Nvidia shows off Post Volta GPUs in Next Generation Drive PX

[AlTech]

Nvidia has shown off their new generation Drive PX which is coming out late next year and journalists have spotted a new generation of GPU.

 

The GPU in the new Drive PX platform for Deep AI learning and Self Driving cars is newer than Volta.

Quote

At GTC Europe 2017 in Germany, NVIDIA today has announced the Drive PX Pegasus, a new entry to the Drive PX family of computing modules for self-driving cars. Building off of the previous Drive PX family members, the Driver PX Pegasus is intended to be the next step in self-driving hardware by being the company's first Level 5 system, meaning it's capable of supporting fully automated driving for fully autonomous vehicles. Put succinctly, this is the holy grail of what NVIDIA has been building towards over the last few years.

The Drive PX Pegasus is very much a forward-looking product. While NVIDIA is announcing it today, they won't even have dev kits available until later next year, and any kind of commercial release is farther off still. Consequently, the specifications for Drive PX Pegasus are equally forward looking: the board features two  unannounced post-Volta next-generation discrete GPUs, which will be doing most of the heavy lifting. To put this in context, NVIDIA has only just started shipping Big Volta (GV100) for compute products, and smaller scale Volta GPUs are not expected until 2018, so we're looking at something quite far into the future. Meanwhile, rounding out the package and serving as the hearts of the Pegasus will be a pair of NVIDIA's upcoming Xavier SoCs, which combine an integrated Volta GPU (complete with tensor cores) with an unnamed octa-core ARM CPU design.

With the Drive PX Pegasus, NVIDIA is targeting commercial applications in robo-taxis and driverless long-haul. It will, in turn, coexist with the to-be-launched Drive PX Xavier, NVIDIA's previously-announced small-scale self-driving hardware that essentially packs Drive PX 2's capabilities into a small 30W board and features a single Xavier SoC. Drive PX Pegasus, by contrast, is essentially a next-generation Drive PX 2, utilizing much more powerful SoCs and GPUs than before.

NVIDIA has stated that Drive PX Pegasus will be air-cooled, although it's likely to be right at the edge of what's reasonable. WIth a 500W TDP, NVIDIA is pushing the envelope on performance in part by packing so much hardware into a single board. Not that a car will have any kind of trouble delivering that kind of power, of course, but it's quite the interesting change in car design when we're talking about significant amounts of power being dedicated to a non-mechanical operation.

 

 

 

I'm really not impressed by Nvidia's arrogance a while ago when Jensen said Pascal was unbeatable this holiday season.

 

If Nvidia has post-volta GPUs ready and coming soon to Corporate then the least they can do is offer GeForce Volta to consumers. Nevermind the fact that GPC (as popularized by "The Good Old Gamer" on YouTube) hasn't increased since Maxwell for gaming.

 

This is ludicrous behavior and despite how I feel about Deep AI learning and self driving cards, this is truly disgusting and turns me off of buying new Nvidia GPUs even more.

 

Source:

https://www.anandtech.com/show/11913/nvidia-announces-drive-px-pegasus-at-gtc-europe-2017-feat-nextgen-gpus

[Max_Settings]

This isn't a surprise. They already have at least the next 3 generations of GPUs now if they wanted to. They just space them out to make money, same as any other manufacturer.

[AlTech]

Just now, Max_Settings said:

This isn't a surprise. They already have at least the next 3 generations of GPUs now if they wanted to. They just space them out to make money, same as any other manufacturer.

I'm not entirely sure about that but in this case they definitely have 1 newer than Volta which isn't surprising since Volta is fairly similar to Pascal.

[JuztBe]

Sadly thats what you get with lack of competition. Intel was sitting on their ass for ~5 years, now nvidia will  do the same.

I just wish that the passion CEOs once had in their fields wouldn't get sucked out so fast after seeing their wallets icreasings exponetially. Imagine EA or Activision ran by people who are passionate about gaming.

[AlTech]

Just now, JuztBe said:

Sadly thats what you get with lack of competition. Intel was sitting on their ass for ~5 years, now nvidia will  do the same.

I just wish that the passion CEOs once had in their fields wouldn't get sucked out so fast after seeing their wallets icreasings exponetially. Imagine EA or Activision ran by people who are passionate about gaming.

Hopefully Dr Lisa Su get's RTG together and releases good products in 2018 which compete.

[mynameisjuan]

14 minutes ago, JuztBe said:

Sadly thats what you get with lack of competition. Intel was sitting on their ass for ~5 years, now nvidia will  do the same.

I just wish that the passion CEOs once had in their fields wouldn't get sucked out so fast after seeing their wallets icreasings exponetially. Imagine EA or Activision ran by people who are passionate about gaming.

Well you cant blame them. New dies require a new process and equipment to do it which doesnt come cheap. So they could either spend money on new equipment again or use the current equipment for longer for profit until they need it. 

 

I get that people get pissed over this but Intel and Nvidia are still businesses. I guess people just forgot the goal of a business is to make money.

[ravenshrike]

19 minutes ago, AluminiumTech said:

Hopefully Dr Lisa Su get's RTG together and releases good products in 2018 which compete.

Um... this is pretty damned clearly a response to AMD's venture with Tesla.

[GoodBytes]

45 minutes ago, Max_Settings said:

This isn't a surprise. They already have at least the next 3 generations of GPUs now if they wanted to. They just space them out to make money, same as any other manufacturer.

Not really. Volta is probably ready for productions. But "Volta 2.0" as I call it (as Nvidia has tendency of doing: new architecture > revision > new architecture > revision) with some exceptions where you have more than 1 revision, and you can have a revision that got a lot of changes due to delays of the next architectures (example: Pascal). As we speak, you probably have another team working on Nvidia next architecture (2-3 GPUs from now). It takes on average 5 years to make a new GPU architecture. So it must suck for the engineer working on the successor of Volta architecture, knowing that what they are working on should result in a much better graphics card than Volta, and Volta isn't even out yet.

 

[System Error Message]

well, how much is the px pegasus gonna cost? Planes can fly themselves btw and they dont use nvidia.

[LordLeewee]

49 minutes ago, Max_Settings said:

This isn't a surprise. They already have at least the next 3 generations of GPUs now if they wanted to. They just space them out to make money, same as any other manufacturer.

yes, this is exactly how it happens.

[michaelocarroll007]

50 minutes ago, JuztBe said:

Sadly thats what you get with lack of competition. Intel was sitting on their ass for ~5 years, now nvidia will  do the same.

I just wish that the passion CEOs once had in their fields wouldn't get sucked out so fast after seeing their wallets icreasings exponetially. Imagine EA or Activision ran by people who are passionate about gaming.

seems like Intel is responding quickly to AMD which either means that they really were sitting and holding back or the huge more amount of money helped Make it be able to respond quicker. 

[GoodBytes]

14 minutes ago, System Error Message said:

well, how much is the px pegasus gonna cost? Planes can fly themselves btw and they dont use nvidia.

LOL not really no.

[DildorTheDecent]

No point in releasing Volta for the gamer crowd at this time. 

 

Nvidia would only be competing with themselves. 

 

Vega would look like a toy. 

[Mira Yurizaki]

1 hour ago, System Error Message said:

well, how much is the px pegasus gonna cost? Planes can fly themselves btw and they dont use nvidia.

That's because:

• Planes operate in free space where there's little possibility of hitting something else. ACAS is really only a concern near airports where there's more airplanes flying around.
• The equations of aerospace are well understood and designs are probably optimized to minimize a computer needing to do something.
  • I mean, we've had realistic enough flight simulators running on personal computers since the 90s.
• People who operate airplanes need years of training before they're allowed to call the shots. Possibly even a couple before they're allowed to ever touch the real thing.

i.e., totally different animal than driving a car where most people are just barely qualified to operate the thing.

 

2 hours ago, JuztBe said:

Sadly thats what you get with lack of competition. Intel was sitting on their ass for ~5 years, now nvidia will  do the same.

Intel's progression on IPC since Sandy Bridge hasn't really changed much since whenever (except from Prescott/Presler to Core). And this is true on AMD as well (except from Piledriver to Zen).

 

On the flipside, this may make developers who were relying on Moore's Law to fix their design issues start designing their applications well enough in the first place.

[System Error Message]

1 minute ago, M.Yurizaki said:

That's because:

• Planes operate in free space where there's little possibility of hitting something else. ACAS is really only a concern near airports where there's more airplanes flying around.
• The equations of aerospace are well understood and designs are probably optimized to minimize a computer needing to do something.
• People who operate airplanes need years of training before they're allowed to call the shots. Possibly even a couple before they're allowed to ever touch the real thing.

i.e., totally different animal than driving a car where most people are just barely qualified to operate the thing.

 

planes can fly themselves even with traffic, and can land themselves too, even take off themselves.

You have various different modes for autopilots in commercial planes, different for private aircraft though.

Planes also have a mode that allows for remote control, this is usually done from another aircraft that can physically see what is going on and can fly near.

[GoodBytes]

1 hour ago, System Error Message said:

planes can fly themselves even with traffic, and can land themselves too, even take off themselves.

You have various different modes for autopilots in commercial planes, different for private aircraft though.

Planes also have a mode that allows for remote control, this is usually done from another aircraft that can physically see what is going on and can fly near.

Maybe in some research lab. But the FAA certification is so extensive... it is for a reason why plane sports 3DFX chips for its graphical interface, and not a GeForce GTX 1080.

 

Flying in the air is simple. Landing and take off if where you need a pilot, and it is most dangerous part of flying. Highly trained pilot is needed, with quick thinking to manage all sorts of changing situation as the plane takes off and lands.

 

If AI in planes was a legit thing, trust me, Air line company would ditch pilots day 1.

[Mira Yurizaki]

17 minutes ago, System Error Message said:

planes can fly themselves even with traffic, and can land themselves too, even take off themselves.

You have various different modes for autopilots in commercial planes, different for private aircraft though.

Planes also have a mode that allows for remote control, this is usually done from another aircraft that can physically see what is going on and can fly near.

You also have to understand that airplanes fly in controlled spaces where there are ground support crew monitoring what's going on. Combined with the fact that free space is very large and even in busy air space you're very unlikely to hit something (if you're within 30 miles to a busy airport, there are probably dozens planes in the air, but you don't see them), this makes it much easier for a computer to fly an airplane.

 

Also take off is stupid easy (seriously, you may as well try and tell me a computerized drag racer is impressive). Landing takes more finesse but it's not that hard either. And again, you're in a controlled environment. Once you have clearance to land, you basically own whatever section of airspace is necessary to get to the runway and on touch down, you basically own the runway until you taxi off it. Nobody else is allowed to take your space.

 

Driving is in a totally different environment that's barely controlled. Everyone and anything is allowed to take your space.

Edited October 10, 2017 by M.Yurizaki

[Sniperfox47]

3 minutes ago, System Error Message said:

planes can fly themselves even with traffic, and can land themselves too, even take off themselves.

You have various different modes for autopilots in commercial planes, different for private aircraft though.

Planes also have a mode that allows for remote control, this is usually done from another aircraft that can physically see what is going on and can fly near.

Planes also recieved signals from ground control and don't rely on computer vision though... They don't need to read traffic signs and detect when little Suzy's dog jumps out in front of them. There are electronic communications in place that *drastically* reduce the complexity of control for a plane.

 

2 hours ago, AluminiumTech said:

If Nvidia has post-volta GPUs ready and coming soon to Corporate then the least they can do is offer GeForce Volta to consumers. Nevermind the fact that GPC (as popularized by "The Good Old Gamer" on YouTube) hasn't increased since Maxwell for gaming.

 

This is ludicrous behavior and despite how I feel about Deep AI learning and self driving cards, this is truly disgusting and turns me off of buying new Nvidia GPUs even more.

Why do you even want them to release Volta to consumers? If you take the v100 they're using in the Teslas you'd be looking at a $1500-2000 graphics card minimum, and if they cut it down and removed the new enterprise level features from it to make it cheaper you'd have Pascal, for all intents and purposes. Volta has no practical benefit on the consumer side, and the dies are *insanely* expensive.

 

What are the Volta features?

• 8-bit/16-bit optimized SM architecture (which gets cut out on consumer systems)
• Higher bandwidth nvlink (which doesn't exist on consumer systems)
• HBM2 (which would be cut out and replaced with GDDR5 for consumer systems)
• Multi-Process Service for QoS assurances while running multiple CUDA applications (datacenter feature that doesn't apply to consumer users)
• Unified memory and address translation with IBM power machines using nvlink (not compatible with x86 machines or PCIe)
• And Cooperative Group APIs for CUDA9 (backported to Maxwell and Pascal, with only some patterns exclusive to Volta)

 

I'd much rather have a consumer Pascal chip with all the reduced precision support left enabled than have a Volta chip with all of that stuff still cut out.

[AlTech]

57 minutes ago, M.Yurizaki said:

ACAS is really only a concern near airports where there's more airplanes flying around.

*TCAS

 

Traffic

Collision

Avoidance

System.

Quote

On the flipside, this may make developers who were relying on Moore's Law to fix their design issues start designing their applications well enough in the first place.

But Moore's Law isn't a real law. It's just an observation one man made over 10 years ago.

[AlTech]

33 minutes ago, Sniperfox47 said:

Planes also recieved signals from ground control and don't rely on computer vision though... They don't need to read traffic signs and detect when little Suzy's dog jumps out in front of them. There are electronic communications in place that *drastically* reduce the complexity of control for a plane.

 

Why do you even want them to release Volta to consumers? If you take the v100 they're using in the Teslas you'd be looking at a $1500-2000 graphics card minimum, and if they cut it down and removed the new enterprise level features from it to make it cheaper you'd have Pascal, for all intents and purposes. Volta has no practical benefit on the consumer side, and the dies are *insanely* expensive.

 

What are the Volta features?

• 8-bit/16-bit optimized SM architecture (which gets cut out on consumer systems)

*FP16 & FP8.

 

VEGA has great FP16 and FP8 perf so there's no reason why GeForce Volta shouldn't have it.

33 minutes ago, Sniperfox47 said:

• Higher bandwidth nvlink (which doesn't exist on consumer systems)
• HBM2 (which would be cut out and replaced with GDDR5 for consumer systems)

Again, VEGA has this and Nvidia could add it to their high end consumer segment.

33 minutes ago, Sniperfox47 said:

• Multi-Process Service for QoS assurances while running multiple CUDA applications (datacenter feature that doesn't apply to consumer users)
• Unified memory and address translation with IBM power machines using nvlink (not compatible with x86 machines or PCIe)

Unified Memory is part of Consumer stuff but not completely.

 

The bit you mentioned isn't part of the consumer stuff.

 

33 minutes ago, Sniperfox47 said:

I'd much rather have a consumer Pascal chip with all the reduced precision support left enabled than have a Volta chip with all of that stuff still cut out.

Volta is also using a different spec for the process.

 

It's using slightly modified spec of 16nm to be able to claim it is "12nm" (which we know it's not true 12nm).

[Taf the Ghost]

37 minutes ago, GoodBytes said:

Maybe in some research lab. But the FAA certification is so extensive... it is for a reason why plane sports 3DFX chips for its graphical interface, and not a GeForce GTX 1080.

Flying in the air is simple. Landing and take off if where you need a pilot, and it is most dangerous part of flying. Highly trained pilot is needed, with quick thinking to manage all sorts of changing situation as the plane takes off and lands.

 

If AI in planes was a legit thing, trust me, Air line company would ditch pilots day 1.

FAA cert takes between 3-5 years, from what I've been told. At least on the Flight Control and other systems. Though modern airliners can take-off & land themselves at this point. 

 

As to Volta 2.0, design & small runs aren't hard to do with Silicon. But Nvidia is the market leader. They're selling millions of the GPUs. That takes some massive Fab time. They're currently selling over 1 Billion USD per quarter in consumer GPUs. Somewhere in the range of 1-2 million per month. They have to accrue return on the initial investment (since they pay for custom aspects of their Nodes), then build up money to cover future R&D costs. Nvidia is now making a lot of money, but they have to spend a lot to do what they're doing.

 

Volta isn't going to be a huge speed improvement over Pascal. It's on a refined 16nm process (thus "12nm"), but it doesn't look like there's huge uArch changes. That probably comes next time. It's going to be a double-refined Maxwell, it seems, but the 1080 replacement (2080? 1180?) should be 25-30% faster, by having more cuda cores and clocking higher, albeit what should be roughly the same size of die.

[Sniperfox47]

32 minutes ago, AluminiumTech said:

*FP16 & FP8.

 

VEGA has great FP16 and FP8 perf so there's no reason why GeForce Volta shouldn't have it.

Again, VEGA has this and Nvidia could add it to their high end consumer segment.

Unified Memory is part of Consumer stuff but not completely.

 

The bit you mentioned isn't part of the consumer stuff.

 

Volta is also using a different spec for the process.

 

It's using slightly modified spec of 16nm to be able to claim it is "12nm" (which we know it's not true 12nm).

Except GP100 had great FP16 and FP8 performance too, and it was cut out for consumers to avoid cannibalizing their enterprise/deep learning Tesla markets. There's also the fact that the 16 bit and 8 bit reduced precision hardware takes up a ton of space on the die and contributes to the ridiculous costs for GP100 and GV100. And probably contributes to the high die cost of VEGA too. There's no reason for them to enable it for the consumer space.

 

GP100 had HBM2 for enterprise stuff too, and it was cut out for consumers because of cost and availability issues. Neither of these issues have disappeared so both of those things would stay cut out in a consumer release. Look at the insane price for VEGA. What would nVidia gain by switching to HBM2 in the consumer market, really? And VEGA *does not* have nvlink, since that's a proprietary interface made by nVidia that's only available in GP100, GV100, and IBM Power CPUs.

 

If VEGA is the architecture you're taking cues from, you're doing it wrong. Is VEGA good for prosumers? Sure. Is VEGA good for deep learning or datacenters? Not in the slightest. The heat generation of it is insane, the power draw is insane, and all that to get not even the same performance as a Tesla P100, much less a Tesla V100. Unless RTG has some ace hidden up their sleeve for the enterprise market, they're in trouble.

 

Unified memory on the consumer side is just memory mirroring, not memory pooling. It's not a full memory sharing implementation like Volta has when combined with Power, just a way to ease design challenges for CUDA developers. But again, it doesn't matter because the new system from Volta wouldn't even be applicable, since consumer chips aren't used with nvlink or Power CPUs.

 

And sure, but you could node shrink Pascal too. And if you took GV100, cut out all the new enterprise features, and cut out all the stuff from GV100 that was cut out from GP100, then you'd basically wind up with a node shrunk GP102...

 

Volta has 0 benefits for consumers, outside maybe dumb prosumers who are trying to run 10-20 rendering workloads on a single GPU all at once where the QoS system would help. If that didn't get cut out.

[Misanthrope]

Legit question: do we actually have cars with these? Nvidia has been going on and on about cars for some time now yet I usually know nothing about car tech and if it's actually being used.

[Sniperfox47]

1 minute ago, Misanthrope said:

Legit question: do we actually have cars with these? Nvidia has been going on and on about cars for some time now yet I usually know nothing about car tech and if it's actually being used.

https://www.nvidia.com/en-us/self-driving-cars/partners/tesla/

Tesla does, among others.

[Guest]

3 minutes ago, Sniperfox47 said:

https://www.nvidia.com/en-us/self-driving-cars/partners/tesla/

Tesla does, among others.

Tbh, Tesla was the only one that I had even heard used one since it isn't marketed much. Nice to hear that it is found in more brands (Too bad its mostly luxury vehicles with the exception of a couple.) Wonder if this will make it a more unified platform for car jacking