Nvidia Pascal Speculation Thread

Status
Not open for further replies.
I understand the concerns of "too much" but on the midrange, 1024bit and 4GB is just about right.

I would like to see a 65W card version of that, getting around GTX 680 performance?
With a 35W Skylake on the side.. That would not be ground-breaking perforrmance at all, but would be fun to see in a small and quiet PC.
 
From WCCFTech: "NVIDIA Pascal GPU Will Be Manufactured on TSMC 16nm FF Node – Flagship Single Chip Card To Feature 16 GB HBM2 VRAM."

A lot of it is rehashing older rumors, but there seems to be some new stuff too, at least some new slides.

Furthermore, NVIDIA has their GTC conference being held in Japan on 18th September 2015 and have already released some material in regards to the new announcements that will be made at the graphics focused event.
Click to expand...
NVIDIA-Pascal-GPU_Roadmap.jpg


NVIDIA-Pascal-GPU_Compute-Performance.jpg


Should I interpret the statement "Pascal ≈ Maxwell + Mixed Precision + 3D Memory + NVLink" to mean that Pascal won't have any major changes compared to Maxwell other than those three features?
 
iMacmatician said:
From WCCFTech: "NVIDIA Pascal GPU Will Be Manufactured on TSMC 16nm FF Node – Flagship Single Chip Card To Feature 16 GB HBM2 VRAM."

A lot of it is rehashing older rumors, but there seems to be some new stuff too, at least some new slides.


NVIDIA-Pascal-GPU_Roadmap.jpg


NVIDIA-Pascal-GPU_Compute-Performance.jpg


Should I interpret the statement "Pascal ≈ Maxwell + Mixed Precision + 3D Memory + NVLink" to mean that Pascal won't have any major changes compared to Maxwell other than those three features?
Click to expand...
The other big feature of Pascal is shared virtual memory (paging data automatically between CPU and GPU).
 
iMacmatician said:
Should I interpret the statement "Pascal ≈ Maxwell + Mixed Precision + 3D Memory + NVLink" to mean that Pascal won't have any major changes compared to Maxwell other than those three features?
Click to expand...
IIRC Jen-Hsun said something along those lines already ages ago

edit: and the TSMC-part isn't old rumors, it's actually fresh news from Korea that NVIDIA only just decided that they'll do GPUs (only) at TSMC
 
http://techfrag.com/2015/09/16/nvidia-pascal-gpu-samsung-loses-manufacturing-to-tsmc/

And to say Pascal is Maxwell with extra parts, err that's a bit of an underwhelming statement, this is going to be a very different chip because of the features they are putting into Pascal. Using HBM the memory bus will have to be different, the different precision at once, the shader array has to be different, caching might be changed because of this too, and what ever other tweaks they will do to improve on Maxwell and of course the inclusion of DP again.
 
But will it have async compute.... j/k

Still, I was hoping to see it's DX12 credentials brought more in line with Skylake, or at least to see it go fully bindless.
 
18 months ago...

http://techreport.com/news/26226/nvidia-pascal-to-use-stacked-memory-proprietary-nvlink-interconnect

If all of this info sounds more than a little familiar, perhaps you'll recall that Nvidia also announced a future, post-Maxwell GPU at GTC 2013. It was code-named Volta and was also slated to feature stacked memory on package. So what happened?

Turns out Volta remains on the roadmap, but it comes after Pascal and will evidently include more extensive changes to Nvidia's core GPU architecture.

Nvidia has inserted Pascal into its plans in order to take advantage of stacked memory and other innovations sooner. (I'm not sure we can say that Volta has been delayed, since the firm never pinned down that GPU's projected release date.) That makes Pascal intriguing even though its SM will be based on a modified version of the one from Maxwell. Memory bandwidth has long been one of the primary constraints for GPU performance, and bringing DRAM onto the same substrate opens up the possibility of substantial performance gains.
Click to expand...
 
Probably the biggest question on my mind (which probably nobody outside Nvidia can answer) for the long term, is their Einstein GPU architecture what comes after Volta?

Maybe we'll see Einstein appear on Nvidia's roadmap at GTC next year (not the upcoming GTC in Japan).
 
Megadrive1988 said:
Probably the biggest question on my mind (which probably nobody outside Nvidia can answer) for the long term, is their Einstein GPU architecture what comes after Volta?

Maybe we'll see Einstein appear on Nvidia's roadmap at GTC next year (not the upcoming GTC in Japan).
Click to expand...

Einstein was a research project, not a product. Ideas from Einstein have probably made it into Nvidia product architectures, but I doubt we will ever see a product named Einstein.
 
RecessionCone said:
Einstein was a research project, not a product. Ideas from Einstein have probably made it into Nvidia product architectures, but I doubt we will ever see a product named Einstein.
Click to expand...

Right, I kind of knew that. Einstein (I believe) was the chip implementation of Bill Daily's Echelon project. (Nvidia's Chief Scientist).

Anyway, from the Pascal and Volta architectures, we'll have years of actual products for the rest of this decade.

The presentation 'Nvidia's Path to ExaScale by Bill Daily at the SC14 conference is fascinating.


Starting at about 16:00 he compares the energy cost of doing some operation on Fermi to that of Volta.
 
Last edited:
NVIDIA Confirmed HBM2 Supply From Both Samsung And SK Hynix
Industry sources have stated that both Samsung Electronics and SK Hynix are planning to mass produce second-gen High Bandwidth Memory (HBM) for NVIDIA’s Pascal lineup of graphics processors. The mass production phase is expected to start during the first quarter of 2016, but before that, pilot production and reliability tests completion will be taking place by the end of 2015.
Click to expand...

http://wccftech.com/nvidia-confirmed-supplied-gen-hbm-samsung-sk-hynix/
 
 
Last edited:
Who are these Wccf-clowns? They seem as bad - or if not even worse - than Charlie D and that guy whatsisname who got ousted from The Register all those years ago ever was.
 
That headline is indeed terrible, but I generally like wccftech because:

1 - They always source their news
2 - Most of the time, they put a "Rumored" tag where it's due
3 - They seem unbiased (their coverage of the Async thing was really fair)
4 - they put out the news very fast and they're usually the first place where I see the news coming from


That said, you just need to take in consideration that they're not pros and they don't have informants of their own.
They're just a news aggregation site and they're pretty straight-forward about that.
 
Status
Not open for further replies.

Similar threads

F
ARM Speculation Thread
Replies
8
Views
2K
Arun
A
D
NVIDIA celebrates 25 years since the GeForce 256
Replies
14
Views
2K
Roly
R
F
RDNA5 Speculation
Replies
2
Views
1K
Erinyes
E
T
AMD CDNA: MI300 & MI400 (Analysis, Speculation and Rumors in 2024)
3 4 5
Replies
82
Views
14K
pTmdfx
P
T
GPU Stability testing for Nvidia cards - PowerSweep
Replies
0
Views
445
T2098
T
Back
Top