Not really. Think about, this is for products that are supposedly due in late 2016 and they are just now figuring out how to successfully produce a chip on this 14nm with any kind of yield. They aren't even close to taping out Zen yet and tapeout is usually 6-8 months or more before release. Looking more and more like AMD will be out of business before Zen even releases.
These production chips are Zen as Arctic Islands is coming out of TSMC. All they need now is to work on yields. Under perfect conditions, Zen can be out in late Q2 next year.
AMD may actually come in earlier to market than expected with Zen.
Thanks to Samsung for bailing out Glofo's normal "late" to market performance expectations.
Technically, the shortest time period from sampling to mass production is only a few months, but it is always longer since so much work needs to be done to verify that you want to go ahead with full production.
It is conceivable, but it would be unheard of, for AMD to be completely satisfied with the initial silicon, yields, AND be ready with the platform by the end of Q2. Especially with a completely new design on a complete new process.
It's a double-whammy of a new design on a new node. If they get it out in Q3 I'd be stunned. I have much more confidence in AM4 being ready, and we know AMD already has a DDR4 controller (a dual DDR3/4 controller, in fact.. I really hope that means Zen will have it and I can use DDR3 instead of DDR4 since I have so much high-end DDR3).
"All they need now is to work on yields. Under perfect conditions, Zen can be out in late Q2 next year."
Not likely to happen. Sure, there will be ES available at that time and few months later we'll have the first slideware and vaporware release. First Zen CPUs (a handful of SKUs) available in stores? Early Q4 at best. Most of the initial lineup available? 2017.
You're saying it will take them 6-8 months to go from sample production to ES? Really? Given that Xen is the only planned 14nm product for AMDs next line-up (Bristol Ridge is still 28nm Excavator, but in a new socket), that seems like a stretch. Sure, it will take longer than the same job on an established and well known/optimized node. But more than half a year for ES? I don't believe that.
@MapRef41N93W He said "finally some good news out of AMD", can you read? Are you an idiot? In your opinion they should just quit right? What a moron you are, learn to read.
Bulldozer was a failled design, and they threw it in the trash. They were always 2 fab gen behind intel, this is going to change. All they need is price competitiveness. Pure performace is not that important anymore. I have an haswell cpu and don't plan to change it until a few years. Gonna swap GPU long before i swap CPU. So give AMD something descent at 200$ and it can go a long way.
The problem with Bulldozer is that each core are underpowered compared to Intel I design. They need to redesign less about the APU aspect rather focus on making beefier cores instead of making less powerful quad core in order to compete with Intel.
That's exactly what Zen is. It's a monster 10 pipeline CPU.
4ALUs, 2 AGUs, and 4FPUs (64-bit mergeable pipes). In theory, this should be able to match up well against Haswell (probably will be a bit below on average).
Funny, considering that for the last, what, five years price is the only thing AMD _can_ compete on. I`m all for the competition of powerful architectures, not race to the ground pricing wars.
Tape out of chip happens before sample production. The article says - 'AMD for their part has already announced that they have taped out several 14LPP designs for GlobalFondries' And this is first sample production for one of those chips.
I hope that amd can deliver a price worthy solution in 2016, you already can see intel is happy and shows it by releasing cpu's which has nothing to offer to me. These new series are absolute nothing new and barely am faster than the previous ones. Hell did not even upgrade yet at all, because its not worth the big pile of money they charge, and since they killed the second hand market you can sell your still perfect working product only for scraps prices. Any game i have runs smooth on either the amd 9590 or the intel 4690
Not sure why people disagree with you, because this is truly awesome news! I am a proud Intel fanboy myself, but it has been sad to watch AMD fall further and further behind on all fronts. But if they catch up on their manufacturing, and a fire is being lit under their engineering staff, then this could place their next gen CPUs at a mere year behind Intel. More importantly, it puts them on a path to potentially catch up to Intel who seems to have been resting relatively easy for the last few product cycles.
They haven't been resting easy, they have just been investing in gains that are of little benefit to the desktop consumer market.
But look at the perf/watt they have been working on the last few cycles? Or their new memory system? Also, while we don't see it, they seem to have been a bit busy in the server market as of late too.
I agree they aren't being threatened in the desktop space, and therefore we aren't seeing much if any improvement on our end. But at the same time, its just because they have taken this chance to focus on other things.
Nor is Carrizo, for all intents and purposes. Imagine being told that your new product which uses half the power of the old one for the same, if not greater performance, is "embarassing". And that's at 15W.
What they managed to do with Carrizo is really impressive. If they finally have access to a process node that isn't four years old I'm hopeful about AMD to at least get back on their feet a bit.
They're priced just fine. In the real world, you can't get a 980Ti for less than $750 that comes with a cooler as nice as the one that comes with the $650 Fury X.
If water cooling doesn't matter to you that much, buy the standard Fury for $100 less and lose almost no performance.
The problem I have grabbing one is the lack of dual DVI-D. The display connectivity choices are abysmal. Hopefully AMD has learned from the backlash over their choices... or someone makes a good Fury card with Dual DVI-D connections.
If a lack of DVI-D is your only problem, why not invest in a couple of £10 DisplayPort to DVI-D adaptors? I have a full set of DP adapters for work, because we have no standard for conference room projectors, and you might need to connect by DVI, HDMI or even VGA.
I for one applaud them for ditching obsolete standards sooner rather than later. Especially given the amazing versatility of DP and the wealth of reasonably priced adapters that can be found.
Sticking a DVI port on the Fury X would have been silly and odd given the otherwise forward looking nature of the card.
If you can afford a $650 GPU, you can afford a $20 adapter dongle.
Actually for people with 1440p this adapters wont work as they support up to 1080p , and adapters from DUAL LINK to Display port that actually use the bandwidth of Dual link so you could use your old 1440p montor that does not have a native displayport connector are expensive , they are active so i am sure they also add lantecy .
Sorry for my ignorance, but could you give a little more info on a few of the items mentioned in the article?
What is 14XM FinFET process? How does it compare to this 14nm FinFET LPP that they were able to produce sample chips from?
Also what was the nature of the licensing with Samsung's 14nm FinFET? How similar is this GloFo 14nm process to Samsung's 14nm process? What are the differences?
14XM finfet no longer exists. Global Foundries failed to make it work. In desperation, they licensed Samsung's 14 nm finfet process in order to not screw over AMD for the umpteenth time by being late to market.
Idk the details of the licensing agreement, but I assume Glofo pays Samsung a fee for each wafer produced plus any fees for Samsung on site training, engineering, etc.
As for 14XM vs Samsung 14nm LPP: GlobalFoundries confirms that this news means the end of the road for its own 14XM process. The firm says Samsung's process tech has two key advantages over 14XM. Samsung's tech is further along in development, so the schedule is more attractive, and Samsung's 14-nm FinFET tech provides better area scaling by cramming more gates into a given area.
As for the licensing agreement: GlobalFoundries will so closely implement Samsung's technology that the two partners are once again talking about portability between fabs. Through a proven level of fab synchronization never previously achieved outside of a single company, Samsung and GLOBALFOUNDRIES will use a coordinated copy-smart approach involving materials, process recipes, integration and tools. The company will also run fab-sync test chips on a regular basis to ensure that the fabs are using the 14nm FinFET process exactly the same.
You read like your handle: Intel. Your crapping yourself because AMD's Fab 8 is exceeding their goals. Last you want is for them to have Apple and AMD full throttle. Intel's days as dictating the industry, like Microsoft, are over.
GloFo, IBM (now absorbed into GloFo) and Samsung were already collaborating on processes and their 32/28 nm process where a Common Platform process (the name of the consortium) that was used across the three companies with the same tools/libraries used to design for all three. They shared technologies between each other, developed it collaboratively and bought the same tools and machines. The collaboration goes back decades, as it's expensive to develop and build all tools for just one or two fabs without sharing some with others. GloFo just wasn't going to develop any competing process to the common one. Their 22 nm process in East Fishkill is a proprietary process for example. Considering they had a common platform for 32/28 nm with IBM and Samsung as well as STMicroelectronics that they all contributed too, it's nothing new.
Things like 28 nm FD-SOI was a failure too but that doesn't mean they have got a free ride on someone else's tech. GloFo collaborated with STMicro on 28 nm FD-SOI but designs like ST-E's mobile AP's/ModAP's never made it to commercial production and there just weren't any demand for it. AMD opted for 32 nm SOI and 28 nm SHP at GF rather than 32 nm FD-SOI (their other proprietary process) or 28 nm HPP, at the time GloFo also was also manufacturing using the 32/28 nm Common Platform process and now they choose not to use 14XM and thus didn't force GloFo to develop and manufacture using two slightly different processes. Dual-sourcing 32/28 nm has mostly been working out for all of the companies collaborating though. It makes the choice at the fabless companies easier, they could choose to design for common platform or TSMC basically. The shared 14LPE/LPP-process is already used at three different fabs and will probably expand to one or two more. Canceling 14XM doesn't mean they had to run out and license 14LPP for lots of money as they have contributed to the research and development, though it means the process Samsung was further along on became the process which allows multi-sourcing and the one that is implemented across different companies. The alternative would have been to finish 14XM as a proprietary tech and use 14LPE/LPP as the common process, but that would have meant were little demand for 14XM and at that time there wasn't really any design for 14XM any how, and obviously nobody working with them on a proprietary process for 14 nm.
Any word on whether AMD will be moving their Radeon GPUs over to GloFo? It will be a boon for them if they won't have to share foundry time with the likes of Apple and Nvidia.
I'm glad to see some good news from AMD for a change. I really hope they can turn their situation around.........they've been making me rather nervous lately.
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.
58 Comments
Back to Article
Shadow7037932 - Thursday, November 5, 2015 - link
Finally some good news out of AMD.MapRef41N93W - Thursday, November 5, 2015 - link
Not really. Think about, this is for products that are supposedly due in late 2016 and they are just now figuring out how to successfully produce a chip on this 14nm with any kind of yield. They aren't even close to taping out Zen yet and tapeout is usually 6-8 months or more before release. Looking more and more like AMD will be out of business before Zen even releases.effingterrible - Thursday, November 5, 2015 - link
They did a tape out back in July.iwod - Thursday, November 5, 2015 - link
And that is due to Samsung's 14nm Tech, not because they manage to do it themsevles.looncraz - Friday, November 6, 2015 - link
Companies license tech from each other all the time. Intel even licenses tech from AMD and vice versa.Morawka - Saturday, November 7, 2015 - link
Architecture tech is easy, Fabbing is hardkoko4kaka - Saturday, November 14, 2015 - link
They're both remarkably difficult endeavours actually.testbug00 - Sunday, November 8, 2015 - link
Tape out under a month ago for Zen.Intel999 - Friday, November 6, 2015 - link
These production chips are Zen as Arctic Islands is coming out of TSMC. All they need now is to work on yields. Under perfect conditions, Zen can be out in late Q2 next year.AMD may actually come in earlier to market than expected with Zen.
Thanks to Samsung for bailing out Glofo's normal "late" to market performance expectations.
looncraz - Saturday, November 7, 2015 - link
Technically, the shortest time period from sampling to mass production is only a few months, but it is always longer since so much work needs to be done to verify that you want to go ahead with full production.It is conceivable, but it would be unheard of, for AMD to be completely satisfied with the initial silicon, yields, AND be ready with the platform by the end of Q2. Especially with a completely new design on a complete new process.
It's a double-whammy of a new design on a new node. If they get it out in Q3 I'd be stunned. I have much more confidence in AM4 being ready, and we know AMD already has a DDR4 controller (a dual DDR3/4 controller, in fact.. I really hope that means Zen will have it and I can use DDR3 instead of DDR4 since I have so much high-end DDR3).
Arnulf - Sunday, November 8, 2015 - link
"All they need now is to work on yields. Under perfect conditions, Zen can be out in late Q2 next year."Not likely to happen. Sure, there will be ES available at that time and few months later we'll have the first slideware and vaporware release. First Zen CPUs (a handful of SKUs) available in stores? Early Q4 at best. Most of the initial lineup available? 2017.
Valantar - Sunday, November 8, 2015 - link
You're saying it will take them 6-8 months to go from sample production to ES? Really? Given that Xen is the only planned 14nm product for AMDs next line-up (Bristol Ridge is still 28nm Excavator, but in a new socket), that seems like a stretch. Sure, it will take longer than the same job on an established and well known/optimized node. But more than half a year for ES? I don't believe that.testbug00 - Sunday, November 8, 2015 - link
6-8mo? A correct numberYou're missing the other correct number.
testbug00 - Sunday, November 8, 2015 - link
1H isn't realistically possible. Unless it's on pape only.readbeforeyoucomment - Friday, November 6, 2015 - link
IT IS GOOD NEWS PEOPLE, They are improving!@MapRef41N93W
He said "finally some good news out of AMD", can you read?
Are you an idiot? In your opinion they should just quit right? What a moron you are, learn to read.
Michael Bay - Friday, November 6, 2015 - link
It is, if you consider someone coming out of a clinical death into coma an improvement.Not to mention your exemplary case of AMD butthurt.
Da W - Friday, November 6, 2015 - link
Bulldozer was a failled design, and they threw it in the trash.They were always 2 fab gen behind intel, this is going to change.
All they need is price competitiveness.
Pure performace is not that important anymore. I have an haswell cpu and don't plan to change it until a few years. Gonna swap GPU long before i swap CPU.
So give AMD something descent at 200$ and it can go a long way.
pugster - Friday, November 6, 2015 - link
The problem with Bulldozer is that each core are underpowered compared to Intel I design. They need to redesign less about the APU aspect rather focus on making beefier cores instead of making less powerful quad core in order to compete with Intel.looncraz - Saturday, November 7, 2015 - link
That's exactly what Zen is. It's a monster 10 pipeline CPU.4ALUs, 2 AGUs, and 4FPUs (64-bit mergeable pipes). In theory, this should be able to match up well against Haswell (probably will be a bit below on average).
It comes down to clock speed in the end.
Michael Bay - Saturday, November 7, 2015 - link
Funny, considering that for the last, what, five years price is the only thing AMD _can_ compete on.I`m all for the competition of powerful architectures, not race to the ground pricing wars.
medi03 - Monday, November 9, 2015 - link
AMD's Jaguar cores had better perf/watt than anything Intel's, according to Anand.ant6n - Friday, November 6, 2015 - link
Can we get some moderation on this?How about an option to flag abusive comments.
Mr Perfect - Friday, November 6, 2015 - link
Agreed, personal attacks deserve moderating.prtskg - Friday, November 6, 2015 - link
Tape out of chip happens before sample production. The article says -'AMD for their part has already announced that they have taped out several 14LPP designs for GlobalFondries'
And this is first sample production for one of those chips.
bronan - Monday, November 9, 2015 - link
I hope that amd can deliver a price worthy solution in 2016, you already can see intel is happy and shows it by releasing cpu's which has nothing to offer to me. These new series are absolute nothing new and barely am faster than the previous ones. Hell did not even upgrade yet at all, because its not worth the big pile of money they charge, and since they killed the second hand market you can sell your still perfect working product only for scraps prices.Any game i have runs smooth on either the amd 9590 or the intel 4690
CaedenV - Friday, November 6, 2015 - link
Not sure why people disagree with you, because this is truly awesome news! I am a proud Intel fanboy myself, but it has been sad to watch AMD fall further and further behind on all fronts. But if they catch up on their manufacturing, and a fire is being lit under their engineering staff, then this could place their next gen CPUs at a mere year behind Intel. More importantly, it puts them on a path to potentially catch up to Intel who seems to have been resting relatively easy for the last few product cycles.I for one would love to see a little competition!
tekeffect - Friday, November 6, 2015 - link
Agreed I'm an Intel guy and boy have they been comfortable for too long. Any news that might get these 2 competing again is good newsRefuge - Monday, November 9, 2015 - link
They haven't been resting easy, they have just been investing in gains that are of little benefit to the desktop consumer market.But look at the perf/watt they have been working on the last few cycles? Or their new memory system? Also, while we don't see it, they seem to have been a bit busy in the server market as of late too.
I agree they aren't being threatened in the desktop space, and therefore we aren't seeing much if any improvement on our end. But at the same time, its just because they have taken this chance to focus on other things.
sabot00 - Thursday, November 5, 2015 - link
Good news for them. They need a cutting-edge process node if they want Zen to have any chance at all.spikebike - Thursday, November 5, 2015 - link
I'm hoping AMD can execute well, the CPUs, APUs, and GPUs are looking kinda embarassing both on performance and performance/watt.mayankleoboy1 - Friday, November 6, 2015 - link
Their GPU's are not embarassingsilverblue - Friday, November 6, 2015 - link
Nor is Carrizo, for all intents and purposes. Imagine being told that your new product which uses half the power of the old one for the same, if not greater performance, is "embarassing". And that's at 15W.gijames1225 - Friday, November 6, 2015 - link
What they managed to do with Carrizo is really impressive. If they finally have access to a process node that isn't four years old I'm hopeful about AMD to at least get back on their feet a bit.Refuge - Monday, November 9, 2015 - link
If only they could get them put into some fucking laptops!!!! Gah!Samus - Friday, November 6, 2015 - link
No, they're really good, but not competitively priced.looncraz - Saturday, November 7, 2015 - link
They're priced just fine. In the real world, you can't get a 980Ti for less than $750 that comes with a cooler as nice as the one that comes with the $650 Fury X.If water cooling doesn't matter to you that much, buy the standard Fury for $100 less and lose almost no performance.
The problem I have grabbing one is the lack of dual DVI-D. The display connectivity choices are abysmal. Hopefully AMD has learned from the backlash over their choices... or someone makes a good Fury card with Dual DVI-D connections.
markbanang - Saturday, November 7, 2015 - link
If a lack of DVI-D is your only problem, why not invest in a couple of £10 DisplayPort to DVI-D adaptors? I have a full set of DP adapters for work, because we have no standard for conference room projectors, and you might need to connect by DVI, HDMI or even VGA.Valantar - Sunday, November 8, 2015 - link
I for one applaud them for ditching obsolete standards sooner rather than later. Especially given the amazing versatility of DP and the wealth of reasonably priced adapters that can be found.Sticking a DVI port on the Fury X would have been silly and odd given the otherwise forward looking nature of the card.
If you can afford a $650 GPU, you can afford a $20 adapter dongle.
David_K - Thursday, November 12, 2015 - link
Actually for people with 1440p this adapters wont work as they support up to 1080p , and adapters from DUAL LINK to Display port that actually use the bandwidth of Dual link so you could use your old 1440p montor that does not have a native displayport connector are expensive , they are active so i am sure they also add lantecy .medi03 - Monday, November 9, 2015 - link
AMD's mid range GPUs rock, I mean R9 380 and R9 390.380 has much better perf than 960, while consuming marginally more power in games (7-30w).
In the high end, Fury is a mixed bag, and I don't care about lower end so didn't follow.
Cliff34 - Friday, November 6, 2015 - link
I just hope they can get their chips out quickly to compete with Intel.Mugur - Friday, November 6, 2015 - link
LPP at 0.8V for Core device? Isn't it mobile territory?psychobriggsy - Friday, November 6, 2015 - link
You should look at GF's 22FDX node then - that starts at 0.4V.Andrei Frumusanu - Friday, November 6, 2015 - link
That is not the nominal VDD, 14LPE for example goes down to 575mV.looncraz - Saturday, November 7, 2015 - link
Sure, until you clock the CPU into the 4GHz range :pikjadoon - Friday, November 6, 2015 - link
Finally. Glad to hear some good news out of GloFo and AMD. Let's gooooo, competition!Minor type: *principal
soliloquist - Friday, November 6, 2015 - link
Sorry for my ignorance, but could you give a little more info on a few of the items mentioned in the article?What is 14XM FinFET process? How does it compare to this 14nm FinFET LPP that they were able to produce sample chips from?
Also what was the nature of the licensing with Samsung's 14nm FinFET? How similar is this GloFo 14nm process to Samsung's 14nm process? What are the differences?
Intel999 - Friday, November 6, 2015 - link
14XM finfet no longer exists. Global Foundries failed to make it work. In desperation, they licensed Samsung's 14 nm finfet process in order to not screw over AMD for the umpteenth time by being late to market.Idk the details of the licensing agreement, but I assume Glofo pays Samsung a fee for each wafer produced plus any fees for Samsung on site training, engineering, etc.
soliloquist - Saturday, November 7, 2015 - link
Did some digging:As for 14XM vs Samsung 14nm LPP:
GlobalFoundries confirms that this news means the end of the road for its own 14XM process. The firm says Samsung's process tech has two key advantages over 14XM. Samsung's tech is further along in development, so the schedule is more attractive, and Samsung's 14-nm FinFET tech provides better area scaling by cramming more gates into a given area.
As for the licensing agreement:
GlobalFoundries will so closely implement Samsung's technology that the two partners are once again talking about portability between fabs. Through a proven level of fab synchronization never previously achieved outside of a single company, Samsung and GLOBALFOUNDRIES will use a coordinated copy-smart approach involving materials, process recipes, integration and tools. The company will also run fab-sync test chips on a regular basis to ensure that the fabs are using the 14nm FinFET process exactly the same.
mdriftmeyer - Saturday, November 7, 2015 - link
You read like your handle: Intel. Your crapping yourself because AMD's Fab 8 is exceeding their goals. Last you want is for them to have Apple and AMD full throttle. Intel's days as dictating the industry, like Microsoft, are over.Alexey291 - Sunday, November 8, 2015 - link
Actually he sounds calm and collected. And most importantly reasonable. You sound angry and upset.Are you sure you didn't mean to log into verge or engadget instead?
Michael Bay - Monday, November 9, 2015 - link
Don`t insult Engadget. After vergins went away, it became decent again.Penti - Sunday, November 8, 2015 - link
GloFo, IBM (now absorbed into GloFo) and Samsung were already collaborating on processes and their 32/28 nm process where a Common Platform process (the name of the consortium) that was used across the three companies with the same tools/libraries used to design for all three. They shared technologies between each other, developed it collaboratively and bought the same tools and machines. The collaboration goes back decades, as it's expensive to develop and build all tools for just one or two fabs without sharing some with others. GloFo just wasn't going to develop any competing process to the common one. Their 22 nm process in East Fishkill is a proprietary process for example. Considering they had a common platform for 32/28 nm with IBM and Samsung as well as STMicroelectronics that they all contributed too, it's nothing new.Things like 28 nm FD-SOI was a failure too but that doesn't mean they have got a free ride on someone else's tech. GloFo collaborated with STMicro on 28 nm FD-SOI but designs like ST-E's mobile AP's/ModAP's never made it to commercial production and there just weren't any demand for it. AMD opted for 32 nm SOI and 28 nm SHP at GF rather than 32 nm FD-SOI (their other proprietary process) or 28 nm HPP, at the time GloFo also was also manufacturing using the 32/28 nm Common Platform process and now they choose not to use 14XM and thus didn't force GloFo to develop and manufacture using two slightly different processes. Dual-sourcing 32/28 nm has mostly been working out for all of the companies collaborating though. It makes the choice at the fabless companies easier, they could choose to design for common platform or TSMC basically. The shared 14LPE/LPP-process is already used at three different fabs and will probably expand to one or two more. Canceling 14XM doesn't mean they had to run out and license 14LPP for lots of money as they have contributed to the research and development, though it means the process Samsung was further along on became the process which allows multi-sourcing and the one that is implemented across different companies. The alternative would have been to finish 14XM as a proprietary tech and use 14LPE/LPP as the common process, but that would have meant were little demand for 14XM and at that time there wasn't really any design for 14XM any how, and obviously nobody working with them on a proprietary process for 14 nm.
r3loaded - Saturday, November 7, 2015 - link
Any word on whether AMD will be moving their Radeon GPUs over to GloFo? It will be a boon for them if they won't have to share foundry time with the likes of Apple and Nvidia.koko4kaka - Saturday, November 14, 2015 - link
I doubt it. GloFo has been an albatross around AMD's neck for the past few years. If anything AMD would prefer to move their CPUs over to TSMC.Nathan_Foster - Sunday, November 8, 2015 - link
I'm glad to see some good news from AMD for a change. I really hope they can turn their situation around.........they've been making me rather nervous lately.testbug00 - Sunday, November 8, 2015 - link
Zen launch non- paper in 2016.Zen launch in paper 2016.
Both true statements I feel. Is Zen always Zen!
Glideslope - Monday, November 9, 2015 - link
Well done GloFo. Can only improve from here on out.