apple would not know the total capacity TSMC had for 7nm, and how will TSMC handle their 7nm expansion if Apple decide to take over whatever capacity they had, which has nothing to do with Apple. Not to mention Qualcomm second source with Samsung incase anything happens to TSMC.
That assumes that the A12 will be in all the new phones released this year. I believe that they are sitting on large amounts of A11 chips so they might show up in the LCD model as a differentiator?
Do you really think that this will be the first smartphone SoC shipped in a product? It seems that Apple will have a good claim on that. Or do you just not want to mention a non Android, non AOSP product? While we won’t know for certain which cores they will use, it’s not important, because whichever they choose will be highly modified anyway.
Hmm... "though both pairs of A76’s are the same IP, this mid-pair are very likely to actually be more efficient when running at the same frequency" I don't think so. It depends on power rails implemented but it's still both more costly and uter stupid to implement two areas of them for four core's instead implementing one with cuple more rails. Future on it's also utterly stupid to implement mid tier CPU core's along with developing (property) scheduler for them instead just making a hand brake based on relative utilisation (in two points; optimal & sustainable frequency). Come on now? 178% more power efficient. What it sips 3W into battery when pushed hard. 100% less than something is zero. Two huge not optimized for application use NNPU's. That must be a new record in black silicone on any mobile SoC. I am still for the DSP's that are also optimised for NNPU task's. At least that way we would see a benefit in the multimedia capabilities.
Another rather bad design regarding me. At least how it looks like this time around they will improve regarding GPU capabilities and power consumption.
I have no idea what you're trying to say regarding the A76s. The two pairs are on their own voltage rails. If you're TDP limited in a quad-design, then clocking down a pair of them and putting them on a separate rail is only beneficial.
178% better perf/W, it's a growing figure. The estimates check out.
NPUs have their place as dedicated silicon, you need to be able to run imaging and inferences alongside each other. That can't be done with just one DSP. The Kirin 970 already had a Cadence DSP of similar capability of what you see in Snapdragons - alongside the NPU. The 980 won't be any different.
I was under the assumption that Cadance had not implemented floating point support (fixed point only) or SMT in their DSPs. They have very performant DSPs, I don't question that..but not quite to the level of the QDSP6v6. Last I checked they were falling short of QDSP6v5 after they introduced DMT. Unfortunately Cadence doesn't rely on BDTI so reliable benchmarks are hard to come by.
I am very up to date in regards to the DSPs that immediately "matter" to me Hexagon 400-600 series and also many CEVA DSPs. Admittedly it can be difficult to keep up with all of them. I do see they added support as of 2017. It seems the C5, Q6 and P6 are geared towards vision and AI applications, but doesn't seem to be a multithreaded architecture the way that Hexagon is. So I agree you would need a NPU if you were using one of their DSPs, the same is not true of Hexagon
I am trying to say about mid pair of the A76's how they are just another waist of silicon & will slow things down additionally. I am for just a pair of big cores. It's smarter to put only a pair of them add more power rails & limit their up scaling based on high utilisation on let's say 1.6GHz & 2GHz. You save on silicon & they don't have penalty of another migration, you also save on DTP they use as you know they use a lot even while only idling. Would be good if we could also limit SMP tasks that can use more than two core's only on smal in order one's & trow in more of those (small core's).
I figured out for what 178% figure stands for but I have hard time believing it all together how G76 will be 78% faster per/W. 70% is my best educated guess & still per cluster. But we will see that only when we get silicone.
Now look at it this way. NNPU on the Kirin 970 whose huge chuck of silicone that we didn't use to often & it whose really bad design (with out ¼ precision). It looks like they didn't do anything regarding improving design and they are now pushing two of those. Really? For me that is as bad as it can get. I never mentioned neither only one nor MP1 DSP's nor most modern ADSP's are single threaded. I am not against NN just against specialised single purpose accelerators on the SoC's with both limited DTP & price point. As we simply can't put in big enough FPGA (or EFPGA) to be useful I think program-able DSP adopted for NN are best match. Putting even two MP4 Tensilica DSP's would eat lot less silicone while being much more useful & still good enough for NN tasks. We will talk more when you do a Mate review.
Andrei, it's mainly because of that DSP capability similar to Hexagon etc. that I don't get why they still don't have 4Kp60 video encoding. Isn't all that imaging muscle plus the fast memory bus basically screaming for it? Did anybody ask what keeps the 980 from 4Kp60 recording? Is it the software?
Yes, but isn't it surprising that the video encoding block was "left out" amid all the other improvements? That's what I meant with my question (should have asked it better, I guess). Is this Huawei's take on what's important or not for flagship phones in 2018/19? Is so, I believe they are quite wrong. It does matter; to me (and many others), the ability to record 4kp60 in decent quality is a key feature that may well decide which phone I spend > 750 dollars or Euros on. Won't be a Huawei now this time around.
The greatest technology matters nothing, when the politics behind it are wrong!
These chips are put into the most personal computers people have, but the loyalty of these digital servants or brain extensions then lie with the manufacturer, Google and quite possible with the government of either the manufacturer or Google instead of the *owner*.
You are 'sold' a computer, you can neither trust nor control nor fully make your own or use as you see fit.
If you see my argument a few a pages back, you'd know how people are against it and supporting the corporate, because of the majority of the people not knowing or caring about, we are at loss. And it will be increasing exponentially. Android for example, with the version 9, a.k.a Pie. The application API targeting lockdown is going on, like banning hidden API techniques, Undocumented APIs, Drawing over other apps. And the battery stats from Nougat -> Oreo and forth is downgraded as well, Google's answer is the security issue causing to read the package level Mah data apparently seems posing as a threat. And the API targeting of the market for the updating apps stays at 1 level less than the current API now that P is launched all the app updates should target O's not just 8.0 it must be 8.1 something like that. I don't know how hard they enforce this.
Not to mention the SafetyNet. And The A/B partition system to throw a wrench causing unified partitions for Kernel/Recovery. But development has been going good for now.
People don't care since they use social media at max forget the tuning and tweaking the OS. With as a service models this goes further away from hand, the Win10 and also their walled garden love, similar to that of Google's proprietary SystemUI in Pixel to the latest background/recent switcher exclusive to Pixel Launcher, but they opened the latter. Also not to forget how they abandoned the Browser/Messaging apps. LineageOS takes it further and adds security and features. And yet people blast the tinkerers/enthusiasts.
I wonder how much ELV is actually used in the prodution of such a chip?
From what I understand you'd still try to use multi-patterning for everything say along the x-axis (cheapest), multi-masking (rather expensive) and a bit of z-axis in a way to compensate for the structure stretched out from the patterning and then ELV (horribly expensive) for the essential y-axis precision...
But I guess that's already far to 2D oriented and simplyfied =:-O
Did any consumer software end up using the NPU on the 970? I remember Huawei made a big deal about it, but I don't remember it being used for much. Their "AI" camera stuff wasn't all that impressive in practice.
lol exactly what I said on another site. The NPU is basically useless at the moment and this most likely will not change for the effective life of this product, that's why Qualcomm's solution is more sensible while this is more a marketing stunt and waste of silicon while cutting corners elsewhere like in GPU.
if this is expected jump for other product or at least similar (read cpu, gpu, memory, flash etc) then Intel as a basic example will be in massive trouble when AMD launches 7nm for cpu-gpu on 7nm well ahead of Intel, hopefully the cost will actually end up being that much less costly and much higher volume then they managed to do with 16/14/12nm the past 2 years.
really tired of the massive price gouge BS from ram makers as well as the price increase being aimed directly at miners "chewing up the already constrained supplies"
anyways, likely to be a VERY interesting 4Q2018 and 1-2Q2019 that is for darn sure.
Yup, kudos to Andrei for a nice and informative article, and to the HiSilicon designers for a job seemingly well done. It will be very interesting to get real world performance data from independent tests.
We've yet to see when Huawei will actually ship, last year the 970 launched weeks before A11 and shipped weeks later. They literally *stole* the spotlight and the title of the first SoC with NPU. Huawei and their marketing stunts. smh
Both Huawei and Samsung has a modem better than Qualcomm for now, until Qualcomm ship their Snapdragon 855 with X24. And Intel isn't anywhere close to those three..... just why Apple still have Intel is quite beyond me at this moment.
And Huawei has done something I long thought Apple would do first, its own WiFI Chipset. Huawei ships lots of ONT and world number one in 4G CPE shipment, lots of other M2M devices in China, so their WiFi volume, in 100s of millions unit per year make a lot of sense to design and Fab their own WiFi. Apple ships 300M+ unit with WiFi ( iPad, iPhone, Mac ) and yet they have not made such a move ( yet ).
For now, Huawei is looking like a much stronger competitor to Apple than Samsung ever was.
The question here is: Are there mobile (cell) networks out there that can actually take advantage of those theoretical maximum transmission rates? Aide from that, Huawei is one of the big global players in wireless and wired networking technology, so them coming up with their own modem makes a lot of sense. I wonder if IP issues will keep their newest creation out of the US for now. AFAIK, that's a key reason why Samsung still uses QC's SoCs in their Galaxy devices sold over here.
I look forward to seeing sustained performance for devices with the Kirin 980, particularly in gaming and emulation, and how it compares to SD845 and the 835 of the Moto Z3. Motorola claims they have the Adreno 540 clocked at 850 mhz in the Z3, afaik that is the first time it has been clocked higher than 710mhz in a shipping device. This is quiet an interesting year in the mid and high tiers
Andrei I have a quesiton about the density of this chip. If we assume the chip is 95mm² and it packs 6.9B transistors that works out to a density of 76.2MTr/mm² This is quite a bit lower than Intels claimed 100MTr/mm² of their 10nl process, I though TSMCs 7nm was supposed to be equal to or ahead of Intel?
We don't know the die size yet. However don't expect Intel to get anywhere 100mt/mm^2 - the process that will be released will be more like 12nm. And typical densities are significantly lower than claimed, for example Intel's own marketing admits that real TSMC 20nm chips have better density than Intel 14nm.
@Wilco1: Actual vs. theoretical transistor density has been the subject of much discussion for years, if not decades. Your statement about Intel admitting that their 14 nm has lower density than TSMC's 20 nm got me curious. Could you provide a link to a source for that? Would appreciate it - Thanks!
Thanks for the reference, appreciate it! I looked at the slides, and Intel did say that their 22nm process was defintely below TSMC's 20 nm in transistor density, and their somewhat distorted graph suggests that their 14 nm process (in 2015) was about equal to TSMC's A8-based SoCs in transistor density. They also point out that the ARM SoCs have a large % of elements that are amenable to high density, whereas Intel's CPUs contain a large % of elements that are not (Intel's take on that, not mine). I guess the only true apples-to-apples comparison would be a comparison of either ARM SoCs to ARM SoCs from two different fabs, or x86/x64 CPUs of the same class. Would love to see some slides like this for EPYC vs. Xeon on current nodes. However, the take home message remains: As of 2018, Chipzilla has surrendered the crown of being the most advanced microprocessor manufacturer to TSMC, and Samsung is now 2nd, with Intel dropping to 3rd. I look forward to see how much improvement in power efficiency and speed the move to 7 nm will bring. If the upsides are compelling enough, Intel may regret sitting this one out.
Intel is still in the lead, though that may change depending on how TSMC delivers on 7nm next year, and whether Intel will sort out their yield issues. (Also next year.)
As of right now; Intels 10nm process has the highest density of any of their competitors, just like Intels 14nm process is well ahead of TSMCs and Samsung’s 14/16nm process.
The chart referred to in the marketing materials is misleading. You can’t compare the density in a power sipping cellphone SOC with the density in a 90W computer CPU (or even 15W laptop CPU.)
(Just one reason why the result you get will be misleading: The density of SRAM cells is 3 times bigger than the density of logic cells. A cellphone SOC contains twice as many SRAM cells than an Intel CPU.)
Nonsense. Intel has lost its lead a long time ago. TSMC 10nm beats Intel 14nm by a huge margin (and 14++nm by even more), and 7nm chips will be on the market in a few weeks.
That comparison is between actual chips aimed at similar markets, same number of cores, on-chip GPU, similar TDP, similar performance, so it's a much better comparison than theoretical densities.
As of right now there is no Intel 10nm process that is in volume production, we have to wait until next year. But then TSMC will be on their 2nd generation 7nm process.
Look, if you don’t understand why the comparison is meaningless, then I’m not going to waste my time trying to explain a fanboy the difference between a process meant for mobile chips, and one meant for HP desktop CPUs.
The wackiest thing here isn’t just your fanboy hate of Intel (Find a sports team to cheer for, seriously!) It’s the fact that you’re claiming that TSMCs process is more sense than Intels, when anyone who reads the chart titled “Normalized for composition” can see that it’s clearly not the case, and that Intels 14nm process is significantly more dense than both TSMCs and Samsung’s.
Congratulations you're the best ever Intel fanboy! If you don't understand why the "normalized decomposition" is pure marketing bullshit hen you're even dumber than it seems.
Of course you have no idea whether Intels 10nm process will be closer to 12nm or not.
And your claim that TSMC 20nm chips are more dense than Intels 14nm chips is likewise wrong.
Intels marketing makes no such claim, in fact they make quite the opposite claim: That Intels 14nm handily beats both TSMC and Samsung’s comparable process in density.
The actual densities achieved in real chips prove what a process can achieve. And Intel's own slides clearly show that 20nm has better density than Intel 14nm. You can bash your bible as much as you like but that's not going to change hard facts.
Just look at the “normalized for composition” graph to see Intels advantage.
And the fact that you either don’t know or don’t understand why the other graph is meaningless, shows that you don’t really understand the subject: Just another fanboy who treats CPU architectures like rival soccerclubs.
Reading this makes me want to trash my p20 pro :) Usually I buy the best hw I can afford but I find ridiculous spending 1359 euro for an Iphone x, and in this case I went straight for the camera(s). Couldn't be happier. In my consumer perspective, the 980 won't be a reason to upgrade, but a waterproof phone which I can use to take picture underwater would :=)
ARM has to look at the chips for supercomputers where Intel sells its 8 billion transistors Xeon chips for $2000 while this 8 billion transistors ARM marvel go just for $20. Since most supercomputers are memory bandwidth bound anyway the ARM will easily find its place there
Look up the Fujitsu's next ARM supercomputer..it will be the first exascale and the fastest in the world.
Btw your comparison with Intel is a perfect illustration of just how disgustingly overrated Intel is. Their chips have nothing to do with high tech any more..it's just a monopoly built on borderline criminal practices.
"..The A64FX is made up of 8,786 million transistors and will be manufacturered to a 7nm process. Fujitsu claims that it will be the first chip to take advantage of ARM's Scalable Vector Extensions instruction set, developed specifically for high-performance computing. Fujitsu claims that it could also be used for artificial intelligence applications.
Sporting 48 cores with two-four assistant cores to assist with processes such as input/ouput, jitter reduction and asynchronous MPI [multiple protocol interface]. For floating-point calculations, Fujitsu claims the A64FX will be able to achieve 2.7 teraflops for 64-bit (FP64) operations, more than 5.4 teraflops for 32-bit (FP32) operations, and more than 10.8 teraflops for 16-bit (FP16) operations."
it would be instructive to see an instruction profile of a xeon machine in 'supercomputer' use?? RISC came to be just because lots o folks were convinced that all those high-cycle instructions weren't used much or that useful when they were. could it be that ARM architecture is sufficient for this use case??
If you’re curious about the topic of Intel Xeons in supercomputers, you could always contact some of the many universities/institutions that are running a Xeon based supercomputer.
There’s certainly no lack of them.
Out of the ten most powerful supercomputers in the world as of June, three were using Xeon CPUs and 2 were using Xeon Phi.
If you take a look at the Supercomputing Green Top 500, EIGHT of the top ten supercomputers were Xeon machines.
You may want to give these people a call and tell them they’re doing it all wrong, and need to put some ARM processors in their supercomputers instead because RISC.
If you take a look at the ten most powerful supercomputers in the world, three are using Xeon CPUs and two are running on POWER9 processors.
On the Green Top500, Top 10 consists of 8 Xeon and 2 POWER9 supercomputers.
But you’re basically saying, that all the thousands of people who have built these machines, people who have studied or taught computer science for decades are doing it all wrong, yes?
They just need to throw some 20$ cellphone CPUs in their machines instead to really kick it up a notch.
All top performers in the Top500 are clusters and even if some of them use Xeons the peak performance they got comes from the use of GPU or Phi or IBM. No, people are not stupid as seen by broad use of Intel chips in this list. But the explanation of this is because just few short years back the 64 bit ARM or GPU did not exist which precluded any their serious use in supercomputing.
Besides in many cases supercomputers are memory bandwidth bound so any small advantage of some processors over the others becomes insignificant. Xeons are good processors, they just don't deserve anymore almost two orders their production cost.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
120 Comments
Back to Article
darkich - Friday, August 31, 2018 - link
Hoenstly this sounds like the best android SoC in years..everything just falls in place.And shows just how much Samsung is in a bad timing place with their Note series..technically ther Exynos 9820 could've been ready by October but waiting for Snapdragon 855 would take too long..so they have to settle for the "old" 10nm SoC's that will get spanked by Huawei.
yeeeeman - Friday, August 31, 2018 - link
Huawei announcing this doesn't mean you can buy it tomorrow. It will probably be in phones at the end of this year or more likely start of next year. By that time, Note 9 will be half a year old, Qcom will have the 855 launch and Huawei won't have any headstart. All of the SoC providers are gated by 7nm process.Ian Cutress - Friday, August 31, 2018 - link
Launch on October 16th with Mate 20vladx - Friday, August 31, 2018 - link
And most likely December for the Honor Magic 2 with the same Kirin 980.yeeeeman - Friday, August 31, 2018 - link
Ok then, it sounds very good. If this social holds true to its promises it might push Huawei to new heightsabufrejoval - Friday, August 31, 2018 - link
Kirin 960 based tablets launching now, Kirin 970 based HiKey Linaro developer boards just available...They must be sitting on big piles of old chips, but I'd still like to see the newest and best in those other form factors, including a notebook.
Ian Cutress - Friday, August 31, 2018 - link
Or they want to control their latest parts tightly in their premium product linesZolaIII - Friday, August 31, 2018 - link
It's not the worst one but still on the lower part of those from a design stand point.tmnvnbl - Friday, August 31, 2018 - link
This indeed looks like a great performer, and I like how Huawei shares a lot of numbers. They seem too good to be true, so I am still a bit skeptical. Yet I don't necessarily see anything wrong with what they have shared from their testing methods and comparisons.Except for the GPU turbo that suddenly fixes their GPU performance, this really seems something fishy, so looking forward into the planned article for that feature. Huawei seems to mix up decent performance numbers and marketing nonsense in their announcements, but at least that makes up for an interesting exercise for the reader.
Also, let's hope they have put a lot of effort into their scheduler. Using triple cluster systems makes this problem already a lot harder, and we see that Samsung has messed this part up too.
SydneyBlue120d - Friday, August 31, 2018 - link
Really curious to know if there is L5 support for both GPS and Galileo.Ian Cutress - Friday, August 31, 2018 - link
Says right there in the articleSydneyBlue120d - Friday, August 31, 2018 - link
In the article and image only GPS is shown, and the Kirin 970 didn't support Galileo AFAIK :)Correct me if I'm wrong.
levizx - Friday, August 31, 2018 - link
You're wrong.SydneyBlue120d - Saturday, September 1, 2018 - link
Ok, can you share a screenshot of GPS test beta showing Galileo is being used with a Kyrin 970 then?Thanks a lot!
negusp - Sunday, September 2, 2018 - link
You're a dense idiot. Read the article.psychobriggsy - Monday, September 3, 2018 - link
Very rude to a reasonable question, especially when the article doesn't mention Galileo support. It only mentions L1 and L5 support from the GPS satellites.Still, L5 is welcome.
shabby - Friday, August 31, 2018 - link
Finally more then 30gb/sec of bandwidth, seems we were stuck at 30 for eons.jjj - Friday, August 31, 2018 - link
Sounds like scheduling will make or break this one.Their own WiFi solution adds a risk too- ofc best case scenario, it would be an upside.
Does it support uMCP memory solutions?
Dr. Swag - Friday, August 31, 2018 - link
Looks like TSMC 7nm density is looking good... Similar transistor count to GP104 for 1/3 the size. I can't wait to see what this means for 7nm amd and nvidia gpus.shabby - Friday, August 31, 2018 - link
This means Nvidia gpus will start at $999...darkich - Monday, September 3, 2018 - link
1/3 the size??I think it's even far less than that actually.
Santoval - Monday, September 3, 2018 - link
The density is quite less than I expected actually. Assuming a 98 to 99 mm^2 die size the transistor density is just ~70 million transistors per mm^2.psychobriggsy - Monday, September 3, 2018 - link
Well, there's an assumption there about the die size being very close to 100mm^2, rather than, say, 90mm^2. Hopefully an accurate measurement can be made soon enough.And this is real world transistor density, not perfect theoretical transistor density. The A76 cores, for example, will likely be made with bigger transistors to get extra performance, which lowers density.
Quantumz0d - Friday, August 31, 2018 - link
What's the use when the Huawei locks down bootloader and pump bloat without any means to remove. Also no US availability even if its available, the Pro versions without jacks, QHD panels or SD slots are favored.To add VLC banned them due to aggressive background killing except for their stupid bloat a.k.a iOS clone EMUI.
Better to use an Exynos with BL unlocked + Custom ROM community to get a full control and streamlined optimized UX.
Plus comparing a 7nm + New cortex chip to old is hilarious lol, Let's see how it stacks up with A11/12 GPU perf or even the Adreno.
To add a subjective opinion, these all benchmarks can gtfo vs real world usage, having used Sultan XDA kernel + ROM I can't think any developer who can match that UX in smoothness + perf with custom written drivers and tweaks with hardcoded underclocking. This stupid Kirin BS can barely get to that level of QComms CAF and it's large community.
centurio9 - Friday, August 31, 2018 - link
Yeah, none of that has anything to do with the SoC in question but OK...Andrei Frumusanu - Friday, August 31, 2018 - link
Ignore the trolls.Quantumz0d - Friday, August 31, 2018 - link
Wow, so calling out the elephant in the room and their glaring flaws means troll. ThanksThe benchmark part that i said is not to dismiss your articles in anyway, perhaps I worded it wrong. I actually meant the way OEMs use them to manipulate like Samsung 9810 did.
levizx - Friday, August 31, 2018 - link
What elephant? 99.99% of users DON'T about bootloader. SD card and 3.5mm is done and dusted, none of the popular has them besides Samsung. Their declining sales speaks volumes - most people don't care.You are always gonna choose Samsung anyway, why bother ranting here?
melgross - Friday, August 31, 2018 - link
I agree. Some people just like to pretend that “features” that are only good for an almost unmeasurable fraction, are of great importance. They’re not. The truth is that no manufacturer cares about that tiny fraction other than manufacturers who themselves only have tiny sales, and so have to appeal to that tiny fraction for a large part of theirs.Quantumz0d - Sunday, September 2, 2018 - link
Other than who have tiny sales ? Go and speak about that with Xiaomi India and OnePlus in Indian market which is the world's fastest growing and highest market for Smartphones. And all of these OEMs phones have full unlockable bootloaders.Just because you ceded the control of the devices you own and how Apple and Samsung rule the world doesn't mean that's generalized. There are still people who care to have control over what they own. If the 9810 was locked down, What would have Andrei done ? Dump the Samsung brand and move over. End of the story, he's an XDA developer and contributed much enough. What if the option didn't exist ?
A shame that people thinking this way. Soon you'll be using a thin client on your smartphone and run the OS over their servers and bend to them. Don't worry they got you covered, since 99.9% people are fine, that's what going to happen. GL
Quantumz0d - Sunday, September 2, 2018 - link
Other than who have tiny sales ? Go and speak about that with Xiaomi India and OnePlus in Indian market which is the world's fastest growing and highest market for Smartphones. And all of these OEMs phones have full unlockable bootloaders.Just because you ceded the control of the devices you own and how Apple and Samsung rule the world doesn't mean that's generalized. There are still people who care to have control over what they own. If the 9810 was locked down, What would have Andrei done ? Dump the Samsung brand and move over. End of the story, he's an XDA developer and contributed much enough. What if the option didn't exist ?
A shame that people thinking this way. Soon you'll be using a thin client on your smartphone and run the OS over their servers and bend to them. Don't worry they got you covered, since 99.9% people are fine, that's what going to happen. GL
kn0w1 - Sunday, September 2, 2018 - link
The phones are consumer devices, aimed at...consumers. Companies are right in locking down bootloaders when they know that an extreme minority will end up causing more headaches for them. OnePlus is still a niche player, aiming at a captive audience. Xiaomi is increasingly raising the wait time for their bootloader unlocks.Developers and enthusiasts alike should be aiming at developer devices. The Android One program has made unlocked devices available at the lower-end. If all phones are shipped with locked bootloaders, then move your development and tinkering to development boards. You can choose the devices to have control over rather than expecting the OEM to cede control.
Eris_Floralia - Friday, August 31, 2018 - link
They used NBA2K18 and PUBG for GPU comparison:NBA2K18: avg FPS 59.3 for Kirin 980, and 48.5 for S845.
Kirin 980: 43.7mW/frame, S845: 64.5mW/frame
PUBG: avg FPS 58.8 (980) vs 54.1 (S845)
38.7mW/frame (980) vs 45.2mW/frame (S845)
Eris_Floralia - Friday, August 31, 2018 - link
Shoudn't take these numbers too seriously thoughyhselp - Friday, August 31, 2018 - link
The sad thing is that, given his talents, Andrei is unlikely to stay at AnandTech forever, if much longer at all. I really love these SoC pieces. Thanks. I hope Apple's A12 makes it as well.Kirin 980, and supposedly the next QC flagship, might just be what we hoped E9810'd be, and with a GPU that could give Adreno a run for its money. Fingers crossed.
Amandtec - Friday, August 31, 2018 - link
Assuming talented people seek only filthy lucre.porcupineLTD - Friday, August 31, 2018 - link
The most likely reason he came back is that Imagination fired him (lucky for us) after Apple dropped them and they had to make staff reductions. I hope his talents go unnoticed for as long as possible, but when the time comes he will leave and not look back (most of what he receives in comments seems to be hate).Andrei Frumusanu - Friday, August 31, 2018 - link
I resigned IMG because I did not like the life and prospects in the UK.porcupineLTD - Friday, August 31, 2018 - link
Glad to be wrong.Total Meltdowner - Friday, August 31, 2018 - link
Keep it up Andre, I really enjoy your articles here.jospoortvliet - Saturday, September 1, 2018 - link
Same here.GreenReaper - Friday, August 31, 2018 - link
AnandTech is now owned by a UK PLC. How long before Brexit's cruel bite takes its toll?RaduR - Saturday, September 1, 2018 - link
Aha , so I knew that Andrei is the same person writing IMG blogs. I still miss that company they made good products . Never taken mobile seriously .Still my Dreambox receiver and Atheros router still survives with MIPS CPU :)
eastcoast_pete - Saturday, September 1, 2018 - link
Their loss, our gain. Your SoC articles are a real highlight of this site, at least IMO. They are one of the reasons why I look at this site frequently.yeeeeman - Sunday, September 2, 2018 - link
I agree. He is the only writer left on this site who actually has the same depth and passion as Anand.yeeeeman - Sunday, September 2, 2018 - link
Bravo Andrei! Te sustinem 😉Speedfriend - Monday, September 3, 2018 - link
yeeeeman - show some respect. I am sure customers don't come to your place of employment and tell you that you aren't any good at your job.....or maybe they do?Quantumz0d - Friday, August 31, 2018 - link
Yes the SoC articles are really fantastic. Most unique in the whole journalism. I'm also really thankful to Andrei for his top grade deep dive analysis. Not to forget the 9810 parts esp the tuning article he published and his tweaks to that SoC, that helped a lot with the s9 community. Plus his thoughts in comments also the A11 snips from the same. Its an honor tbh to be able to have access to them.Coming to the second point I don't think Adreno can be beaten so easily. Samsung LSIs GPU division just scaled up recently I don't think we can see one from them next year.
And on a side note, this SoC can't be customized or tuned due to Huawei stance on Bootloader lock and no. CAF/Exynos level commitment.
eastcoast_pete - Friday, August 31, 2018 - link
Thanks for the article, Andrei! The Kirin 980 looks like a definite step up from the 970, the one fly in the spec-ointment is the somewhat puzzling inability to encode 4K at 60 frames/second; the 980 should have the raw horsepower to do so. Hopefully they have a high bitrate 1080p60 mode with both EIS and OIS enabled (and got rid of EIS artifacts).@Andrei: I have a question is about availability: Given that Apple has pretty much a right of first refusal regarding TSMC's 7 nm fab capacities, has Huawei confirmed that the 980 is in volume production now or that a large production run has already completed? I am curious because now that GloFo is out at 7 nm, many companies out there state "we'll do 7 nm with TSMC", but the numbers just don't add up. Did Huawei mention expected sales numbers for their 980-bearing flagships for Q4 2018 and for 2019? Did they make sense to you in light of TSMC's obligation to Apple?
Andrei Frumusanu - Friday, August 31, 2018 - link
I wouldn't be concerned about volume as Huawei's product lineup is pretty staggered and vastly lower volume compared to Apple, the Mate 10 to date "only" shipped 10 million units. I don't expect the Mate 20 to suddenly be that more successful that they run into volume issues, but who knows.eastcoast_pete - Friday, August 31, 2018 - link
Thanks Andrei! The next company where that question will be acute for is Qualcomm; AFAIK, they typically ship several times the volume of their snapdragon flagship SoCs compared to the top-dog Kirin, and QC wanted to go TSMC exclusievly for 7 nm. Might be interesting, given that Apple and QC have a serious beef with each other right now, so Apple could squeeze QC at the fab level.eastcoast_pete - Friday, August 31, 2018 - link
missing last line: ..at the fab level by ordering some more A12s just when QC wants its chips to be made.melgross - Friday, August 31, 2018 - link
Apple is estimated to be between 20-25% of TSMCs business, so likely, QC wouldn’t have a problem. And companies don’t order hundreds of million of dollars of parts out of schedule to hinder a rival.eastcoast_pete - Friday, August 31, 2018 - link
Actually, the numbers are more the other way around. eetimes published some estimates (from multiple sources) that Apple is taking around 75% - 80% of TSMC's 7 nm capacity, with Qualcomm, Huawei, AMD and everybody else getting whatever is left over. Let's not forget that Apple has what is basically a first right of refusal for TSMC's 7 nm capacity, which was a key part of them giving TSMC the exclusive for their A12 and possible others going forward. As for not ordering large inventory just to hinder a rival: Why wouldn't they (Apple)? 1. Apple and QC are arguing over billions of dollars in license fees, and 2. SoCs are definitely ordered in bulk once they are past sampling stage (comes with how they are made). Of course, I am not privy to Apple's strategy vs. QC, but Apple is known to play hardball and even dirty if that helps them win. Although, in this case, I don't exactly feel sorry for QC, either.porcupineLTD - Saturday, September 1, 2018 - link
Isn't that capacity for 2018 only? The only products that will ship in volume in 2018 will be the A12 and Kirin 980. TSMC will expand capacity in 2019 and will start volume production of their 7nm+ in Q2-Q3. Also Apple needs a lot of chips up front (tens of millions) huawei need 10-20 million a year and I believe qualcom needs a number similar to apple (even if there are more android sales than iphone only a tiny fraction of those have flagship socs and an significant part of those are Kirin and Exynos) so with the usual release of flagship android phones in Q2-Q3 2019 there should be no problem.eastcoast_pete - Saturday, September 1, 2018 - link
The 75-80% for Apple is for 2018, AFAIK. Huawei will likely be okay, while they ship a lot of phones, they don't/won't sell most of them with 980s inside, and had apparently reserved ahead of time. The situation is different for QC: a lot of "flagship" Android phones (almost all bound for the US) are equipped with the respective top snapdragon, so their volume requirements are bigger. The eetimes analysis suggests that QC will get about the same sliver of TSMC's 7 nm capacity as Huawei, but QC will want a lot more chips that that, and they didn't want to wait for Samsung to come online. They may not have much choice now, though. Also, while 2019 might well bring a more relaxed supply situation, if you're hoping to ship your new flagship phones in 2018, you'll need those SoCs to start coming in about now.The 7 nm capacity crunch has become a game of musical chairs, and now the music has stopped as GloFo out of it altogether, so there will be some fabless customers left standing. The main one is likely AMD, I believe it's unlikely we'll see 7 nm VEGAs on 7 nm anytime soon. With QC, I wouldn't be surprised if a number of their execs are flying to South Korea a lot these days.
Speedfriend - Monday, September 3, 2018 - link
Given that Apple didn't order extra SOC test machines this year, the A12 is probably coming in smaller numbers than the A11....iwod - Saturday, September 1, 2018 - link
Highly unlikely to near impossible. Apple would not know the total capacity TSMC had for 7nm, and how will TSMC handle their 7nm expansion if Apple decide to take over whatever capacity they had, which has nothing to do with Apple. Not to mention Qualcomm second source with Samsung incase anything happens to TSMC.V900 - Monday, September 3, 2018 - link
Given that Apple has more or less paid for TSMCs 7nm fabs, it’s rather likely that they know exactly the capacity, actually.Speedfriend - Monday, September 3, 2018 - link
That assumes that the A12 will be in all the new phones released this year. I believe that they are sitting on large amounts of A11 chips so they might show up in the LCD model as a differentiator?melgross - Friday, August 31, 2018 - link
Do you really think that this will be the first smartphone SoC shipped in a product? It seems that Apple will have a good claim on that. Or do you just not want to mention a non Android, non AOSP product? While we won’t know for certain which cores they will use, it’s not important, because whichever they choose will be highly modified anyway.Achtung_BG - Friday, August 31, 2018 - link
Pls more info for Balong 5000 and 5G technology.Andrei Frumusanu - Friday, August 31, 2018 - link
I added a paragraph under the modem bit, unfortunately we don't have any more info on that part.centurio9 - Friday, August 31, 2018 - link
Can't wait to see it next year in some ~400$ Honor device :DZolaIII - Friday, August 31, 2018 - link
Hmm... "though both pairs of A76’s are the same IP, this mid-pair are very likely to actually be more efficient when running at the same frequency" I don't think so. It depends on power rails implemented but it's still both more costly and uter stupid to implement two areas of them for four core's instead implementing one with cuple more rails. Future on it's also utterly stupid to implement mid tier CPU core's along with developing (property) scheduler for them instead just making a hand brake based on relative utilisation (in two points; optimal & sustainable frequency).Come on now? 178% more power efficient. What it sips 3W into battery when pushed hard. 100% less than something is zero.
Two huge not optimized for application use NNPU's. That must be a new record in black silicone on any mobile SoC. I am still for the DSP's that are also optimised for NNPU task's. At least that way we would see a benefit in the multimedia capabilities.
Another rather bad design regarding me. At least how it looks like this time around they will improve regarding GPU capabilities and power consumption.
Andrei Frumusanu - Friday, August 31, 2018 - link
I have no idea what you're trying to say regarding the A76s. The two pairs are on their own voltage rails. If you're TDP limited in a quad-design, then clocking down a pair of them and putting them on a separate rail is only beneficial.178% better perf/W, it's a growing figure. The estimates check out.
NPUs have their place as dedicated silicon, you need to be able to run imaging and inferences alongside each other. That can't be done with just one DSP. The Kirin 970 already had a Cadence DSP of similar capability of what you see in Snapdragons - alongside the NPU. The 980 won't be any different.
Wardrive86 - Friday, August 31, 2018 - link
I was under the assumption that Cadance had not implemented floating point support (fixed point only) or SMT in their DSPs. They have very performant DSPs, I don't question that..but not quite to the level of the QDSP6v6. Last I checked they were falling short of QDSP6v5 after they introduced DMT. Unfortunately Cadence doesn't rely on BDTI so reliable benchmarks are hard to come by.ZolaIII - Friday, August 31, 2018 - link
Think you didn't bean following the DSP scene for a very long time.Wardrive86 - Friday, August 31, 2018 - link
I am very up to date in regards to the DSPs that immediately "matter" to me Hexagon 400-600 series and also many CEVA DSPs. Admittedly it can be difficult to keep up with all of them. I do see they added support as of 2017. It seems the C5, Q6 and P6 are geared towards vision and AI applications, but doesn't seem to be a multithreaded architecture the way that Hexagon is. So I agree you would need a NPU if you were using one of their DSPs, the same is not true of HexagonZolaIII - Friday, August 31, 2018 - link
I am trying to say about mid pair of the A76's how they are just another waist of silicon & will slow things down additionally. I am for just a pair of big cores. It's smarter to put only a pair of them add more power rails & limit their up scaling based on high utilisation on let's say 1.6GHz & 2GHz. You save on silicon & they don't have penalty of another migration, you also save on DTP they use as you know they use a lot even while only idling. Would be good if we could also limit SMP tasks that can use more than two core's only on smal in order one's & trow in more of those (small core's).I figured out for what 178% figure stands for but I have hard time believing it all together how G76 will be 78% faster per/W. 70% is my best educated guess & still per cluster. But we will see that only when we get silicone.
Now look at it this way. NNPU on the Kirin 970 whose huge chuck of silicone that we didn't use to often & it whose really bad design (with out ¼ precision). It looks like they didn't do anything regarding improving design and they are now pushing two of those. Really? For me that is as bad as it can get. I never mentioned neither only one nor MP1 DSP's nor most modern ADSP's are single threaded. I am not against NN just against specialised single purpose accelerators on the SoC's with both limited DTP & price point. As we simply can't put in big enough FPGA (or EFPGA) to be useful I think program-able DSP adopted for NN are best match. Putting even two MP4 Tensilica DSP's would eat lot less silicone while being much more useful & still good enough for NN tasks. We will talk more when you do a Mate review.
eastcoast_pete - Friday, August 31, 2018 - link
Andrei, it's mainly because of that DSP capability similar to Hexagon etc. that I don't get why they still don't have 4Kp60 video encoding. Isn't all that imaging muscle plus the fast memory bus basically screaming for it? Did anybody ask what keeps the 980 from 4Kp60 recording? Is it the software?Andrei Frumusanu - Saturday, September 1, 2018 - link
Video encoding is done by the video encoding block, not the DSP.eastcoast_pete - Saturday, September 1, 2018 - link
Yes, but isn't it surprising that the video encoding block was "left out" amid all the other improvements? That's what I meant with my question (should have asked it better, I guess). Is this Huawei's take on what's important or not for flagship phones in 2018/19? Is so, I believe they are quite wrong. It does matter; to me (and many others), the ability to record 4kp60 in decent quality is a key feature that may well decide which phone I spend > 750 dollars or Euros on. Won't be a Huawei now this time around.abufrejoval - Friday, August 31, 2018 - link
The greatest technology matters nothing, when the politics behind it are wrong!These chips are put into the most personal computers people have, but the loyalty of these digital servants or brain extensions then lie with the manufacturer, Google and quite possible with the government of either the manufacturer or Google instead of the *owner*.
You are 'sold' a computer, you can neither trust nor control nor fully make your own or use as you see fit.
Rooting is a right of mankind!
GreenReaper - Friday, August 31, 2018 - link
Like many human "rights", they only matter if you are willing to fight - and die - for them.Quantumz0d - Monday, September 3, 2018 - link
If you see my argument a few a pages back, you'd know how people are against it and supporting the corporate, because of the majority of the people not knowing or caring about, we are at loss. And it will be increasing exponentially. Android for example, with the version 9, a.k.a Pie. The application API targeting lockdown is going on, like banning hidden API techniques, Undocumented APIs, Drawing over other apps. And the battery stats from Nougat -> Oreo and forth is downgraded as well, Google's answer is the security issue causing to read the package level Mah data apparently seems posing as a threat. And the API targeting of the market for the updating apps stays at 1 level less than the current API now that P is launched all the app updates should target O's not just 8.0 it must be 8.1 something like that. I don't know how hard they enforce this.Not to mention the SafetyNet. And The A/B partition system to throw a wrench causing unified partitions for Kernel/Recovery. But development has been going good for now.
People don't care since they use social media at max forget the tuning and tweaking the OS. With as a service models this goes further away from hand, the Win10 and also their walled garden love, similar to that of Google's proprietary SystemUI in Pixel to the latest background/recent switcher exclusive to Pixel Launcher, but they opened the latter. Also not to forget how they abandoned the Browser/Messaging apps. LineageOS takes it further and adds security and features. And yet people blast the tinkerers/enthusiasts.
abufrejoval - Friday, August 31, 2018 - link
I wonder how much ELV is actually used in the prodution of such a chip?From what I understand you'd still try to use multi-patterning for everything say along the x-axis (cheapest), multi-masking (rather expensive) and a bit of z-axis in a way to compensate for the structure stretched out from the patterning and then ELV (horribly expensive) for the essential y-axis precision...
But I guess that's already far to 2D oriented and simplyfied =:-O
Pointers? Links?
cfenton - Friday, August 31, 2018 - link
Did any consumer software end up using the NPU on the 970? I remember Huawei made a big deal about it, but I don't remember it being used for much. Their "AI" camera stuff wasn't all that impressive in practice.s.yu - Sunday, September 2, 2018 - link
lol exactly what I said on another site. The NPU is basically useless at the moment and this most likely will not change for the effective life of this product, that's why Qualcomm's solution is more sensible while this is more a marketing stunt and waste of silicon while cutting corners elsewhere like in GPU.Dragonstongue - Friday, August 31, 2018 - link
if this is expected jump for other product or at least similar (read cpu, gpu, memory, flash etc) then Intel as a basic example will be in massive trouble when AMD launches 7nm for cpu-gpu on 7nm well ahead of Intel, hopefully the cost will actually end up being that much less costly and much higher volume then they managed to do with 16/14/12nm the past 2 years.really tired of the massive price gouge BS from ram makers as well as the price increase being aimed directly at miners "chewing up the already constrained supplies"
anyways, likely to be a VERY interesting 4Q2018 and 1-2Q2019 that is for darn sure.
Entropyq3 - Saturday, September 1, 2018 - link
Yup, kudos to Andrei for a nice and informative article, and to the HiSilicon designers for a job seemingly well done. It will be very interesting to get real world performance data from independent tests.zodiacfml - Saturday, September 1, 2018 - link
Enabled by the latest process node. I was surprised Huawei to launch this early as I thought Apple has all the 7nm output for the first few months.s.yu - Sunday, September 2, 2018 - link
We've yet to see when Huawei will actually ship, last year the 970 launched weeks before A11 and shipped weeks later. They literally *stole* the spotlight and the title of the first SoC with NPU. Huawei and their marketing stunts. smhCedarWind - Saturday, September 1, 2018 - link
If the Pixel 3 will use Snapdragon 845 it will be far behind any phone using the new ARM A76 coresiwod - Saturday, September 1, 2018 - link
Both Huawei and Samsung has a modem better than Qualcomm for now, until Qualcomm ship their Snapdragon 855 with X24. And Intel isn't anywhere close to those three..... just why Apple still have Intel is quite beyond me at this moment.And Huawei has done something I long thought Apple would do first, its own WiFI Chipset. Huawei ships lots of ONT and world number one in 4G CPE shipment, lots of other M2M devices in China, so their WiFi volume, in 100s of millions unit per year make a lot of sense to design and Fab their own WiFi. Apple ships 300M+ unit with WiFi ( iPad, iPhone, Mac ) and yet they have not made such a move ( yet ).
For now, Huawei is looking like a much stronger competitor to Apple than Samsung ever was.
RaduR - Saturday, September 1, 2018 - link
Only if they get rid of that custom UI...eastcoast_pete - Saturday, September 1, 2018 - link
The question here is: Are there mobile (cell) networks out there that can actually take advantage of those theoretical maximum transmission rates?Aide from that, Huawei is one of the big global players in wireless and wired networking technology, so them coming up with their own modem makes a lot of sense. I wonder if IP issues will keep their newest creation out of the US for now. AFAIK, that's a key reason why Samsung still uses QC's SoCs in their Galaxy devices sold over here.
Wardrive86 - Saturday, September 1, 2018 - link
I look forward to seeing sustained performance for devices with the Kirin 980, particularly in gaming and emulation, and how it compares to SD845 and the 835 of the Moto Z3. Motorola claims they have the Adreno 540 clocked at 850 mhz in the Z3, afaik that is the first time it has been clocked higher than 710mhz in a shipping device. This is quiet an interesting year in the mid and high tierstsk2k - Saturday, September 1, 2018 - link
Andrei I have a quesiton about the density of this chip.If we assume the chip is 95mm² and it packs 6.9B transistors that works out to a density of 76.2MTr/mm²
This is quite a bit lower than Intels claimed 100MTr/mm² of their 10nl process, I though TSMCs 7nm was supposed to be equal to or ahead of Intel?
tsk2k - Saturday, September 1, 2018 - link
Correction 72,6MTr/mm²Wilco1 - Saturday, September 1, 2018 - link
We don't know the die size yet. However don't expect Intel to get anywhere 100mt/mm^2 - the process that will be released will be more like 12nm. And typical densities are significantly lower than claimed, for example Intel's own marketing admits that real TSMC 20nm chips have better density than Intel 14nm.eastcoast_pete - Saturday, September 1, 2018 - link
@Wilco1: Actual vs. theoretical transistor density has been the subject of much discussion for years, if not decades. Your statement about Intel admitting that their 14 nm has lower density than TSMC's 20 nm got me curious. Could you provide a link to a source for that? Would appreciate it - Thanks!Wilco1 - Sunday, September 2, 2018 - link
See https://www.semiwiki.com/forum/files/2015_Investor... page 18.eastcoast_pete - Sunday, September 2, 2018 - link
Thanks for the reference, appreciate it! I looked at the slides, and Intel did say that their 22nm process was defintely below TSMC's 20 nm in transistor density, and their somewhat distorted graph suggests that their 14 nm process (in 2015) was about equal to TSMC's A8-based SoCs in transistor density. They also point out that the ARM SoCs have a large % of elements that are amenable to high density, whereas Intel's CPUs contain a large % of elements that are not (Intel's take on that, not mine). I guess the only true apples-to-apples comparison would be a comparison of either ARM SoCs to ARM SoCs from two different fabs, or x86/x64 CPUs of the same class. Would love to see some slides like this for EPYC vs. Xeon on current nodes.However, the take home message remains: As of 2018, Chipzilla has surrendered the crown of being the most advanced microprocessor manufacturer to TSMC, and Samsung is now 2nd, with Intel dropping to 3rd. I look forward to see how much improvement in power efficiency and speed the move to 7 nm will bring. If the upsides are compelling enough, Intel may regret sitting this one out.
V900 - Monday, September 3, 2018 - link
Nonsense.Intel is still in the lead, though that may change depending on how TSMC delivers on 7nm next year, and whether Intel will sort out their yield issues. (Also next year.)
As of right now; Intels 10nm process has the highest density of any of their competitors, just like Intels 14nm process is well ahead of TSMCs and Samsung’s 14/16nm process.
The chart referred to in the marketing materials is misleading. You can’t compare the density in a power sipping cellphone SOC with the density in a 90W computer CPU (or even 15W laptop CPU.)
(Just one reason why the result you get will be misleading: The density of SRAM cells is 3 times bigger than the density of logic cells. A cellphone SOC contains twice as many SRAM cells than an Intel CPU.)
s.yu - Monday, September 3, 2018 - link
Interesting!Wilco1 - Monday, September 3, 2018 - link
Nonsense. Intel has lost its lead a long time ago. TSMC 10nm beats Intel 14nm by a huge margin (and 14++nm by even more), and 7nm chips will be on the market in a few weeks.That comparison is between actual chips aimed at similar markets, same number of cores, on-chip GPU, similar TDP, similar performance, so it's a much better comparison than theoretical densities.
As of right now there is no Intel 10nm process that is in volume production, we have to wait until next year. But then TSMC will be on their 2nd generation 7nm process.
V900 - Monday, September 3, 2018 - link
LOLOLOL!Look, if you don’t understand why the comparison is meaningless, then I’m not going to waste my time trying to explain a fanboy the difference between a process meant for mobile chips, and one meant for HP desktop CPUs.
The wackiest thing here isn’t just your fanboy hate of Intel (Find a sports team to cheer for, seriously!)
It’s the fact that you’re claiming that TSMCs process is more sense than Intels, when anyone who reads the chart titled “Normalized for composition” can see that it’s clearly not the case, and that Intels 14nm process is significantly more dense than both TSMCs and Samsung’s.
Wilco1 - Monday, September 3, 2018 - link
Congratulations you're the best ever Intel fanboy! If you don't understand why the "normalized decomposition" is pure marketing bullshit hen you're even dumber than it seems.V900 - Monday, September 3, 2018 - link
Of course you have no idea whether Intels 10nm process will be closer to 12nm or not.And your claim that TSMC 20nm chips are more dense than Intels 14nm chips is likewise wrong.
Intels marketing makes no such claim, in fact they make quite the opposite claim: That Intels 14nm handily beats both TSMC and Samsung’s comparable process in density.
Wilco1 - Monday, September 3, 2018 - link
Given the 10nm issues, they will have to relax CPP/MPP far more than for the ++ processes: https://semiaccurate.com/2018/08/02/intel-guts-10n...The actual densities achieved in real chips prove what a process can achieve. And Intel's own slides clearly show that 20nm has better density than Intel 14nm. You can bash your bible as much as you like but that's not going to change hard facts.
V900 - Monday, September 3, 2018 - link
It showed no such thing. 😁Just look at the “normalized for composition” graph to see Intels advantage.
And the fact that you either don’t know or don’t understand why the other graph is meaningless, shows that you don’t really understand the subject:
Just another fanboy who treats CPU architectures like rival soccerclubs.
Wilco1 - Monday, September 3, 2018 - link
Look at page 18 - it shows the indisputable area advantage for TSMC and Samsung in real silicon.The rest is marketing bullshit, if you believe 30+% of a SoC is SRAM, I can give you a great price on this bridge I have for sale.
RaduR - Saturday, September 1, 2018 - link
Looking forward for a real life test . I'm afraid these A76 CPUs will throttle a lot and eat your battery fast .Battery life today is more important than CPU power . Also GPU is important as average user profile is : Whatsapp, Facebook , YouTube.
Also for business users that open a lot of pdf , and office docs GPU is important . Will just see how it performs in real life .
In the end competition is good . Intel was sleeping for years until Zen. We don't want QCOMM to do the same , isn't it ?
umano - Saturday, September 1, 2018 - link
Reading this makes me want to trash my p20 pro :) Usually I buy the best hw I can afford but I find ridiculous spending 1359 euro for an Iphone x, and in this case I went straight for the camera(s). Couldn't be happier. In my consumer perspective, the 980 won't be a reason to upgrade, but a waterproof phone which I can use to take picture underwater would :=)SanX - Sunday, September 2, 2018 - link
ARM has to look at the chips for supercomputers where Intel sells its 8 billion transistors Xeon chips for $2000 while this 8 billion transistors ARM marvel go just for $20. Since most supercomputers are memory bandwidth bound anyway the ARM will easily find its place theredarkich - Monday, September 3, 2018 - link
Look up the Fujitsu's next ARM supercomputer..it will be the first exascale and the fastest in the world.Btw your comparison with Intel is a perfect illustration of just how disgustingly overrated Intel is.
Their chips have nothing to do with high tech any more..it's just a monopoly built on borderline criminal practices.
darkich - Monday, September 3, 2018 - link
"..The A64FX is made up of 8,786 million transistors and will be manufacturered to a 7nm process. Fujitsu claims that it will be the first chip to take advantage of ARM's Scalable Vector Extensions instruction set, developed specifically for high-performance computing. Fujitsu claims that it could also be used for artificial intelligence applications.Sporting 48 cores with two-four assistant cores to assist with processes such as input/ouput, jitter reduction and asynchronous MPI [multiple protocol interface]. For floating-point calculations, Fujitsu claims the A64FX will be able to achieve 2.7 teraflops for 64-bit (FP64) operations, more than 5.4 teraflops for 32-bit (FP32) operations, and more than 10.8 teraflops for 16-bit (FP16) operations."
Smart Japs.
V900 - Monday, September 3, 2018 - link
Intel rarely sells “2000$ Xeon chips”.You seem to confuse the list price with the price actual customers (enterprise) pays for them.
While Xeon is a huge moneymaker for Intel, none of their big enterprise customers actually pay the list price, in most cases it’s significantly less.
They’re still much more expensive than ARM chips, but the kind of customers who need the fastest CPUs available don’t really care.
FunBunny2 - Monday, September 3, 2018 - link
it would be instructive to see an instruction profile of a xeon machine in 'supercomputer' use?? RISC came to be just because lots o folks were convinced that all those high-cycle instructions weren't used much or that useful when they were. could it be that ARM architecture is sufficient for this use case??V900 - Monday, September 3, 2018 - link
No, that’s not the case.Your understanding is also about two decades behind the industry, if you want to bring up the ancient and meaningless RISC vs. CISC.
V900 - Monday, September 3, 2018 - link
If you’re curious about the topic of Intel Xeons in supercomputers, you could always contact some of the many universities/institutions that are running a Xeon based supercomputer.There’s certainly no lack of them.
Out of the ten most powerful supercomputers in the world as of June, three were using Xeon CPUs and 2 were using Xeon Phi.
If you take a look at the Supercomputing Green Top 500, EIGHT of the top ten supercomputers were Xeon machines.
You may want to give these people a call and tell them they’re doing it all wrong, and need to put some ARM processors in their supercomputers instead because RISC.
darkich - Tuesday, September 4, 2018 - link
The first exascale supercomputer will have ARM cores.See my comment above
V900 - Monday, September 3, 2018 - link
If you take a look at the ten most powerful supercomputers in the world, three are using Xeon CPUs and two are running on POWER9 processors.On the Green Top500, Top 10 consists of 8 Xeon and 2 POWER9 supercomputers.
But you’re basically saying, that all the thousands of people who have built these machines, people who have studied or taught computer science for decades are doing it all wrong, yes?
They just need to throw some 20$ cellphone CPUs in their machines instead to really kick it up a notch.
SanX - Monday, September 3, 2018 - link
All top performers in the Top500 are clusters and even if some of them use Xeons the peak performance they got comes from the use of GPU or Phi or IBM. No, people are not stupid as seen by broad use of Intel chips in this list. But the explanation of this is because just few short years back the 64 bit ARM or GPU did not exist which precluded any their serious use in supercomputing.Besides in many cases supercomputers are memory bandwidth bound so any small advantage of some processors over the others becomes insignificant. Xeons are good processors, they just don't deserve anymore almost two orders their production cost.
mazz7 - Tuesday, September 4, 2018 - link
This one is really good in paper, let us see the execution, hope it doesnt include issue like thermal and software optimization :)NeonFlak - Tuesday, September 4, 2018 - link
This particular article seemed to be 40% about the author and 60% about the hardware. Ego much?maroon1 - Friday, September 7, 2018 - link
Don't believe these slides. It is not going to be as good as those slides make it as always. just wait for some independent testssaranya - Friday, September 28, 2018 - link
Great content thanks for sharing this informative blog which provided me technical information keep posting.http://www.besanttechnologies.in/devops-training-i...
kannanbsc - Monday, August 24, 2020 - link
Great blogs we are enjoying ithttp://glimonlinetrainings.com/ansys-online-traini...