Overclock.net banner

1 - 13 of 13 Posts

·
Registered
Joined
·
724 Posts
Discussion Starter #1
Quote:
To repeat what we concluded earlier, FP64 is completely useless for graphics, and it takes up a lot of die area. That's especially true for the dedicated SIMDs needed to run FP64 alongside the main FP32-and-FP16 pipeline, as with the GP100 design. To keep costs down for consumers, I'm expecting Nvidia to effectively remove FP64 in the chips that arrive to power GeForce models. It'll still be there because it can't disappear completely, but it'll probably just be 1/32-rate like we got in GM200.

Then there's HBM2. I'd have argued for its inclusion in GeForce Pascals a few months ago, but GDDR5X is on the way. This memory doubles the prefetch length and also should come with a fairly large increase in effective clock speed. It'll be cheaper to use than HBM2 at similar aggregate bandwidths, and it's cheaper to implement at the on-chip PHY level-not to mention the savings from the lack of an interposer and stack packaging. GDDR5X also doesn't have strict rules tying bandwidth to capacity. That lets Nvidia use memory sizes other than 4GB, 8GB, 12GB, or 16GB on its GeForce products, compared to the limitations of HBM2.

Given those guesses, I think there's at least one consumer chip that's still really big, but quite a bit smaller than 610 mm². It probably has similar overall throughput to GP100 in the metrics we care about for graphics, and it'll probably come with less memory capacity. Even so, it should still have plenty of overall bandwidth. Some rumours say this chip is called GP102. I think it'll have 56 to 60 SMs, 1/32nd FP64 throughput, and more than 8GB of 384-bit GDDR5X. If it exists, then it's likely destined for a Titan-class card first, and maybe a enthusiast's favourite "Ti" product later on.

Nvidia is also likely working on a GM200 replacement for the pair of high-end GeForce non-Tis that make up the meat of the enthusiast market these days. It's likely called GP104. That chip will likely also have token FP64 throughput-remember that these are GPUs, not HPC cards. I also bet it'll have 8GB of 256-bit GDDR5X, 40 SMs or thereabouts, and all the associated machinery in terms of texturing and backend throughput that implies, in a die of around 300 mm².
Source

And now Rys Sommefeldt has the same opinion that I did when I first saw Pascal. There'll be a GP102 chip that's similar to GP100 in a lot of ways, but it won't be nearly as powerful, and it probably won't feature HBM v2. And it won't be 610mm².
 

·
Registered
Joined
·
724 Posts
Discussion Starter #4
Quote:
Originally Posted by 0razor1 View Post

So we're not looking at radically different schedulers and innards as compared to maxwell 2?
I think that's the implication, now that I've finished reading through the article more thoroughly. Pascal is a reworked Maxwell with an emphasis on working FP64 capabilities back into the Quadro and Tesla families, and figuring out HBM v2 and interposer technology. That's not a bad thing by any means, and we'll still see performance jumps thanks to the new process tech and supposedly higher frequencies, but it'll disappoint anyone who thought it would be a radical shift.

Volta will do that, I think, being a completely new architecture on a process that by then will be more mature. Pascal just lays the groundwork for NVIDIA to follow it up with something better in 2017.
 

·
Premium Member
Joined
·
10,978 Posts
Quote:
Originally Posted by Mand12 View Post

Am I wrong in seeing Maxwell and Pascal as similar to ye olde Tick-Tock?
Yep
 

·
Frog Blast The Vent Core
Joined
·
6,114 Posts
I was hoping for, you know, discussion. Not just telling me I'm wrong and not bothering to explain why.
 

·
I <3 narcissists
Joined
·
6,492 Posts
What a junk article. Titled like it is something useful and informative but it's just an overview involving a card that is meaningless to the GeForce audience and a bunch of rumors mashed together. People complain about WCCF a lot but at least they don't try to hide what their stories are.
 

·
Super Moderator
Joined
·
3,662 Posts
Thread moved, since this is really more of an editorial than news.
 

·
Premium Member
Joined
·
10,885 Posts
Quote:
Originally Posted by bigjdubb View Post

What a junk article. Titled like it is something useful and informative but it's just an overview involving a card that is meaningless to the GeForce audience and a bunch of rumors mashed together. People complain about WCCF a lot but at least they don't try to hide what their stories are.
Yeah, should be titled Theorizing Pascal Architecture instead. There's no exploration there, it's all just speculation.
 

·
Registered
Joined
·
724 Posts
Discussion Starter #11
Quote:
Originally Posted by superhead91 View Post

Thread moved, since this is really more of an editorial than news.
Agreed, and thank you. I'll keep this in mind in the future for editorials.
 

·
Super Moderator
Joined
·
3,662 Posts
Quote:
Originally Posted by CataclysmZA View Post

Agreed, and thank you. I'll keep this in mind in the future for editorials.
:thumb:
 

·
Registered
Joined
·
1 Posts
Quote:
Originally Posted by bigjdubb View Post

What a junk article. Titled like it is something useful and informative but it's just an overview involving a card that is meaningless to the GeForce audience and a bunch of rumors mashed together. People complain about WCCF a lot but at least they don't try to hide what their stories are.
Sorry you didn't like it, I tried to make it relevant to GeForce customers after the refresher on Maxwell, using GP100 to get context for what might happen in the GeForce products. I even set it up in the intro like that, so you wouldn't be hoodwinked into thinking it was a review. It's a pre-release look at the microarchitecture using public information, with a view to using that to try and guess at GeForce configurations. I got the first chip config right, too.
Quote:
Originally Posted by superhead91 View Post

Thread moved, since this is really more of an editorial than news.
There is unique content in there that hasn't been published before (how the FP16 hardware likely works), but you're right, it's mostly editorial.
Quote:
Originally Posted by Forceman View Post

Yeah, should be titled Theorizing Pascal Architecture instead. There's no exploration there, it's all just speculation.
Yep, sorry the title didn't make that more clear. I explained the piece's reason for being in the intro, though, to save people clicking through to something they didn't want to read.

Thanks for reading
smile.gif
 
1 - 13 of 13 Posts
Top