• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

NVIDIA Volta with GDDR6 in early 2018?

Looks like it is about AI/Deep learning, which points to Quadro cards. I will watch it all, as pretty interesting and the quality is good here and no pauses either!
 
Tesla V100 is here! The sheer size of it! 812mm2!

15 FP32 TFLOPS.
PdzpLKa.png
 
Rroff said:
Hence why it wasn't possible before 12FF and really envisioned for 10nm. Architecturally it is very different to the P100.
Click to expand...

Most because of those new Tesnor TFLOP tech added, purely for Deep Learning and A.I.

Looking at FP32 at 15 TFLOPS, it's not the jump I was expecting compared to Vega 10's 12.5 TFLOPS on the MI25 Instinct at the moment.

I can't imagine how expensive and difficult it must be to produce that behemoth die.

Also It's due Q4, and early 2018 for the none Rack DGX-1
 
LoadsaMoney said:
Wonder how many will be cancelling their pre-ordered Vega' for this, when it arrives before they get their cards :p
Click to expand...

Only those that have the option of taking out another mortgage. I wonder what this will translate to when it eventually reaches the consumers.
 
Rroff said:
Hence why it wasn't possible before 12FF and really envisioned for 10nm. Architecturally it is very different to the P100.
Click to expand...
Biggest update since Fermi really. The new tensor operation looks to do 2xFP16 MUL and FP32 add at insane rates to give the 120TFLOPs for appropriate mixed precision matrix multiplication, perfect for deep learning.

Definitely a GPU designed for 10nm but squeezed into the 12nm half-node shrink 810mm is huge but acceptable for such a high margin part.
 
N19h7m4r3 said:
Most because of those new Tesnor TFLOP tech added, purely for Deep Learning and A.I.

Looking at FP32 at 15 TFLOPS, it's not the jump I was expecting compared to Vega 10's 12.5 TFLOPS on the MI25 Instinct at the moment.

I can't imagine how expensive and difficult it must be to produce that behemoth die.

Also It's due Q4, and early 2018 for the none Rack DGX-1
Click to expand...


GV100 isn't designed to push the boundaries of Fp32 performance though, it has to dedicate die area to FP64 which Vega lacks, and there are then lots of other architectural changes such as the mixed-precision tensor ops. Apples to oranges comaprisons really. As Doom112 says, a consumer gaming version of this architecture might be at 20TFPs for example.
 
D.P. said:
GV100 isn't designed to push the boundaries of Fp32 performance though, it has to dedicate die area to FP64 which Vega lacks, and there are then lots of other architectural changes such as the mixed-precision tensor ops. Apples to oranges comaprisons really. As Doom112 says, a consumer gaming version of this architecture might be at 20TFPs for example.
Click to expand...

20 TFLOPS is very optimistic if we compare P100 vs GP102 for TFLOPs.
 
You must log in or register to reply here.
Back
Top Bottom