GPU Rumors: AMD Plans 20nm but NVIDIA Waits for 16nm

Subject: General Tech, Graphics Cards | December 28, 2014 - 09:47 PM |
Tagged: radeon, nvidia, gtx, geforce, amd

According to an anonymous source of WCCFTech, AMD is preparing a 20nm-based graphics architecture that is expected to release in April or May. Originally, they predicted that the graphics devices, which they call R9 300 series, would be available in February or March. The reason for this “delay” is a massive demand for 20nm production.

View Full Size

The source also claims that NVIDIA will skip 20nm entirely and instead opt for 16nm when that becomes available (which is said to be mid or late 2016). The expectation is that NVIDIA will answer AMD's new graphics devices with a higher-end Maxwell device that is still at 28nm. Earlier rumors, based on a leaked SiSoftware entry, claim 3072 CUDA cores that are clocked between 1.1 GHz and 1.39 GHz. If true, this would give it between 6.75 and 8.54 TeraFLOPs of performance, the higher of which is right around the advertised performance of a GeForce Titan Z (only in a single compute device that does not require distribution of work like what SLI was created to automate).

Will this strategy work in NVIDIA's favor? I don't know. 28nm is a fairly stable process at this point, which will probably allow them to get chips that can be bigger and more aggressively clocked. On the other hand, they pretty much need to rely upon chips that are bigger and more aggressively clocked to be competitive with AMD's slightly more design architecture. Previous rumors also hint that AMD is looking at water-cooling for their reference card, which might place yet another handicap against NVIDIA, although cooling is not an area that NVIDIA struggles in.

Source: WCCFTech

December 28, 2014 | 10:12 PM - Posted by arbiter

I expect a few ppl to get their panties in a bunch over me saying this but, Is this another timetable AMD will miss yet again?

December 28, 2014 | 10:31 PM - Posted by Anonymous (not verified)

Depends They could just be playing it safe by opting to go with 20nm because 16nm might have some unexpected delays, maybe glo-flo and Samsung are giving AMD a good deal for the 20nm hard to tell right now. All this being said if the AMD gpu's come to market in mid 2015 then they will be ahead of Nvidia anyway. As a blanket statment the whole industry is going to start missing wafer sizes including Intel who has a really strong wafer supply.

December 28, 2014 | 10:47 PM - Posted by arbiter

Knowing 16nm nvidia gpu's coming out in coming months after AMD 20nm part, could keep ppl from buying as well since could be price drops that come with it and to see what kinda performance nvidia brings to the table as well.

December 29, 2014 | 12:33 AM - Posted by Anonymous (not verified)

your avatar should say "Are you Nvidiot?"

December 29, 2014 | 10:46 AM - Posted by Anonymous (not verified)

As soon as I read the first post, I Got that WCCF T vibe, but it's good to see a process node shrink, and the difference between 20nm and 16 nm is not that great, so maybe Ronald will show up and the three ring circus will begin.

With all the fabs heading towards 14nm, at least there will be a more even parity among process nodes, and then everybody will have to get more out of reengineering their CPU/GPU/SOC microarchitectures rather than simply relying on process node shrinks to provide better power usage, and performance. Maybe the SOC/GPU makers will focus more on drivers, and getting better graphics capabilities on their SOC/APU products, as well as the desktop variants. I'm looking forward to a tablet with the PowerVR wizard, and ray tracing ability built into the GPU, hopefully the discrete GPUs will get this ability also, now that the process node shrinks will allow for more functionality on the GPU's die. AMD needs a to get a custom ARMv8 ISA based competitor to Nvidia's Tegra K1 Denver, and force more innovation in the Tablet SOC market, Nvidia has the lead, outside of the closed hardware/software ecosystem of Apple's tablet SOC, so the OEM's market needs a competing product to keep Nvidia on its toes.

I'll take whatever shrink they can give, as long as I can look forward to better laptop/mobile discrete GPUs, with good driver support for OpenGL, Mantle, DX, whatever, and both windows, and Linux OSs, and that includes Win. 7. Having Mantle available for 7 is great news, but Mantle support for Linux need to get here ASAP.

December 29, 2014 | 07:45 AM - Posted by Anonymous (not verified)

Retail NVIDIA 16nm cards will be out probably 10-12 months after AMD 20nm hit shelves. Only people that currently have OC 780Ti cards will be waiting that long. Everyone else either already upgraded to 970/980 or snagged some really cheap 290/X cards.

December 28, 2014 | 11:17 PM - Posted by pdjblum

Yeah, keep hoping they step up with a happy surprise, but freesync seems to be well behind schedule and this seems to be the same. Now nvidia can sit on the real Maxwell enthusiast flagship gpu, the 210 I gather, for a good while longer.

Hard to root against Nvidia's only competition, but AMD is doing everything it can to alienate us.

December 28, 2014 | 11:07 PM - Posted by jchambers2586

I am not waiting on AMD I am buying a pair of GTX970's for SLI I would buy a pair of R9290x's but the power requirements are too high I have a new 850 watt PSU from evga I am not changing the PSU for a cross fire set up. AMD missed the boat for my money this upgrade cycle I am not waiting I need something now.

December 28, 2014 | 11:50 PM - Posted by arbiter

that was more the Chip fabs that caused the miss not AMD.

December 29, 2014 | 10:56 AM - Posted by Anonymous (not verified)

Apple grabbed all the bleeding edge fab capacity from the fabs, and the Fab partners are more responsible for getting their processes working with their clients designs. Apple had the big bucks to throw at the fabs, and offer them some financial assistance in return for first dibs on the capacity. Those tall stacks of Benjamins scream.

December 30, 2014 | 12:12 AM - Posted by arbiter

Its a bit more complicated then that. Apple uses a smaller chip, Arm cpu's are low power and small. Which that was worked out. The process at larger sizes that are needed for GPU's was delayed a bit.

December 29, 2014 | 06:10 PM - Posted by Mad_Dane

HUH? A 850 watt can easily drive 2 290X's!

Standard version of a 290X has a 6 pin and an 8 pin PCI-E connectors, the math goes (75x2 from mobo PCI-E slots) (150x2 From the 8 pin PCI-E) (75x2 From the 6 pin PCI-E)=600 watts, that leaves you 250 watts for CPU, SSD's, HDD's and fans, so unless you run an AMD FX9590 or an obscene amount of drives and fans, you will be just fine.

December 29, 2014 | 01:23 AM - Posted by Anonymous (not verified)

AMD only used water in the past, because their only option was to raise clocks. It also added to cost of the card. If AMD uses water with 16nm flagships I'll be shocked.

December 29, 2014 | 01:26 AM - Posted by Anonymous (not verified)


December 29, 2014 | 05:59 PM - Posted by Mad_Dane

Why would you be surprised? Intel chips has shown us that a move to a 3D chip design will make the temperatures spike more aggressively.

December 29, 2014 | 02:45 AM - Posted by Anonymous (not verified)

I dont see the point in any of this...

Until there are 21:9 OLED f/a/gsync monitors out there i dont see the point i ANY video cards.

December 29, 2014 | 04:39 AM - Posted by JohnGR

I am not expecting gpus at 20nm. AMD could make a few low end gpus for mobile or desktop market at 20nm, but I don't expect something major like a new midrange or a new hi end in that process.

December 29, 2014 | 06:46 AM - Posted by Paul K (not verified)

The only problem with reporting from WCCFTech is that they have such a history of making stuff up that when you repost their rumours and talk about it you give more credence than the rumour deserves. It could be true but WCCFtech have a terrible reputation because they make stories like this up for the clicks.

Honestly its best to avoid them as a source, they are worse than Fudzilla and Semiaccurate in accuracy.

December 29, 2014 | 07:38 AM - Posted by MarkT (not verified)

I disagree with Paul K.

December 29, 2014 | 03:21 PM - Posted by renz (not verified)

accurate or not i agree with him about wccftech click bait tactic. to them it is better if they have something to put out rather than nothing at all. so what they do is pick up all the rumor that they can get on the net and then make article out of it. then add their own speculation to it. then all of a sudden we heard new rumor that never being discuss on another forum before. if it's about GPU i believe VCZ more.

December 30, 2014 | 12:16 AM - Posted by arbiter

I agree as well, when I seen this linked to WCCFTech and source was Anonymous. Who would have such access to both sides and what their future plans?

December 29, 2014 | 10:52 AM - Posted by H1tman_Actua1

AMD doing what they always do, riding the coat tails of NVidia and falling off constantly


December 29, 2014 | 11:21 AM - Posted by Anonymous (not verified)

AMD got no boost from Nvidia, and does just fine considering their limited financial resources. Really the coattail riding was mostly from that x86 license that IBM forced Intel to give to AMD and others at the beginning of the PC market, and it's a good thing IBM did this, otherwise there would be no PC market outside of Intel's control. If only IBM had forced the cross licensing of the OS it used on their PCs, or bought the CPM OS outright, and forced M$ into more competition, things would be much better OS wise. They all rode the coattails of IBM at one time, hopefully the Power8's that are up for ARM style licensing will find their way into some gaming PCs, once the third party licensees get ahold of a Power8 license, and some more IBM coattail riding can happen on the licensed IP side, and give Intel some serious competition for the CPU market. It would be great it those Nintendo rumors pan out, and AMD gets to integrate its graphics, with a power8 based Nintendo gaming console, how long would take others, maybe even AMD, and Nvidia to begin offering their own APUs/SOCs with a licensed power8 design, there're doing it with ARM, so a licensed Power8 based product from AMD, Nvidia, and others is not out of the question.

December 29, 2014 | 01:38 PM - Posted by Anonymous (not verified)

Guessing Scott missed the WCCFTech Sisoft leak that had a AMD Fiji scoring way higher than the leaked full blown Maxwell.

Nvidia big Maxwell leak high scored a 55.2GB
AMD Fiji leaked high scored 63.6GB

Looking at the dates of the Sisoft leaks. AMD higher score came a week earlier.

December 29, 2014 | 03:25 PM - Posted by renz (not verified)

well even so it is still speculation. is there a way to really know that the info is not fake (even the article from wccftech itself bearing rumor tag)? and PCPer for one is not a rumor/speculating site. they post rumor from time to time but not like it is their bread and butter like Wccftech. you can't expect PCPer to post all the rumor and speculation out there.

December 29, 2014 | 04:39 PM - Posted by Anonymous (not verified)

Scott is basing his Maxwell prediction of between 6.75 and 8.54 TeraFLOPs of performance on rumors. If you believe that rumor you kind of have to believe that AMD has a better performing product based on the scores.

If you also take the Chiphell leaks they all point to AMD having a faster product than Nvidia.

Unless you want to be selective on which rumors you believe.

December 29, 2014 | 06:19 PM - Posted by renz (not verified)

rumor is rumor. there is nothing selective about it. selective in which rumor to believe? since when rumor become a fact? i've seen some people take all the these speculation as a straight fact.

December 29, 2014 | 09:48 PM - Posted by Anonymous (not verified)

ChipHell have been on the mark for most leaks including the 780 and 980. So i would take their word over everyone else's. The problem is AMD are faster according to Chip-hell so nVidia people are changing the goal posts or in denial. We will find out soon, but I definitely believe one company out of the two will start getting far ahead since they have started using different foundries.

December 29, 2014 | 09:52 PM - Posted by Anonymous (not verified)

Well Chiphell have been very accurate in the past for 680, 780 and 980. So why are they suddenly wrong now? Is it bcoz they say AMD is going to wreck nVidia?

December 30, 2014 | 04:16 AM - Posted by renz (not verified)

oh come on this is nothing to do with AMD vs Nvidia. accurate or not rumors are rumors. can you prove something like captain jack really exist? is there real fact to back it up right now? if there is then give me the link. not from the leak article or leak source but one that is directly from AMD themselves. as long as it is not something official from AMD then it is not final.

December 30, 2014 | 04:32 PM - Posted by arbiter

Yea, AMD has yet to make a chip that is fast and power efficient. That captain jack leak shows both which puts some doubt on it. r9 285 Tonga, AMD claimed 60watt drop in TDP but really it was only a 10 watt drop. So I will not believe it til AMD puts the chip out and proves it legit and not fudged numbers or someone else's attempt to BS people.

December 30, 2014 | 07:30 PM - Posted by Anonymous (not verified)

Like Nvidia is doing by putting average power usage numbers.

Got it. No bias here.

December 29, 2014 | 06:25 PM - Posted by Scott Michaud

Actually I didn't see that.

As for reporting on rumours, it's complicated. We don't want to be spouting crap, but they can still be interesting for our readers even if they don't always work out. It's not even a matter of being first -- personally, I'm a few days behind most stories anyway (unless we're given advanced notice or we luck out).

December 29, 2014 | 03:28 PM - Posted by renz (not verified)

any word on when AMD going to release Mantle public SDK? they promised to release it this year right? we still got a day or so before 2014 ends.

December 29, 2014 | 09:44 PM - Posted by Anonymous (not verified)

Well I hope AMD don't release Mantle SDK for free bcoz they spent allot of money and time in this tech. AMD is a company, not a charity OK!

December 30, 2014 | 12:24 AM - Posted by arbiter

AMD already claimed they would and if they pull out and don't do it well kinda puts the fork in it as dead. If AMD tries to charge what they claimed they will release for free, Less companies have monetary incentive to put it in their game, they won't.

The tech ONLY works on AMD gpu's. SO if they charge for it well any chance of others adopting it ends. Before you say Physx works only on Nvidia gpu's, is not completely true as it can work on cpu though much slower.

December 30, 2014 | 04:17 AM - Posted by renz (not verified)

if you say so then you must also agree with nvidia proprietary approach.

December 30, 2014 | 04:43 PM - Posted by arbiter

Um last i checked right outta the gate most ALL nvidia's stuff does work on AMD cards, or will work in a system with an AMD gpu. Unlike mantle where its locked to AMD gpu's Only. AMD said they would make it open source but over a year but we have yet to see that happen as of now.

Other thing with AMD they let PR talk about everything they are working on even when they don't have a working product yet. Nvidia from what I seen stays quiet about what they are working on til they got a working one ready to show off. AMD should take that to heart as well instead of ranting off what they are going to do and in the end not living up to everything or even having it ready when they claim they will have it ready.

December 30, 2014 | 04:17 AM - Posted by renz (not verified)

if you say so then you must also agree with nvidia proprietary approach.

December 30, 2014 | 11:45 AM - Posted by Josh Walrath

So yeah, I've been reading about this...

Couple of things.  20 nm planar is bad for large GPUs.  Voltage/frequency scaling is not appropriate for such a design.  I had hoped that someone would have gone for 20 nm FDSOI planar during this time, as it has slightly superior characteristics than Intel's 22 nm TriGate (aka FinFET).  That would have been a logical jump for the GPU guys as they would have had really nice density scaling and some major improvements in frequency and power.  Alas, nobody went that direction for this particular node (except for some small runs at ST Micro) and some rumors that GLOBALFOUNDRIES was looking in that direction for RF applications.

AMD might go 20 nm planar for a smaller, low power design.  Not for a big GPU though.  Next gen Kabini variants are more appropriate for such a process.  Apple's latest SOCs are all low power and smaller designs than what typically goes for a GPU or a desktop processor today.

I think there is some credence to AMD looking at GF for a 28 nm SHP production of GPUs, but you have to wonder how much extra development they need in the design stage to get that to work effectively.  TSMC and GF certainly do not share common standard cells between their 28 nm processes, so redesigns for each production line are needed.  They could potentially get a 10% improvement in power performance over TSMC's process, plus a few % improvement in density.  It certainly will not be the jump that they were hoping for, but between the latest GCN designs with regards to power consumption, every extra bit will count when competing with the latest Maxwell GPUs from NVIDIA.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.