Nvidia unveils its first Volta GPU, the Tesla V100

[midian182]

At the start of Nvidia’s annual GPU Technology Conference, one of CEO Jen-Hsun Huang's announcements during his keynote, which lasted over two hours, was the company’s first GPU based on its seventh-generation Volta architecture, the Tesla V100.

Built on a 12-nanometer FFN process, the GPU is a big leap forward from the Pascal-based Tesla P100. It features 21 billion transistors and 5120 CUDA cores with boost clock speeds of 1455MHz, along with 16GB of 4096-bit HBM2 memory (from Samsung) running at 900GB/s, all packed into an 815 square millimeter die.

Huang claims the V100, which cost $3 billion in R&D, is the most complex (and largest) chip that can be created with current semiconductor physics. It features new SM Processor architecture, which improves FP31 and FP64 performance, and is 50 percent more energy efficient than Pascal. It also comes with new Tensor cores designed for deep learning applications.

The V100 features an updated NVLink 2.0 interface that enables 300GB/s transfer speeds. The technology is now 10 times faster than standard PCIe connections, according to Huang.

The peak computation units for the V100 are as follows:

• 7.5 TFLOP/s of double precision floating-point (FP64) performance;
• 15 TFLOP/s of single precision (FP32) performance;
• 120 Tensor TFLOP/s of mixed-precision matrix-multiply-and-accumulate.

Nvidia also revealed that eight of the V100 chips would be used in an updated version of its DGX-1 supercomputer. Huang said the machine has the power to replace 400 servers, which explains its $149,000 price tag. The DGX-1V will arrive in Q3; those on a tighter budget may want to consider Nvidia’s "personal AI supercomputer" - the DGX Station. It contains four Tesla V100 GPUs and costs $69,000.

As noted by PCWorld, the V100’s reveal doesn’t necessarily mean Volta-based GeForce cards will be here in the next couple of months, but it does give an idea of what we can expect to see in the consumer GPUs.

You can find out more about the Tesla V100 here. Check out the table below to see how it stacks up against the previous five years of Tesla accelerators.

Permalink to story.

https://www.techspot.com/news/69275-nvidia-unveils-first-volta-gpu-tesla-v100.html

 

[VitalyT]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts? This is how I was sold my last car - a lot more efficient, they said,...right

And on a practical note, and something that's been tormenting me: If one is to stretch those 21bln transistors into a single thread, how long would it be?

 

[frostyshield]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts?

50% extra performance does that

 

[Kenrick]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts?

Judging in how they described volta's features, its enormous cuda cores, and new architecture, they are saying it is 50% more powerful than the pascal variant while having the same TDP of 300W.

Video cards launches are more exciting than CPU during the past decade. Now lets see who is the king of efficiency and performance in the next round of gaming cards.

 

[NigelWright]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts?

In order to do as much with a lower TPD a smaller process would need to be used than 12 nM. To have a lot lower thermal(power) footprint both a process size reduction AND architectural changes would likely be required. It's solid state physics. You can bend some rules, but not all.

 

[Squid Surprise]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts? This is how I was sold my last car - a lot more efficient, they said,...right

And on a practical note, and something that's been tormenting me: If one is to stretch those 21bln transistors into a single thread, how long would it be?

Yes, well, don't confuse efficiency with ACTUALITY.... Yes, they use the same amount of energy - the new card simply does twice as much!

Just like your old car could go 300km on a 50L tank and your new car can go 600km on the same 50L... they both still USE 50 litres of gas - your new car is just more efficient

 

[ZackL04]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts?

Judging in how they described volta's features, its enormous cuda cores, and new architecture, they are saying it is 50% more powerful than the pascal variant while having the same TDP of 300W.

Video cards launches are more exciting than CPU during the past decade. Now lets see who is the king of efficiency and performance in the next round of gaming cards.

When was the last time AMD held that crown?? I dont see them dethroning Nvidia any time soon either

 

[ForgottenLegion]

Wow.

 

[amghwk]

AMD's Radeon is f**ked.

I'm anticipating most of the critics and whiners here will be current Pascal owners, or AMD fans.

Me? I'm just a humble Maxwell owner.

 

[Jibberish18]

Well supposedly AMD may counter with this: http://www.anandtech.com/show/10905/amd-announces-radeon-instinct-deep-learning-2017/2

But we all know it won't be either as powerful or as efficient. AMD always has to give up something. It's the sad truth but that's okay. That's their niche I guess.

 

[Rockstarrrr]

Ahem, so it is 50% more power efficient than Pascal, and yet it consumes the exact same 300 Watts?

Judging in how they described volta's features, its enormous cuda cores, and new architecture, they are saying it is 50% more powerful than the pascal variant while having the same TDP of 300W.

Video cards launches are more exciting than CPU during the past decade. Now lets see who is the king of efficiency and performance in the next round of gaming cards.

When was the last time AMD held that crown?? I dont see them dethroning Nvidia any time soon either

They should make something competitive first, let alone "dethroning" nVidia. And judging by their portfolio in the last couple of years, they will never be competitive in the high-end again. Fury-X was their last "hurrah" and it failed.

And I`m not happy about it at all, nVidias prices are just simply unacceptable to me.

 

[ZackL04]

They should make something competitive first, let alone "dethroning" nVidia. And judging by their portfolio in the last couple of years, they will never be competitive in the high-end again. Fury-X was their last "hurrah" and it failed.

And I`m not happy about it at all, nVidias prices are just simply unacceptable to me.

I agree but I dont think nvidias prices are outrageous. What do you expect for outrageous pixel pushing gaming power? A 1080 can be found for around $400 on sale if you look hard enough. Thats crazy when u consider its faster than a 980ti and that card was $600 for a long time.

 

[VitalyT]

50% extra performance does that

Volta = Viagra

 

[mosu]

From your table: Tesla GP100 has 25.08 million transistors /mm2 and Volta GV100 has 25.88 millions per square mm. so 12 nm vs 16 nm is just a gimmick. Also we should take performance numbers presented with a grain of salt and just wait for the real( graphic processor) thing, when will see it on sale and not on a screen.So let's better wait for real world comparison, when Vega will be last year's graphic card.

 

[Rockstarrrr]

I agree but I dont think nvidias prices are outrageous. What do you expect for outrageous pixel pushing gaming power? A 1080 can be found for around $400 on sale if you look hard enough. Thats crazy when u consider its faster than a 980ti and that card was $600 for a long time.

Maybe in USA, not in Europe. And 980-Ti was two years ago and it was the fastest (hence, you pay premium for it). You could say its much faster than 680 and it costed 499... In EU 1060 is 300 euros (325 dollars), 1070 420e (510$), 1080 530e (575) on avg. and so on. Way overpriced, 1080-Ti is almost 1000 dollars. USA prices are much better, but not everyone lives there.

 

[Evernessince]

So AMD's Vega is going more more triangles per clock and Nvidia is going more compute. It sounds like each competitor is moving to improve what the other does better.

The chip size is massive, it will be interesting to see how much they cut it down for the consumer version.

 

[Stingy McDuck]

Surprise, that Jen-Hsun Huang's image on top was actually rendered in real time in one of these GPUs.

 

[Kenrick]

Surprise, that Jen-Hsun Huang's image on top was actually rendered in real time in one of these GPUs.

haha. Can't you see his hair is the result of NVIDIA HairWorks.

 

[cliffordcooley]

Me? I'm just a humble Maxwell owner.

Me? I'm just a humble Kepler owner.

 

[Zeromus]

My 1080 Ti already outdated af

 

[RBDx3]

My 1080 Ti already outdated af

Nah, I'm in the same boat with you in having grabbed a GTX 1080 Ti when they launched.

If we assume their timeline will be identical for the consumer variants of the Volta platform as it was with Pascal, we've got about another year of good use out of the GTX 1080 Ti. We'll most likely see, what, the GTX 1180 Ti (guessing?) around this time in 2018.

My tech-industry-two-cents-sided-brain assumption, however, is that with AI and Deep Learning being pushed FAR more now than it ever has been, they wanted to at least get the initial Volta-based platforms out specifically for those reasons alone. There's even more revenue and market share to be captured now in that area of the industry than there was this time last year, and it's only going to continue to surge.

And I mean, come on....it's not like anyone at NVIDIA is even remotely beginning to think about losing one bit of sleep over anything AMD has on their road map...

 

[RBDx3]

Me? I'm just a humble Maxwell owner.

Me? I'm just a humble Kepler owner.

And you're even then STILL light years ahead of the RX580

 

[Theinsanegamer]

Me? I'm just a humble Maxwell owner.

Me? I'm just a humble Kepler owner.

And you're even then STILL light years ahead of the RX580

funny as the joke is, that isnt technically true. If he has a 780ti, its current performance is the equivalent of a 980, which is 580 territory. Anything lower, and the 580 is both faster and better perf/watt compared to kepler.

 

[ddferrari]

Apparently the CEO has a punctuality issue- it's like he's always getting to these events at the last second on his motorcycle and didn't have time to take off his leather jacket.