• 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.

The NVidia GV100 News Thread

Man of Honour
OP
Joined
21 May 2012
Posts
31,940
Location
Dalek flagship
I thoughtI would start a news thread for big Volta as it could get quite interesting if I regularly keep the OPs updated.

If people would like to discuss big Volta and post items that could be used in the OP that would be very useful.:)
 
Soldato
Joined
24 Jun 2004
Posts
10,977
Location
Manchester
Holy Crap.

Tensor cores for mat-mat multiplication? 120TF/s throughput?! If it's practical to extract even a small fraction of this potential, then there are going to be a lot of people in the scientific computing field buying a V100.

I'm working on dense linear algebra right now, so I'm going to need to get my hands on one ASAP. Just need to figure out where best to go begging for the funds :p

I'm sure these won't make it to the G-Force range (minimal utility for gaming), but for a lot of HPC applications this could be a game-changer.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,808
Holy Crap.

Tensor cores for mat-mat multiplication? 120TF/s throughput?! If it's practical to extract even a small fraction of this potential, then there are going to be a lot of people in the scientific computing field buying a V100.

I'm working on dense linear algebra right now, so I'm going to need to get my hands on one ASAP. Just need to figure out where best to go begging for the funds :p

I'm sure these won't make it to the G-Force range (minimal utility for gaming), but for a lot of HPC applications this could be a game-changer.

Good luck with raising $3Bn :p gonna be a few months before they are generally available to industry (and at normal industry prices) unless you have deep pockets.
 
Soldato
Joined
24 Jun 2004
Posts
10,977
Location
Manchester
Good luck with raising $3Bn :p gonna be a few months before they are generally available to industry (and at normal industry prices) unless you have deep pockets.

Probably. But it depends :p

We had access to the new Xeon Phi (Knight's Landing) a good 9 months before they were generally available. Big boss man here has pretty close ties with Intel, as some of the work we do is related to optimising the Intel Math Kernal Library. Not sure if he can get anything from Nvidia, but I've seen stranger things happen.

More realistically, in academia, you mentally add on about a 9 to 18 month delay between trying to get some funding, and actually having it to spend. So if I start searching now, I *might* be able to buy one before it's superseded by V110 :p
 
Caporegime
Joined
18 Oct 2002
Posts
32,615
Would you care to explain this for numpties like me?
For particular types of matrix multiplication and accumulation that can leverage mixed precision floating point, the GV100 is stupidly fast. 10x faster than pascal. The main application will be in deep learning, the GV100 is basically accelerating neural network updates and propagation. This is why Nvidia have specially supported Google's Tensor Flow library.


There is a reasoning companies will be paying $18K per single chip for this beast.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,808
That's why I can't understand why all the Gamers were getting excited about it, We're never going to see a GTX Gaming card with anything close to this, Are we?

No but the tech advancement will likely bring other advantages with it that can be applied to GeForce and likewise the potential stats when stripping out the other stuff gives room for a lot of potential.
 
Back
Top Bottom