[SA] News of Nvidia’s Pascal tapeout and silicon is important

[raghu78]

http://semiaccurate.com/2016/02/01/news-of-nvidias-pascal-tapeout-and-silicon-is-important/

A scathing article on the so called articles on the web regarding Pascal tapeout. An interesting point to debate is does this mean AMD has a 3-4 month lead in time to market on FINFET GPUs. Looking forward to a strong contest in the next generation GPUs as AMD failed to show up for the fight last gen.

 

[ShintaiDK]

Assuming you believe the author and the "site".

 

[Azix]

I wouldnt be surprised if they are late. Ultimately nvidia is second class to AMD and intel. AMD just happened to fail miserably selling CPUs so they ended up where they are. They still seem more capable at introducing new tech.

 

[PPB]

Assuming you believe the author and the "site".

He sure has delivered with a lot more of predictions that you did with yours (that are even sig worthy for some users around here). That is for sure.

 

[xthetenth]

Assuming you believe the author and the "site".

At least he has decent info to draw conclusions that aren't much better than random chance from.

I mean we know what his conclusion's going to be, what's in question is how much evidence he's got to get there with, and this isn't bad for his oeuvre.

 

[NTMBK]

Assuming you believe the author and the "site".

No belief needed. In this article he takes the same evidence we all have and just applies common sense, unlike the other tech sites who reported on this.

 

[tviceman]

Charlie is basing his premise on Nvidia showing pre-production silicon to the public. When was the last time Nvidia did this with GPU's? I'm sure it's happened, but I can't remember when and what GPU it was, which goes to show that Nvidia hardly ever shows off pre-production GPU's. They show off Tegra early several times because they were trying to become an important supplier in the SoC space (which ultimately failed).

Case in point, when you are the undisputed leader in a particular market and not feeling significant pressure from the competition, you don't have to show your hand and attempt to drum up excitement. Also, this popped up 3 days ago: http://www.kotaku.com.au/2016/01/rumours-nvidia-will-unveil-their-flagship-pascal-gpu-in-april/ The article states GP100 comes in April, and the rest come starting in June. If so, that will mark a shift back to big die first, which will be impressive if true. Assuming Nvidia shows off pre-production silicon in April and doesn't mention release dates, then yes Nvidia has a problem. Otherwise it's business as usual.

 

[xthetenth]

The combination of showing silicon, the silicon that got shown, and what's in that shipping manifest are compelling together. It's entirely possible that it might happen sooner (IE they showed different silicon because it's more impressive and a feeling that AMD would be executing but only with smaller chips, and the testing equipment is for a chip after the first (few)), but I think it's not the likeliest outcome.

 

[FatherMurphy]

Also, didn't Bill Dally (NVIDIA Chief Scientist) say in a presentation at SC15 (November 2015) that Pascal silicon was "in the lab." I found the slides he used but can't find the video link right now.

http://images.nvidia.com/events/sc15/pdfs/SC5102-path-exascale-computing.pdf

 

[DooKey]

Aren't you supposed to put the source in the title? I wouldn't have bothered to click on this thread if the rules had been followed.

 

[Silverforce11]

You guys don't have to think too hard.

Ask yourself, when was the last time JHH had to stand in front of media and investors and show them a fake product and why did it happened?

Really it is an embarrassment for their CEO to do that, why would he have to do that barring exceptional circumstances?

 

[coercitiv]

Case in point, when you are the undisputed leader in a particular market and not feeling significant pressure from the competition, you don't have to show your hand and attempt to drum up excitement.

An undisputed leader, yet their language is... ambiguous. One could almost argue they did feel a particular pressure to show their hand and attempt to drum up excitement.

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

Nvidia may very well be on their way for a properly scheduled launch of their next gen GPUs, but no public information available is able to support that, and the article in the OP exploits just that.

 

[FatherMurphy]

You guys don't have to think too hard.

Ask yourself, when was the last time JHH had to stand in front of media and investors and show them a fake product and why did it happened?

Really it is an embarrassment for their CEO to do that, why would he have to do that barring exceptional circumstances?

Haven't we gotten past this misleading comparison? In 2009 (or whenever), JHH was actually demonstrating the Fermi cards and arguably misled everyone about the state of production of Fermi.

Here, JHH was not marketing Pascal, he was announcing the Drive PX 2 with availability much later in the year. Pascal was not the focal point (note the complete lack of details about the GPUS, other than some FLOP numbers that don't distinguish between dGPU, iGPU, and CPU cores). If you are telling me that the shipping Drive PX 2 will have Maxwell, not Pascal, you are onto something. But the device he held up was a prop relating to car hardware, not a demonstration of the Pascal GPU.

 

[3DVagabond]

Charlie is basing his premise on Nvidia showing pre-production silicon to the public. When was the last time Nvidia did this with GPU's? I'm sure it's happened, but I can't remember when and what GPU it was, which goes to show that Nvidia hardly ever shows off pre-production GPU's. They show off Tegra early several times because they were trying to become an important supplier in the SoC space (which ultimately failed).

Case in point, when you are the undisputed leader in a particular market and not feeling significant pressure from the competition, you don't have to show your hand and attempt to drum up excitement. Also, this popped up 3 days ago: http://www.kotaku.com.au/2016/01/rumours-nvidia-will-unveil-their-flagship-pascal-gpu-in-april/ The article states GP100 comes in April, and the rest come starting in June. If so, that will mark a shift back to big die first, which will be impressive if true. Assuming Nvidia shows off pre-production silicon in April and doesn't mention release dates, then yes Nvidia has a problem. Otherwise it's business as usual.

That's not what he's basing the premise on. The main thing he's basing it on is that the testing equipment has just been shipped (at the time) and these sites were claiming it was Pascal silicon. That is really what the whole article is about. He's using the "faked silicon" as more evidence that they don't in fact have any. Like you I'll ask, When was the last time we saw nVidia do this? (Rhetorical question, as we all know.)

 

[Silverforce11]

@FatherMurphy

He held a product up and claimed the GPUs are Pascal.

"Two next generation nvidia GPUs based on the Pascal architecture!"

https://www.youtube.com/watch?v=2ZHxGyQqQw4&feature=youtu.be&t=133

Do you really think if they actually had Pascal ready for that engineering sample, they wouldn't use it??

 

[3DVagabond]

Also, didn't Bill Dally (NVIDIA Chief Scientist) say in a presentation at SC15 (November 2015) that Pascal silicon was "in the lab." I found the slides he used but can't find the video link right now.

http://images.nvidia.com/events/sc15/pdfs/SC5102-path-exascale-computing.pdf

And JHH claimed to have Pascal silicon in his hand.

 

[FatherMurphy]

I would add that the only statement that leads me to believe Nvidia has working silicon is Bill Dally's comment at SC15. The Zauba stuff... even Charlie assumes for the purposes of his article that the Zauba information applies to Pascal, but he doesn't seem convinced.

 

[sontin]

Also, didn't Bill Dally (NVIDIA Chief Scientist) say in a presentation at SC15 (November 2015) that Pascal silicon was "in the lab." I found the slides he used but can't find the video link right now.

http://images.nvidia.com/events/sc15/pdfs/SC5102-path-exascale-computing.pdf

Minute 16: http://images.nvidia.com/events/sc15/SC5102-path-exascale-computing.html

Do you really think if they actually had Pascal ready for that engineering sample, they wouldn't use it??

Why didnt AMD allow the journalist to take photos of the die?

 

[coercitiv]

Here, JHH was not marketing Pascal, he was announcing the Drive PX 2 with availability much later in the year. Pascal was not the focal point (note the complete lack of details about the GPUS, other than some FLOP numbers that don't distinguish between dGPU, iGPU, and CPU cores).

Had those GPUs not been proven to be current gen, would you consider them as proof Pascal is ready?

 

[AnandThenMan]

Haven't we gotten past this misleading comparison? In 2009 (or whenever), JHH was actually demonstrating the Fermi cards and arguably misled everyone about the state of production of Fermi.

There was never any direct evidence that an actual Fermi GPU was running somewhere unseen, but a whole bunch of evidence that it was at best a hacked up prototype that in no way resembled a stand alone GPU. If Fermi was actually up and running no way does JHH show a fake board with wood screws and the end sawed off, let's use some common sense here.

Same applies with his "showing" of Pascal.

 

[FatherMurphy]

@FatherMurphy

He held a product up and claimed the GPUs are Pascal.



https://www.youtube.com/watch?v=2ZHxGyQqQw4&feature=youtu.be&t=133

Do you really think if they actually had Pascal ready for that engineering sample, they wouldn't use it??

I get that. But he was showing off the Drive PX 2 as a prototype of hardware that would be available in half a year. He was stating what the Drive PX 2 would have. Do you think that those were actually next-gen Tegras on the board?

I would acknowledge that the fact that the demos (correct me if I am wrong) shown to the public on Drive PX 2 were being done by a Titan X could suggest that they don't currently have in sufficient quantity the specific Pascal GPU intended for the PX 2. With regard to the Big Pascal rumors, though, the lack of the PX 2 Pascal wouldn't have any bearing on the status of Big Pascal, as Big P won't be on the PX 2.

 

[Kenmitch]

And JHH claimed to have Pascal silicon in his hand.

He didn't lie. The company that made the silicone that was holding the acrylic box together was named Pascal. Very misleading but it's the nature of the biz.

You can buy Pascal silicone here if your really want to be the 1st one on the block to have some.

http://www.pascal-rosier.com/boutique.asp?Idbranche=128#

 

[Silverforce11]

Why didnt AMD allow the journalist to take photos of the die?

If you want to talk about AMD, you can start a new thread on why they didn't showcase Polaris properly according to you.

Journalists were given access to see it themselves to confirm the die size.

 

[Silverforce11]

I get that. But he was showing off the Drive PX 2 as a prototype of hardware that would be available in half a year. He was stating what the Drive PX 2 would have. Do you think that those were actually next-gen Tegras on the board?

I would acknowledge that the fact that the demos (correct me if I am wrong) shown to the public on Drive PX 2 were being done by a Titan X could suggest that they don't currently have in sufficient quantity the specific Pascal GPU intended for the PX 2. With regard to the Big Pascal rumors, though, the lack of the PX 2 Pascal wouldn't have any bearing on the status of Big Pascal, as Big P won't be on the PX 2.

That's not his language. You are interpreting what he said when he clearly say in the present tense.

With "this right here, 6x the deep learning throughput compared to Titan X!" etc. he's holding "8 tflops of performance!" etc.

The question you haven't answered and the crucial one, if they had Pascal chips ready for the Drive PX2, would they show that or a fake board?

 

[sontin]

If you want to talk about AMD, you can start a new thread on why they didn't showcase Polaris properly according to you.

Journalists were given access to see it themselves to confirm the die size.

You are bringing it up. AMD hasnt shown Polaris to the public. Only a few guys were allowed to look at a Chip which AMD named Polaris. They werent even allowed to look into the PCs.