[ Various] NVIDIA could launch Pascal-based mobile GPUs by end-2016

[PPB]

How exactly you danger a product by having your ceo have it in the pcb being shown to tje public instead of you current gen mobile product? Why would anyone do that if you have that product on schedule and with at least A0 silicon available?

Some people are just delusional here.

 

[arandomguy]

Because you can get information even if it is just ballpark information that the the company may not want to reveal yet? When initial pictures were released there immediately started to be discussions about why that chip was somewhat large. Now we are back to having zero information on the speculative die size.

When AMD demoed Polaris and showed the GPU die itself they still limited access to control information.

As a quick preface here, while RTG demonstrated a Polaris based card in action we the press were not allowed to see the physical card or take pictures of the demonstration. Similarly, while Raja Koduri held up an unsoldered version of the GPU used in the demonstration, again we were not allowed to take any pictures. So while we can talk about what we saw, at this time it’s all we can do. I don’t think it’s unfair to say that RTG has had issues with leaks in the past, and while they wanted to confirm to the press that the GPU was real and the demonstration was real, they don’t want the public (or the competition) seeing the GPU before they’re ready to show it off.

http://www.anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture

Why did AMD not allow physical board access or pictures? Surely they must have reasons and there is no dispute that there is working silicon in this case. By not allowing pictures we can't use those to estimate die size. Without physical board access we still do not know the memory configuration.

 

[littleg]

Yes. Of course. NV will be straight out of the gate on a new process with new memory and a new uarch with no delays and a free unicorn with every purchase. Absolutely.

 

[MrTeal]

At supercomputing 2015 in november it was mentioned in talks of nv engenieers and i believe this guys way more than any press folks. It seems the Zauba information is right and they have GP100 silicon since July 2015. So GP100 will launch first probably in a few months, of course just in HPC, not consumer. But other gpus might be late because of NVs concentration on the big chip.

They don't have to demo it in public like AMD because the circumstances are different. AMD wants publicity so people wait for polaris before buying a gpu now. Nv is selling extremely well at the moment, so no need to danger this by showing something of pascal.

Were you at Supercomputing 2015, or do you have a link to where nVidia engineers were discussing it?

 

[jpiniero]

At supercomputing 2015 in november it was mentioned in talks of nv engenieers and i believe this guys way more than any press folks. It seems the Zauba information is right and they have GP100 silicon since July 2015. So GP100 will launch first probably in a few months, of course just in HPC, not consumer. But other gpus might be late because of NVs concentration on the big chip.

It makes sense for nVidia to focus on GP100 first due to protecting HPC from Intel and Knights Landing. Wouldn't be surprised that it's ready now but they have to wait for HBM2 availability. But I don't think there's any reason why they wouldn't sell a Titan version, although expect the Titan to be $1500. But yes I would say at least one of the Polaris models would be out before the rest of the GP10x lineup.

 

[prtskg]

TSMC brand new 16nm node and feature size versus Samsung revised 14nm node. Samsung probably takes the edge here, they have been in high volume production of 14nm for almost a full year now.

14nm lpp reached volume production at Samsung just few days ago. 1 year ago lpe reached volume production.
http://forums.anandtech.com/showthread.php?t=2460609

 

[Captain Picard]

Right on, IF they can pull it off with Polaris (beating NV to the punch by a material amount of time), I will be VERY impressed & will become an AMD GPU customer again.

As I have said on these forums, the first co to put out a high end 14/16nm GPU gets my money. I don't care if that is AMD or NV.

I'm with you on this one. Whichever company releases new cards first this year is getting my money. What I've read so far indicates it might be AMD. Haven't had a Radeon since my 4850, which was a solid card for the time I used it.

Looking forward to upgrading from my GTX 660Ti.

 

[PPB]

Because you can get information even if it is just ballpark information that the the company may not want to reveal yet? When initial pictures were released there immediately started to be discussions about why that chip was somewhat large. Now we are back to having zero information on the speculative die size.

When AMD demoed Polaris and showed the GPU die itself they still limited access to control information.

http://www.anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture

Why did AMD not allow physical board access or pictures? Surely they must have reasons and there is no dispute that there is working silicon in this case. By not allowing pictures we can't use those to estimate die size. Without physical board access we still do not know the memory configuration.

You think companies of this caliber dont get access to some information of their competitors beforehand apart from public disclosements from those competitors? How do you think competing products end up stacked so closely between them beforehand? Because companies have access to at least an aproximation of their competitor impending launch product's characteristics and can bin and tune their own to compete more effectively, and also adjust their PR approach for better sell their own products (or in some sad cases, dismissing the competitor's)

Reality tends to be a lot simpler and should be that if NV had at least A0 silicon to show on the PX board, they would have. Having your CEO look like a fool showing something that is NOT what you are claiming it is when lots of closeup takes were done showing the truth. In that sense, this is a lot like the bumpgate. A publicly traded company doesnt fall for what would be otherwise childish mistakes if those could be avoided.

 

[nvgpu]

https://www.zauba.com/import-graphics-processor-integrated-circuits-hs-code.html

GRAPHICS PROCESSOR INTEGRATED CIRCUITS BGLW5
GRAPHICS PROCESSOR INTEGRATED CIRCUITS NT0EH
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 433X3
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 2E73U
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 3R08A
GRAPHICS PROCESSOR INTEGRATED CIRCUITS, JM601

http://www.3dcenter.org/news/reihen...cht-gp100-gp102-gp104-gp106-gp107-gp10b-gv100

GP100, GP102, GP104, GP106, GP107, GP10B already referenced in drivers.

http://www.geeks3d.com/forums/index.php/topic,4247.0.html

Pascal

-D__CUDA_ARCH__=600

-D__CUDA_ARCH__=610

-D__CUDA_ARCH__=620

http://www.gputechconf.com/

Nvidia doesn't have to show you anything. Nvidia will demonstrate their products to their customers in April at GTC 2016.

 

[Pariah]

Reality tends to be a lot simpler and should be that if NV had at least A0 silicon to show on the PX board, they would have. Having your CEO look like a fool showing something that is NOT what you are claiming it is when lots of closeup takes were done showing the truth. In that sense, this is a lot like the bumpgate. A publicly traded company doesnt fall for what would be otherwise childish mistakes if those could be avoided.

If Nvidia was unveiling Pascal and showed a Maxwell based video card instead, you would have a point. Nvidia was not unveiling Pascal here, they were demonstrating a non-consumer product they were working on and showed a mock up of it.

Similarly, while Raja Koduri held up an unsoldered version of the GPU used in the demonstration, again we were not allowed to take any pictures....I don’t think it’s unfair to say that RTG has had issues with leaks in the past, and while they wanted to confirm to the press that the GPU was real and the demonstration was real, they don’t want the public (or the competition) seeing the GPU before they’re ready to show it off.

This is AMD's method of not revealing anything. Nvidia says you can take all the pictures you want, since we aren't revealing Pascal yet and this isn't a Pascal GPU in our mock up. Two different methods of getting to the same end result. Neither company is going to tip its hand before it has to. Whining about one method while having no problem with the other is hypocritical.

 

[KaRLiToS]

If Nvidia was unveiling Pascal and showed a Maxwell based video card instead, you would have a point. Nvidia was not unveiling Pascal here, they were demonstrating a non-consumer product they were working on and showed a mock up of it.



This is AMD's method of not revealing anything. Nvidia says you can take all the pictures you want, since we aren't revealing Pascal yet and this isn't a Pascal GPU in our mock up. Two different methods of getting to the same end result. Neither company is going to tip its hand before it has to. Whining about one method while having no problem with the other is hypocritical.

Why are you trying to defend Nvidia over here? Many tech sites released articles about how dishonest it was for Nvidia to do that... and even illegal.

 

[3DVagabond]

https://www.zauba.com/import-graphics-processor-integrated-circuits-hs-code.html

GRAPHICS PROCESSOR INTEGRATED CIRCUITS BGLW5
GRAPHICS PROCESSOR INTEGRATED CIRCUITS NT0EH
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 433X3
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 2E73U
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 3R08A
GRAPHICS PROCESSOR INTEGRATED CIRCUITS, JM601

http://www.3dcenter.org/news/reihen...cht-gp100-gp102-gp104-gp106-gp107-gp10b-gv100

GP100, GP102, GP104, GP106, GP107, GP10B already referenced in drivers.

http://www.geeks3d.com/forums/index.php/topic,4247.0.html

Pascal

-D__CUDA_ARCH__=600

-D__CUDA_ARCH__=610

-D__CUDA_ARCH__=620

http://www.gputechconf.com/

Nvidia doesn't have to show you anything. Nvidia will demonstrate their products to their customers in April at GTC 2016.

I see GM204. GM206, and GK208. Am I missing the rest.

I'll add the link I'm talking about so as not to cause confusion. https://www.zauba.com/import-graphics-processor-integrated-circuits-hs-code.html

 

[Techhog]

https://www.zauba.com/import-graphics-processor-integrated-circuits-hs-code.html

GRAPHICS PROCESSOR INTEGRATED CIRCUITS BGLW5
GRAPHICS PROCESSOR INTEGRATED CIRCUITS NT0EH
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 433X3
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 2E73U
GRAPHICS PROCESSOR INTEGRATED CIRCUITS 3R08A
GRAPHICS PROCESSOR INTEGRATED CIRCUITS, JM601

http://www.3dcenter.org/news/reihen...cht-gp100-gp102-gp104-gp106-gp107-gp10b-gv100

GP100, GP102, GP104, GP106, GP107, GP10B already referenced in drivers.

http://www.geeks3d.com/forums/index.php/topic,4247.0.html

Pascal

-D__CUDA_ARCH__=600

-D__CUDA_ARCH__=610

-D__CUDA_ARCH__=620

http://www.gputechconf.com/

Nvidia doesn't have to show you anything. Nvidia will demonstrate their products to their customers in April at GTC 2016.

Having a bit of a crisis worried that AMD might be first?

 

[Pariah]

Why are you trying to defend Nvidia over here? Many tech articles released articles about how dishonest it was for Nvidia to do that... and even illegal.

Illegal? Please say that was a joke, because that's hilarious. Nvidia could stand up there and say the PX2 is delicious on a kaiser roll with cheese and mustard and that it will cure blindness if you lick it just right without having to worry about whether any of that is actually true. They weren't making a sales pitch. Volvo has already signed on, while Audi, BMW, Daimler and Ford have all helped with the development of the product. You think none of these companies have worked with the actual product yet while somehow already knowing how it performs in real world testing? The companies that might actually use this product have known about it far longer than you have, and Nvidia has no legal obligation to show the actual product to internet engineers like you.

 

[3DVagabond]

Illegal? Please say that was a joke, because that's hilarious. Nvidia could stand up there and say the PX2 is delicious on a kaiser roll with cheese and mustard and that it will cure blindness if you lick it just right without having to worry about whether any of that is actually true. They weren't making a sales pitch. Volvo has already signed on, while Audi, BMW, Daimler and Ford have all helped with the development of the product. You think none of these companies have worked with the actual product yet while somehow already knowing how it performs in real world testing? The companies that might actually use this product have known about it far longer than you have, and Nvidia has no legal obligation to show the actual product to internet engineers like you.

No I don't. You have anyone from these companies saying they have? Anyone from any other company? Anyone saying they've seen one in the flesh?

All we have is Jen-Hsun doing is smoke and mirrors routine again. I would love to see something substantial.

 

[jpiniero]

Having a bit of a crisis worried that AMD might be first?

AMD was first on 28 nm (by almost 3 months btw), didn't end up mattering much.

 

[Pariah]

No I don't. You have anyone from these companies saying they have? Anyone from any other company? Anyone saying they've seen one in the flesh?

All we have is Jen-Hsun doing is smoke and mirrors routine again. I would love to see something substantial.

There were multiple examples given of what the system had already achieved in testing and which companies had performed the tests. If you've seen any of the companies publicly state those claims were made up, feel free to link to them here.

 

[nvgpu]

The only people having a crisis are the FUDers and slanders that don't know anything but keep believing Charlie Dermerjian the big fat bald loser lies. This is the same person that can't even tell the GM200 from the GM204.

http://forums.anandtech.com/showpost.php?p=37443058&postcount=227

Nvidia has taped out multiple Pascal GPUs since last year while the competitor can only barely tape out 2 GPUs for release this year, 1 notebook GPU with GDDR5 which won't have the performance or bandwidth for enthusiasts in the desktop version and 1 GPU which has to wait for HBM2 mass production availability anyway.

 

[Vesku]

If Nvidia was unveiling Pascal and showed a Maxwell based video card instead, you would have a point. Nvidia was not unveiling Pascal here, they were demonstrating a non-consumer product they were working on and showed a mock up of it.

This is AMD's method of not revealing anything. Nvidia says you can take all the pictures you want, since we aren't revealing Pascal yet and this isn't a Pascal GPU in our mock up. Two different methods of getting to the same end result. Neither company is going to tip its hand before it has to. Whining about one method while having no problem with the other is hypocritical.

Did you watch the video of JHH showing off PX2? As JHH flips from Tegra side to GPU side:

On the back two next generation Nvidia GPUs based on the Pascal architecture

https://youtu.be/Md6KvHrVbMM?t=1342

If we take him at his word, JHH thought he was giving the audience and their cameras a glimpse of Pascal.

 

[dark zero]

Right on, IF they can pull it off with Polaris (beating NV to the punch by a material amount of time), I will be VERY impressed & will become an AMD GPU customer again.

As I have said on these forums, the first co to put out a high end 14/16nm GPU gets my money. I don't care if that is AMD or NV.

They are not allowed to fail again since if they fails is game over for them.

 

[Pariah]

Did you watch the video of JHH showing off PX2? As JHH flips from Tegra side to GPU side:



https://youtu.be/Md6KvHrVbMM?t=1342

If we take him at his word, JHH thought he was giving the audience and their cameras a glimpse of Pascal.

I am aware of what he said. I didn't mean Nvidia literally said you can take pictures because this isn't Pascal. That would be idiotic, but par for the course here.

 

[MrTeal]

I see GM204. GM206, and GK208. Am I missing the rest.

I'll add the link I'm talking about so as not to cause confusion. https://www.zauba.com/import-graphics-processor-integrated-circuits-hs-code.html

The supposition is that the one of August 15th must be GP100, because the value is so high.

 

[JDG1980]

It makes sense for nVidia to focus on GP100 first due to protecting HPC from Intel and Knights Landing. Wouldn't be surprised that it's ready now but they have to wait for HBM2 availability. But I don't think there's any reason why they wouldn't sell a Titan version, although expect the Titan to be $1500.

Yields. Up until now, the biggest thing TSMC has made on 16FF+ is the Apple A9X, which is only 147 mm^2. Do you really think they can get even half-decent yields on a 500+ mm^2 die at this time? Remember that on 28nm, yields weren't even good enough for a public release of Tesla cards until November 2012, eight months after GK104 and nearly a year after AMD launched their first 28nm product. Before that, the tiny allocation of working GK110 chips all went to Oak Ridge. And yields didn't get good enough for a Titan gaming card until February 2013.

We could potentially see a really unusual situation this year if Nvidia focuses on the high end while AMD concentrates on a small-die strategy. Imagine if Nvidia releases a Tesla-only GP100 (~500 mm^2) and a consumer-focused GP104 (~350 mm^2), the latter of which would have about twice the raw power of GM204 and beat GTX 980 Ti by 25-35% in most real-world gaming applications. Meanwhile, AMD releases Polaris 10 at ~125 mm^2 and Polaris 11 at about double that, with Polaris 10 falling between Pitcairn and Tonga in performance and Polaris 11 roughly matching Hawaii. We could conceivably have a situation where each company has one end of the market sewn up for a while, with no real competition from the other.

 

[jpiniero]

Yields. Up until now, the biggest thing TSMC has made on 16FF+ is the Apple A9X, which is only 147 mm^2. Do you really think they can get even half-decent yields on a 500+ mm^2 die at this time?

I don't think GP100 is anywhere near 500 mm2. I am thinking in the 350-400 range. Even then, they'll just have to deal and be creative in binning... plus charging $1500 for a Titan and $5K+ for Teslas (2 die) will help for sure. GP104 is probably close to Polaris 11's size, maybe a bit bigger. GP102 is probably in the 300 range.

Yes, I know. You should downgrade your performance expectations for this generation pretty significantly unless you are willing to pay.

 

[PPB]

AMD was first on 28 nm (by almost 3 months btw), didn't end up mattering much.

This doesnt seem like that time, this seems more like how they were first with a full new generation of Skus on 40nm for almost half a year.