Nvidia Turing Announced

[Taf the Ghost]

Just now, jasonvp said:

Have they done a Volta spin yet that uses GDDR?  As far as I recall, all Volta chips are set up for HBM2, which NVidia certainly will not ship on consumer cards.  So unless they've redone Volta to use GDDR, then I tend to doubt it.  My gut tells me Volta's days are numbered.  But I guess we'll find out in a week.

 

No. They've also released all of a single design with Volta. It would appear Volta might have been as much of a Proof of Design as it is a full product stack. 

47 minutes ago, Swatson said:

Guys remember that Nvidia is hosting a "Geforce Gaming Celebration" at Gamescom in a week. We already had rumors of a new RTX Titan, RTX 2080 and RTX 2070 before this announcement confirmed RTX anything.

 

I am going to guess those are going to be VOLTA based though, tensor cores only which does support RTX but nowhere near as fast as Turing with RT cores

Somehow, earlier today, I called that Turing had to have something beyond Tensor Cores. It appears they've "cracked" ray tracing by simply putting in dedicated hardware, and they feel they can convince businesses to buy it. I guess we'll see.

 

Still 12nm, and still some big GPUs, so the cost is appropriate. I wonder if consumer Turing will get 1 RTX & 1 Tensor Core per SM. That might make the most sense.

[jasonvp]

Just now, Swatson said:

Sure but it's not gonna be slower than Titan V

I don't know, to be completely honest.  I can't speculate to that level.  My brain is going in all sorts of interesting directions (including oozing out of my left ear) with this one.  If they release a Titan <whatever> that trounces on the Titan V and costs less, they'll utterly destroy the Titan V's market.  Done.  Gone.  They'll never sell another one.

 

But... maybe that's OK.  Why did NVidia build the Titan V?  Clearly they had their internal roadmap ready to go and knew when they'd be launching Turing.  Did they have a spin of too many Volta GPUs and needed a way to get rid of them quickly?  I know: slap 'em on a new card, call it a "Titan V", disable NVLink to differentiate it, and charge a modicum amount of money for it compared to other work cards.

 

Maybe.

[S w a t s o n]

4 minutes ago, Taf the Ghost said:

 

Still 12nm, and still some big GPUs, so the cost is appropriate.

Honestly at $2300 for RTX5000 with 60% raytracing capability of the big daddy, that's a VERY good price imo.

[Rune]

Its fun watching people squirm

[Taf the Ghost]

5 minutes ago, jasonvp said:

I don't know, to be completely honest.  I can't speculate to that level.  My brain is going in all sorts of interesting directions (including oozing out of my left ear) with this one.  If they release a Titan <whatever> that trounces on the Titan V and costs less, they'll utterly destroy the Titan V's market.  Done.  Gone.  They'll never sell another one.

 

But... maybe that's OK.  Why did NVidia build the Titan V?  Clearly they had their internal roadmap ready to go and knew when they'd be launching Turing.  Did they have a spin of too many Volta GPUs and needed a way to get rid of them quickly?  I know: slap 'em on a new card, call it a "Titan V", disable NVLink to differentiate it, and charge a modicum amount of money for it compared to other work cards.

 

Maybe.

Part of Nvidia's decision making would have been influenced by Node and memory availability. HBM2 just isn't in supply enough to really want to run a full set of GPUs off that. So even if Volta is only a single design, Turing is the clear descendant from that.

 

So, they took some of their bad dies and sold them to "production" type people for 3k. 

[Taf the Ghost]

3 minutes ago, Swatson said:

Honestly at $2300 for RTX5000 with 60% raytracing capability of the big daddy, that's a VERY good price imo.

Yeah, the RTX5000 really seems like the 2080 & 2080 TI design we'll get over the next 9 months.

[S w a t s o n]

8 minutes ago, Taf the Ghost said:

Yeah, the RTX5000 really seems like the 2080 & 2080 TI design we'll get over the next 9 months.

I am still wondering about ampere, we have rumors circulating for it still and that leaked PCB had a pretty small die area for this GPU but then again Volta was bigger so...i really dont know what to make of any of this. Nvidia really played us well this generation

 

 

Oh wait, if Ampere is purely compute, there would not be any RT cores

[vanished]

I only have one burning question: from a software perspective, how does this work?  Is it a proprietary technology like PhysX, etc. that programs will have to go out of their way to implement?  Or is it like how current games are rendered, where they send "generic" tasks and your card - AMD or nvidia - handles it fine either way?

[S w a t s o n]

1 minute ago, Ryan_Vickers said:

I only have one burning question: from a software perspective, how does this work?  Is it a proprietary technology like PhysX, etc. that programs will have to go out of their way to implement?  Or is it like how current games are rendered, where they send "generic" tasks and your card - AMD or nvidia - handles it fine either way?

RTX is proprietary to nvidia (part of gameworks) and requires RT and Tensor cores. AMD is gonna have to come up with some top-tier shit post-navi....

 

DXR (Direct X Raytracing) is agnostic but AMD doesn't have an implementation yet, Nvidia's is RTX

[vanished]

1 minute ago, Swatson said:

RTX is proprietary to nvidia (part of gameworks) and requires RT and Tensor cores. AMD is gonna have to come up with some top-tier shit post-navi....

 

DXR (Direct X Raytracing) is agnostic but AMD doesn't have an implementation yet, Nvidia's is RTX

That's what I feared... So, one day when this trickles down to consumer cards and starts getting used in games, we're going to have a divide where some games are just nvidia-only games, and some are AMD-only?  Or is it not quite that bad?

[S w a t s o n]

2 minutes ago, Ryan_Vickers said:

That's what I feared... So, one day when this trickles down to consumer cards and starts getting used in games, we're going to have a divide where some games are just nvidia-only games, and some are AMD-only?  Or is it not quite that bad?

Think of it as TressFX vs Hairworks or AMD CHS vs Nvidia PCSS, so unfortunately yes it's gonna be "nvidia games" and "amd games" but hopefully raytracing is so amazing that developers implement both for fear of getting angry yelling. Assuming AMD doesn't get rolled....

 

Metro Exodus will support RTX, so there's probably your first good look at how that plays out, except AMD has nothing to be even implemented in competition yet

[vanished]

Just now, Swatson said:

Think of it as TressFX vs Hairworks or AMD CHS vs Nvidia PCSS, so unfortuantely yes it's gonna be "nvidia games" and "amd games" but hopefully raytracing is so amazing that developers implement both for fear of getting angry yelling. Assuming AMD doesn't get rolled....

Actually wait, you said there's a generic version (DXR), and these RTX cards will run that.  So AMD just has to make something that will run it too, and it will then be obvious to devs that DXR is what they should go with... or am I not getting this right?

[S w a t s o n]

1 minute ago, Ryan_Vickers said:

Actually wait, you said there's a generic version (DXR), and these RTX cards will run that.  So AMD just has to make something that will run it too, and it will then be obvious to devs that DXR is what they should go with... or am I not getting this right?

DXR is part of DX12. RTX runs on DXR, AMD will make something that runs on DXR

[Taf the Ghost]

3 minutes ago, Ryan_Vickers said:

That's what I feared... So, one day when this trickles down to consumer cards and starts getting used in games, we're going to have a divide where some games are just nvidia-only games, and some are AMD-only?  Or is it not quite that bad?

It'll be a while before we get there, but the main thing is that it really will only matter when AMD adds it for consoles. If some Ray Tracing hardware isn't in the next console generation, then it's not going to be an issue for at least 2 more GPU generations.

[Taf the Ghost]

1 minute ago, Ryan_Vickers said:

Actually wait, you said there's a generic version (DXR), and these RTX cards will run that.  So AMD just has to make something that will run it too, and it will then be obvious to devs that DXR is what they should go with... or am I not getting this right?

 

Just now, Swatson said:

DXR is the API like DX12. RTX runs on DXR, AMD will make something that runs on DXR

Correct, but the other thing: who's launching DX12 games right now? We're going to need Ashes of the Benchmark 2 for Ray Tracing in about a year.

[S w a t s o n]

2 minutes ago, Taf the Ghost said:

It'll be a while before we get there, but the main thing is that it really will only matter when AMD adds it for consoles. If some Ray Tracing hardware isn't in the next console generation, then it's not going to be an issue for at least 2 more GPU generations.

1 minute ago, Taf the Ghost said:

 

Correct, but the other thing: who's launching DX12 games right now? We're going to need Ashes of the Benchmark 2 for Ray Tracing in about a year.

Metro Exodus tho

I REALLY like Metro so I might be a little extra hype

[Taf the Ghost]

13 minutes ago, Swatson said:

I am still wondering about ampere, we have rumors circulating for it still and that leaked PCB had a pretty small die area for this GPU but then again Volta was bigger so...i really dont know what to make of any of this. Nvidia really played us well this generation

 

 

Oh wait, if Ampere is purely compute, there would not be any RT cores

My thought is Ampere is the 7nm designs. Nvidia, having enough money, ran tandem design runs when they were planning the next generation. Unlike Intel, Nvidia had contingencies in place for how the Node went. 7nm isn't really worth it yet, so it's better to run out a 12nm generation (since AMD isn't going to run passed you) while bringing in more tech for your Data Center sales.

 

So, the reason Ampere cropped up is that they have taped out designs on 7nm and those are the current code name. Whether it's Data Center only or the post-Turing full generation, I have no clue. Is it the Volta replacement?

[Taf the Ghost]

3 minutes ago, Swatson said:

Metro Exodus tho

I REALLY like Metro so I might be a little extra hype

DX12 and Raytracing. Interesting.  If Metro doesn't sell well, at least the Engine should be a good seller.

[S w a t s o n]

5 minutes ago, Taf the Ghost said:

DX12 and Raytracing. Interesting.  If Metro doesn't sell well, at least the Engine should be a good seller.

It's Metro, it's gonna sell well or well enough at least IM TOTALLY NOT BIASED AT ALL (but seriously the previous ones have been critical successes and pushed GPU rendering boundaries, it's the newer Crysis imo)

 

 

There's no chance Nvidia doesn't bundle it with RTX Geforce Cards imo

[Taf the Ghost]

5 minutes ago, Swatson said:

It's Metro, it's gonna sell well or well enough at least IM TOTALLY NOT BIASED AT ALL (but seriously the previous ones have been critical successes and pushed GPU rendering boundaries, it's the newer Crysis imo)

 

 

There's no chance Nvidia doesn't bundle it with RTX Geforce Cards imo

Which explains some of 4A Games' marketing approach for pushing tech. Good on them. They're good games, but the main thing is they're about to be the "Game Everyone Tests with" for a while again. Much like Metro: Last Light did for a long while.

[S w a t s o n]

My condolences to anyone that bought 1080ti, Titan xP or Titan V (at least you get slow RTX on Titan V) in the last 6 months

Spoiler

 

[Amazonsucks]

2 hours ago, Swatson said:

My condolences to anyone that bought 1080ti, Titan xP or Titan V (at least you get slow RTX on Titan V) in the last 6 months

  Reveal hidden contents

 

Yeah thats why ive been saying to wait for Turing on here for so long. I basically predicted this when i first heard about Turing. Didnt know theyd go so far as to make the RT cores a thing but it makes sense. 

[Taf the Ghost]

25 minutes ago, Amazonsucks said:

Yeah thats why ive been saying to wait for Turing on here for so long. I basically predicted this when i first heard about Turing. Didnt know theyd go so far as to make the RT cores a thing but it makes sense. 

It depends what the new RT cores actually are.  They're some sort of custom unit, but what exactly they are will dictate how much actual power they add for anything but certain rendering tasks.

[ScratchCat]

5 hours ago, Swatson said:

Wait, RTX5000 is only $2300.... so if Titan RTX is that chip it's gonna be cheaper than Titan V, less than $2300 for sure. What if RTX Titan could be RTX6000, which looks to be full raytracing capability Turing?

I actually have no idea at this point but if nvidia can sell cut down Turing as RTX5000 QUADRO (expensive) for only $2300 we might see a full Turing release after all?

The GV100 was 800+mm2 with a non full core in the Titan V at $3000 so the 700mm2 RTX Titan should be less than that.

[Taf the Ghost]

8 minutes ago, ScratchCat said:

The GV100 was 800+mm2 with a non full core in the Titan V at $3000 so the 700mm2 RTX Titan should be less than that.

And GDDR6 rather than HBM, which is a good chunk cheaper as well.