Nvidia Volta Speculation Thread

For one thing, it is common that people run CPU-only code on the supercomputers or clusters, not using the 10,000+ GPUs whatsoever :D
Or so did I read, same thing with the Xeon Phi monster in China.
Perhaps some GPU-heavy, networking-lightweight backburner things can be run at the same time?
 
I wish it would be as simple as that.
Did you ever try to code an efficient fluid solver on multi-core CPU and ported it to GPU ?
Yeah which is why Nvidia has an advantage in some ways over AMD but still playing catch-up in certan markets against Intel; CUDA and importantly their ties to other industry platform-resource management-etc related software and specific scientific-analytics-modelling,etc type software helps Nvidia become more competitive.
And also one reason it takes a long time also in terms of training and code migration sometimes from one system to a new one if the approach is different, but most using Summit should be a bit familiar with the massively parallel CPU-GPU fat node due to their time with Titan.
Worth noting Intel is seeing the same training/coding migration requirements with HPC projects involving Xeon based supercomputers upgrading to most recent KNL Phi.
Cheers

Edit:
Pretty old news now but some may not had seen any of the HPC Tesla presentations and sort of fits with the recent discussions.
This has some of the info regarding Titan with what it does-used for and then the upgrade with Summit (Volta).
There are others with different info but would take time to find them online.
2015 GPUTechConf:
HETEROGENEOUS HPC, ARCHITECTURAL OPTIMIZATION, AND NVLINK : http://on-demand.gputechconf.com/gtc/2015/presentation/S5649-Steve-Oberlin.pdf

Cheers
 
Last edited:
Yeah which is why Nvidia has an advantage in some ways over AMD but still playing catch-up in certan markets against Intel; CUDA and importantly their ties to other industry platform-resource management-etc related software and specific scientific-analytics-modelling,etc type software helps Nvidia become more competitive.
And also one reason it takes a long time also in terms of training and code migration sometimes from one system to a new one if the approach is different, but most using Summit should be a bit familiar with the massively parallel CPU-GPU fat node due to their time with Titan.
Worth noting Intel is seeing the same training/coding migration requirements with HPC projects involving Xeon based supercomputers upgrading to most recent KNL Phi.
Cheers

Edit:
Pretty old news now but some may not had seen any of the HPC Tesla presentations and sort of fits with the recent discussions.
This has some of the info regarding Titan with what it does-used for and then the upgrade with Summit (Volta).
There are others with different info but would take time to find them online.
2015 GPUTechConf:
HETEROGENEOUS HPC, ARCHITECTURAL OPTIMIZATION, AND NVLINK : http://on-demand.gputechconf.com/gtc/2015/presentation/S5649-Steve-Oberlin.pdf

Cheers

With avx-512 actually Xeon and KNL have converged :)
 
With avx-512 actually Xeon and KNL have converged :)
Hehe true in some ways including 'legacy' code support but from the optimisation and certain code migration they are further apart now, such as offloading feature and other programming models.
I think I have the Stampede migration info somewhere but should be online to search for.
Separate to that here is a classic example of why code migration and optimisation is needed for those upgrading to a KNL architecture in HPC environments.
Multi-threaded ATLAS Simulation on Intel Knights Landing Processors: https://cds.cern.ch/record/2220833/files/ATL-SOFT-SLIDE-2016-739.pdf

Also shows scientists-coders do take optimisation seriously, in answer to a poster earlier about need of 10x more performance with Summit.
Cheers
 
Faster hardware is always appreciated, if not requiring massive architectural overhaul.
The latter still may be needed for longer term benefit.
Innovative algorithms should not be underestimated too, or at least be mentioned more.
 
Could Nvidia be prepping a Volta release?

So currently I am on a business trip and I get a message over facebook from somebody that I do not know personally. He posts that Volta just appeared in HW IDs. After the break more.

So here is literally what the source mentions:

"First sign of the new nVIDIA flagship GPU, codenamed Volta has appeared on the horizon and its PCI device ID is 1D81 = Graphics Device [GV100]"

Volta ladies and gents is a next-gen Nvidia series, indeed due for 2017 but not expected until maybe Q2/Q3?. These products have HBM2 memory. Coincidentally next week Nvidia definitely is going to announce something. Everybody is expecting the GeForce GTX 1080 Ti based on Pascal. But could it be that Nvidia already has Volta in their rendering pipeline?
http://www.guru3d.com/news-story/could-nvidia-be-prepping-a-volta-release.html
 
They're releasing GTX 1080 Ti based on GP102 at the end of this month. I'd bet there won't be Volta, or at least not one for consumers, 'till about a year from now

Yes, surely, a good presentation of Volta, open HPC parts available on contracts ( who will delivered.. well ) .. and well 2018 for gaming parts. As previously with Maxwell and Pascal. Untill something have changed in between.. Good marketing, who have work so far ..
 
They're releasing GTX 1080 Ti based on GP102 at the end of this month. I'd bet there won't be Volta, or at least not one for consumers, 'till about a year from now


Why not, 980ti was released in June of 2015, May of 2016. That is 11 months, granted 2 months less of a gtx 1180 from a 1080ti (if its launched), but still, doesn't seem they will be holding Volta back (unless there are birthing problems)
 
A more immediate concern is that the Volta GPU is going to be the computational heart of Summit and Sierra, two of the upcoming pre-exascale supercomputers that the US Department of Energy is deploying under the agency’s CORAL (Collaboration of Oak Ridge, Argonne, and Lawrence Livermore) program. As far as we know, both systems are on track to be installed before the end of 2017 and go into production in early 2018. Although the host GPU of these two systems is the Power9 CPU, about 90 percent of their floating point performance is likely to be provided by the Volta coprocessors. Therefore, the performance of Summit and Sierra will be primarily derived from the capabilities of the Volta silicon.

The Summit supercomputer, in particular, will receive a great deal of scrutiny, since it is expected to deliver somewhere between 150 and 300 peak petaflops of performance when deployed at Oak Ridge National Lab toward the end of the year. That could be enough for the US to recapture the number one spot on the TOP500 list for the first time since 2012, assuming it gets installed in time for a Linpack run before November 2017. That also assumes that China doesn’t come up with a system even bigger than Summit in the interim. As we reported last week, the Tianhe-2A system is now overdue for its deployment, and is likely to be installed in 2017 with a peak capacity well north of 100 petaflops.

Much more of the Volta story should unfold in early May, during NVIDIA’s GPU Technology Conference (GTC), where the new architecture is expected to be introduced. We might even get a peak of what comes after Volta at GTC. But let’s not get ahead of ourselves.
https://www.top500.org/news/speculation-of-nvidia-volta-gpu-ramps-up-in-anticipation-of-2017-debut/
 
So Volta for consumers would be in 2018. If Vega is indeed on 16 nm this year, we could see a 7 nm refresh next year for Vega. The question is if NV will go for 7 nm on Volta in 2018. If Vega is somewhat weaker than big fat Pascal, then it would give more room to NV to just an uarch update and be on the same node and leave the jump to 7 nm for 2019.

Historically, NV has been slower to move to new nodes than AMD and has been slower to adopt new technologies(like HBM). If what I wrote above turns out to be true then 2019 would be 7 nm Navi vs 7 nm Einstein. I hope that we'll get more info on Einstein during this year's GTC.
 
Why not, 980ti was released in June of 2015, May of 2016. That is 11 months, granted 2 months less of a gtx 1180 from a 1080ti (if its launched), but still, doesn't seem they will be holding Volta back (unless there are birthing problems)
Volta has been 2018 product on NVIDIA roadmaps since it returned there, Volta-Tegra will be available in 2018, first supercomputers with Volta are to be installed by the very end of the year and "in production" in 2018 as mentioned by Top-500.org in @pharma post
 
Volta has been 2018 product on NVIDIA roadmaps since it returned there, Volta-Tegra will be available in 2018, first supercomputers with Volta are to be installed by the very end of the year and "in production" in 2018 as mentioned by Top-500.org in @pharma post


end of this year will be 2018 :) Give or take one Q is not a big deal that is in the hands of nV. Now if there are problems, then we will know, if it slides out to Q2 or more of next year.
 
end of this year will be 2018 :) Give or take one Q is not a big deal that is in the hands of nV. Now if there are problems, then we will know, if it slides out to Q2 or more of next year.
Of course - point being that I'd bet it will take about a year for Volta to appear on consumer space, even if GV100 comes out first (does it go straight to supercomputers or is there DVG-1-like option before it remains to be seen). NVIDIA official has apparently confirmed that Volta architecture / GV100 would be presented at GTC this year, but that doesn't mean it would get released there or become available for consumers.
 
The general product roadmap really should not be used as a literal interpretation for actual date of launch.
While the general product roadmap may seem to infer 2018, the specific presentations had it down as 2017, in fact I have a very early 2015 Nvidia HPC presentation that showed Pascal as 2016 and Volta as 2017 along with later presentations.
In same way in those presentations show the IBM/Nvidia Summit (mostly completed by 2017) is to be replaced in 2022 looking for ~20MW supercomputer.

If using Pascal as the 1st iteration of the strategy:
January 2016 announce Tegra Drive PX2
April 2016 announce Pascal and P100
June 2016 GP104 starts to become more available
August 2016 GP102 Titan starts to become available
Mid-late Q3 2016 DGX-1 available accessible to more businesses (we were offered this option with delivery being 1 week).
Late Q3 2016 GP102 Tesla/Quadro starts to become available
Late Q3 Drive PX2 goes into sampling with manufacturers
Early-mid Q4 2016 P100 available in certified solutions own nodes same config as DGX-1 (we were offered this option as well with delivery 2-3 days).
early-mid Q1 2017 GP100 launched as a Quadro product

These dates are pretty important because:
January 2017 announce Tegra Xavier to replace PX2
~May 2017 more indicators that V100 will be announced (GTC this year a month later than laster year that was April)
~Q3 2017 Nvidia has stated this is when Tegra Xavier is going into sampling with manufacturers - same period as the Drive PX2.
~End 2017 Summit supercomputer is meant to be mostly complete, going live is somepoint 2018 but has to take into consideration thorough testing/training users and importantly system admins/coding changes-porting/optimisation/etc.
~Q1 2018 hands-on workshops with it to start training users.
I think one of the other supercomputer projects associated with Summit is meant to be complete by Q1 2018, the 3rd associated project not sure.
This is important as Summit requires over 20,000 V100 dGPU (6xper node and 3,500 nodes but number may had increased as scope criteria may had been upped), and the two other supercomputers have another 20,000 V100 between them.
So that is a heck of a lot of GV100 accelerators just for 3 supercomputer contracts let alone additional orders, consider that the GP104 was meant to have launched according to some with ~30,000 GTX1080.

The only unknown is whether Nvidia will launch a consumer card like 2016 with the GTX1080, I tend to think they will launch very late Q4 2017 a GV104 consumer product.
It is not in their interest to have the gap too large between the 1st Tesla card on a new architecture and the consumer card otherwise it is a headache for the next architecture product cycle, possibly one reason Nvidia accelerated the launch of certain cards such as Pascal Titan.

Anyway everything seen so far ties up with the various presentations from Nvidia and IBM and one of those projects, and so far it looks like Pascal was a risk milestone/stepping stone specifically for Volta, hence why they did the unusual product launch cycle with Pascal but also paid off by providing high revenue.
But this is something I have said regarding the strategy/timeframe for some time now.
Cheers
 
Last edited:
I tend to think they will launch very late Q4 2017 a GV104 consumer product

Why would they? Vega is almost a year late, given that it should have launched together with Pascal. The 1080 Ti will neutralise Vega 10. AMD will not rush out Vega 20 in order not to undercut their Vega 10 buyers. Why would NV, in such a scenario, destroy their business by prematurely undercutting their own product for no reason?

The lack of 1080 Ti for most of 2016 and Q1 of 2017 has shown that if AMD isn't being competitive, then NV has no inclination to get ahead of themselves, which is logical.
A safer bet is that GV104 will be released in spring of 2018 and then full fat GV102 could possibly be released later if Vega 20 - possibly on 7 nm - manages to be significantly better.
 
Why would they? Vega is almost a year late, given that it should have launched together with Pascal. The 1080 Ti will neutralise Vega 10. AMD will not rush out Vega 20 in order not to undercut their Vega 10 buyers. Why would NV, in such a scenario, destroy their business by prematurely undercutting their own product for no reason?

The lack of 1080 Ti for most of 2016 and Q1 of 2017 has shown that if AMD isn't being competitive, then NV has no inclination to get ahead of themselves, which is logical.
A safer bet is that GV104 will be released in spring of 2018 and then full fat GV102 could possibly be released later if Vega 20 - possibly on 7 nm - manages to be significantly better.
Why do you think Vega should have launched together with Pascal?
 
Back
Top