Nvidia unveils the $9,000 Quadro GV100 GPU with RTX technology

[Cal Jeffrey]

Last week Nvidia showed off its real-time ray tracing (RTX) technology at GDC. Microsoft unveiled a new API for DirectX 12 that takes advantage of RTX and Epic Games showed us what it could do with the tech. Nvidia’s RTX is a game changer for 3D graphics. The only problem is the technology is not available on current generation GPUs.

That changed today at the GPU Technology Conference (GTC) when Nvidia announced the Quadro GV100. The GV100 is the company’s next-gen GPU utilizing the Volta architecture with RTX built right in.

The new workhorse features 32GB HBM2 memory, 10,240 CUDA cores, and 640 tensor cores. This gives it 7.4 TFLOPS of power for double-precision rendering or 14.8 TFLOPS for single-precision.

The card also supports NVLink 2 allowing two GV100s to be linked with a shared memory pool, essentially doubling your processing power with 64GB of memory.

Of course, it also comes with support for Nvidia OptiX, DirectX Raytracing, and Vulkan. While we have seen some great tech demos implementing RTX technology in the past week, don’t get your hopes up for a GV100 gaming rig. The new card isn’t meant for consumers.

CEO Jensen Huang made it clear that the card is built with commercial workstations in mind. For now, studios like Disney and ILM will be utilizing the card and RTX technology to shorten rendering time. The Quadro GV100 will retail for $8,999 direct from Nvidia when it becomes available in April.

Permalink to story.

https://www.techspot.com/news/73902-nvidia-unveils-quadro-gv100-rtx-technology.html

 

[Uncle Al]

WOOF!!!! I'll add that one to my Christmas Shopping List .............. for 2050

 

[wiyosaya]

Yes, nVidia cares about gamers!

 

[yRaz]

Yes, nVidia cares about gamers!

there are likely to be under 1000 of these cards made, however there is a demand for them and they aren't free to produce. These cards actually aren't profitable for nVidia to make but they act more of a trophy card for the brand than a functioning product. There is a very real need for these cards and thus people buy them, but they can't make them profitably.

Think of it this way, Ford doesn't make money by selling the new Ford GT at $300,000+. Ford makes money because the Ford GT is a trophy car that attracts attention to their brand so people buy products in the lower end of their brand.

 

[quadibloc]

Yes, nVidia cares about gamers!

Hey, gamers sometimes go to the theater to watch movies too!

 

[VitalyT]

there are likely to be under 1000 of these cards made...

For real? Many super-computers today feature thousands of nVidia Quadro cards. And that's just one niche.

 

[wiyosaya]

Yes, nVidia cares about gamers!

there are likely to be under 1000 of these cards made, however there is a demand for them and they aren't free to produce. These cards actually aren't profitable for nVidia to make but they act more of a trophy card for the brand than a functioning product. There is a very real need for these cards and thus people buy them, but they can't make them profitably.

Think of it this way, Ford doesn't make money by selling the new Ford GT at $300,000+. Ford makes money because the Ford GT is a trophy car that attracts attention to their brand so people buy products in the lower end of their brand.

I highly doubt your assessment since pretty much all of their chips, as has been proven in the past by various exploits that turn consumer cards into quadros, are the same silicon.

As a software engineer, I have worked for several companies where added features were always included in the base software, however, the customer was charged substantially extra for those features. To have n number of different versions of software is unsustainable and costly. I have no reason to believe it is any different with silicon.

Besides, my comment was meant in a sarcastic tone. So far, we have not heard anything about new gamer cards, nor have we heard about mining focused cards. nVidia appears to be in it for the money, and nothing else. They will do anything to push prices up as they have been doing for years. They are a public company and about making profit and nothing else. To me, all this means that nVidia's comments in regards to caring about gamers, are specious! When they come out with mining cards and new gaming cards, then I might change my mind about their intentions - however - even then, I doubt that I will.

 

[yRaz]

For real? Many super-computers today feature thousands of nVidia Quadro cards. And that's just one niche.

I stand corrected

https://en.wikipedia.org/wiki/Titan_(supercomputer

I don't think all new supercomputers being built would use that many but even if one of that size is built then production numbers would far exceed my expectations.

 

[davislane1]

there are likely to be under 1000 of these cards made...

For real? Many super-computers today feature thousands of nVidia Quadro cards. And that's just one niche.

You mean to tell me that Quadro is used by more people/companies than LTT?

The more you know...

 

[Soulburn74]

I guess this joke is finally old.........
But I still can't believe I'm the first person to say........


....... yeah, but can it play crysis?

 

[Brock Kane]

I guess this joke is finally old.........
But I still can't believe I'm the first person to say........


....... yeah, but can it play crysis?

That joke has run its course!

 

[Theinsanegamer]

The card also supports NVLink 2 allowing two GV100s to be linked with a shared memory pool, essentially doubling your processing power with 64GB of memory.

This is much more interesting to me. If NVLINK 2 is fast enough to sync HBM across two cards, it could potentially be fast enough to allow MCM for standard geforce cards, the thing AMD has been working on. Combining both cards into one would eliminate a lot of driver work to enable multiGPU.

Perhaps NVLINK will replace SLI for multi GPU, making it relevant again?

 

[Evernessince]

The card also supports NVLink 2 allowing two GV100s to be linked with a shared memory pool, essentially doubling your processing power with 64GB of memory.

This is much more interesting to me. If NVLINK 2 is fast enough to sync HBM across two cards, it could potentially be fast enough to allow MCM for standard geforce cards, the thing AMD has been working on. Combining both cards into one would eliminate a lot of driver work to enable multiGPU.

Perhaps NVLINK will replace SLI for multi GPU, making it relevant again?

Nvidia would NEVER do that unless it was forced to by AMD. Can you imagine Nvidia offering a technology that would give you Titan performance if you simply bought two 1060's? That's $800 Nvidia losses out on.

FYI shared memory pool is one thing but it's still far away from AMD's Infinity fabric. There are much more to GPUs than simply that.

 

[Theinsanegamer]

The card also supports NVLink 2 allowing two GV100s to be linked with a shared memory pool, essentially doubling your processing power with 64GB of memory.

This is much more interesting to me. If NVLINK 2 is fast enough to sync HBM across two cards, it could potentially be fast enough to allow MCM for standard geforce cards, the thing AMD has been working on. Combining both cards into one would eliminate a lot of driver work to enable multiGPU.

Perhaps NVLINK will replace SLI for multi GPU, making it relevant again?

Nvidia would NEVER do that unless it was forced to by AMD. Can you imagine Nvidia offering a technology that would give you Titan performance if you simply bought two 1060's? That's $800 Nvidia losses out on.

FYI shared memory pool is one thing but it's still far away from AMD's Infinity fabric. There are much more to GPUs than simply that.

Well why would nvidia put it on a 1060? Much like SLI, I imagine it would be a feature exclusive to nvidia's high end xx8x and above cards. They have to find a way to push that 4k144 with HDR somehow.

Sharing the memory pool is an important first step. It indicated that Nvlink 2 has bandwidth capabilities far outstripping SLI or Nvlink 1. Connecting memory pools so each GPU knows what they are working on was a huge challenge due to the bandwidth required.

 

[Shadowboxer]

The card also supports NVLink 2 allowing two GV100s to be linked with a shared memory pool, essentially doubling your processing power with 64GB of memory.

This is much more interesting to me. If NVLINK 2 is fast enough to sync HBM across two cards, it could potentially be fast enough to allow MCM for standard geforce cards, the thing AMD has been working on. Combining both cards into one would eliminate a lot of driver work to enable multiGPU.

Perhaps NVLINK will replace SLI for multi GPU, making it relevant again?

Nvidia would NEVER do that unless it was forced to by AMD. Can you imagine Nvidia offering a technology that would give you Titan performance if you simply bought two 1060's? That's $800 Nvidia losses out on.

FYI shared memory pool is one thing but it's still far away from AMD's Infinity fabric. There are much more to GPUs than simply that.

But Infinity fabric can’t do what Nvidias shared memory pool can!?

 

[Evernessince]

Well why would nvidia put it on a 1060? Much like SLI, I imagine it would be a feature exclusive to nvidia's high end xx8x and above cards. They have to find a way to push that 4k144 with HDR somehow.

Sharing the memory pool is an important first step. It indicated that Nvlink 2 has bandwidth capabilities far outstripping SLI or Nvlink 1. Connecting memory pools so each GPU knows what they are working on was a huge challenge due to the bandwidth required.

The way MCM works, the architecture has to be designed around it. You can't just decided to use it for higher end cards and forget the lower end ones. That would require you to make two different architectures for the same product line and that would completely erase cost savings.

But Infinity fabric can’t do what Nvidias shared memory pool can!?

Yeah, actually Infinity fabric does much more than that. It can not only transfer a massive amount of data in a short period of time but it can do it with very little latency. A shared memory pool is great but it's still a far cry for the data requirements of sharing all the GPU resources.

The big problem with Nvidia's technology is the GPUs still show up as separate. Infinity fabric doesn't have this problem and it means you don't have to worry about limited SLI support and cross GPU latencies.