Titan V is Supported under RTX in new Driver

[purecain]

As the title states, a titan v user over on hard forums just found the info in the latest driver...

as a Titan V owner im over the moon... and performance is around 15% better on the TV without RTX

 

[newtekie1]

Not to burst your bubble, but I'm pretty sure any Pascal card with 8GB of memory or more is supported as well.

 

[DeathtoGnomes]

performance is around 15% better on the TV without RTX

my TV doesnt come with RTX.

 

[Caring1]

Is Ray tracing supported on Pascal using Vulkan? The changelog in the newest driver states:

Vulkan 1.1

• This driver release provides full support for the new Vulkan 1.1 API and passes the Vulkan Conformance Test Suite (CTS) version 1.1.1.2.
• Includes interoperability with CUDA 10.0.
• New extensions for Turing GPUs:
  • VK_NVX_raytracing (also available for Pascal GPUs with 8GB or more video memory, and Volta GPUs)
  • VK_NV_compute_shader_derivatives
  • VK_NV_corner_sampled_image
  • VK_NV_fragment_shader_barycentric
  • VK_NV_mesh_shader
  • VK_NV_representative_fragment_test
  • VK_NV_scissor_exclusive
  • VK_NV_shader_image_footprint
  • VK_NV_shading_rate_image

Not to burst your bubble, but I'm pretty sure any Pascal card with 8GB of memory or more is supported as well.

As quoted above.

 

[rtwjunkie]

As the title states, a titan v user over on hard forums just found the info in the latest driver...

as a Titan V owner im over the moon... and performance is around 15% better on the TV without RTX

Why so excited? Even if top level Pascal cards are enabled, trying to turn on RTX will result in a slideshow. Good luck.

 

[hat]

RTX would crush Pascal... but Titan V already has Tensor cores, so it should be comparable to the new RTX series cards.

 

[rtwjunkie]

RTX would crush Pascal... but Titan V already has Tensor cores, so it should be comparable to the new RTX series cards.

The cores, number and capability are comparable?

 

[WikiFM]

RTX would crush Pascal... but Titan V already has Tensor cores, so it should be comparable to the new RTX series cards.

It doesnt have RT cores, just the Tensor, so the BVH search will be done by CUDA cores. At least the denoising will be done by the Tensor ones, so expect less performance, but not much less like in Pascal who lacks both new cores.

 

[purecain]

ok, I think all those certain of what the gpu is capable of and not without even seeing it run the code are just guessing.
my TV is a beast and im sure with its much beefier gpu compared with turing, the 3000pound card wont have any issues whatsoever. it certainly wont be a slideshow. but now you have me speculating on performance. we dont know until its run. Anyone with a titan V should be able to run and rtx effects and weather its good or bad, all I was saying is that it works but more importantly and what seems to of gone over thhe heads of a few people here is that its included in the driver and supported to use rtx officially. so I dont think the implementation will be that bad. the 'rt core' you speak of isnt what you think. there is functionality there with volta, the card used to implement these effects. and now the new Quadro cards 3000 and 4000....

 

[londiste]

@purecain, reviewers were provided the Star Wars Elevator demo. Some of them ran benchmarks on it on Turing, Titan V and Pascal cards. I am trying to find the specific reviews and videos but Titan V did not do all that well when compared to 2080/2080Ti.

Edit:
Still can't find the review where they ran it on Titan V. Pascal cards are easier to find:
https://www.bjorn3d.com/2018/09/nvi...ce-rtx-2080-rtx-2080-ti-4k-60-fps-or-bust/32/

Edit2:
HardwareLuxx was one of them with Titan V:
https://www.hardwareluxx.de/index.p...tx-2080-founders-edition-im-test.html?start=8

 

[Vayra86]

9 FPS on a top end card.

Best buy ever

 

[Vya Domus]

RTX would crush Pascal... but Titan V already has Tensor cores, so it should be comparable to the new RTX series cards.

RTX uses RT cores not Tensor Cores. I am afraid performance is still going to by abysmal on a TItan V.

 

[RejZoR]

9 FPS on a top end card.

Best buy ever

It's usable at 1080p as showcase (probably not for gaming).

 

[Vayra86]

It's usable at 1080p as showcase (probably not for gaming).

The guys behind Star Wars Elevator demo said it actually ran OK on a single Titan V as shown at GDC. It was shown with "cinematic" quality - 24 FPS at 1080p.
12 FPS at 1440p and 9 FPS at 2160p are not bad with that in mind.

The improvement on 2080/2080Ti is down to RT cores doing they job.

'Cinematic' quality... or in fact slideshow marketed to great heights by the film industry because anything better wasn't feasible or too expensive to implement.

Fixed

 

[londiste]

The guys behind Star Wars Elevator demo said it actually ran OK on a single Titan V as shown at GDC. It was shown with "cinematic" quality - 24 FPS at 1080p.
12 FPS at 1440p and 9 FPS at 2160p are not bad with that in mind.

The improvement on 2080/2080Ti is down to RT cores doing they job.

9 FPS on a top end card.
Best buy ever

Titan V is a previous top end card.
Especially considering the old top end card doing 9 FPS 2080Ti's 34.6 FPS is not bad at all.

 

[Vya Domus]

The guys behind Star Wars Elevator demo said it actually ran OK on a single Titan V as shown at GDC. It was shown with "cinematic" quality - 24 FPS at 1080p.

That demo ran on 4 Titan Vs, so you're looking at 6 fps. Pretty good ay ?

 

[Vayra86]

The guys behind Star Wars Elevator demo said it actually ran OK on a single Titan V as shown at GDC. It was shown with "cinematic" quality - 24 FPS at 1080p.
12 FPS at 1440p and 9 FPS at 2160p are not bad with that in mind.

The improvement on 2080/2080Ti is down to RT cores doing they job.

Titan V is a previous top end card.
Especially considering the old top end card doing 9 FPS 2080Ti's 34.6 FPS is not bad at all.

Come on, wake up and get real. Not bad at all? Its horrible performance and 100% pointless performance at that, for any real time render path. This needs to die ASAP in consumer graphics.

 

[Vya Domus]

This needs to die ASAP in consumer graphics.

It doesn't need to die, it just needs proper hardware and software of which none is ready right now.

 

[Vayra86]

It doesn't need to die, it just needs proper hardware and software of which none is ready right now.

As long as it needs dedicated hardware to this degree / % of die space, no, it really doesn't deserve our investment...

 

[londiste]

Come on, wake up and get real. Not bad at all? Its horrible performance and 100% pointless performance at that, for any real time render path. This needs to die ASAP in consumer graphics.

What exactly are you trying to say? 34.6 FPS at 4K does not sound bad at all.
Especially for something that everyone keeps saying is not viable and will not take off and is pointless and too computationally expensive etc.

That 9 vs 34 illustrates exactly what Nvidia did with RT cores on Turing.
Titan V is bigger than 2080Ti and more powerful when it comes to compute. A bit lower clocks when everything else is bigger and more powerful definitely do not explain the difference.
Do you even realise how powerful Titan V is? 1080Ti does the same Star Wars demo at 4-5 FPS.

As long as it needs dedicated hardware to this degree / % of die space, no, it really doesn't deserve our investment...

What % of die space would you estimate RT cores take up?

 

[Vayra86]

What exactly are you trying to say? 34.6 FPS at 4K does not sound bad at all.
Especially for something that everyone keeps saying is not viable and will not take off and is pointless and too computationally expensive etc.

That 9 vs 34 illustrates exactly what Nvidia did with RT cores on Turing.
Titan V is bigger than 2080Ti and more powerful when it comes to compute. A bit lower clocks when everything else is bigger and more powerful definitely do not explain the difference.
Do you even realise how powerful Titan V is? 1080Ti does the same Star Wars demo at 4-5 FPS.

What % of die space would you estimate RT cores take up?

We're talking about vaguely benchmarked demo's with very little data to verify and quality settings that are still being tweaked across all the content we see. And the funny thing is, nobody really notices those tweaks because everything you get is a blurry mess anyway. 34,6 FPS tells you exactly nothing of Turing's relative performance to anything else. Regardless, that means that the 'best' you can get right now is less than half the FPS you'd really be able to get out of the card - the card that costs over 1K to get your hands on that is.

As for the die space: Its significant, and the raw performance per shader has not really increased from Pascal - and Pascal was a slight reduction from Maxwell clock for clock.

12,5% / 1/8th of the SMs seems pretty close to me and that doesn't even take into account the doubled cache size which puts it closer to 20% (!!). Effectively Turing = 3,5 years of arrested performance. Nothing really improved, the die just got a whole lot bigger and thus more expensive. All for some silly rays.

 

[londiste]

34,6 FPS tells you exactly nothing of Turing's relative performance to anything else. Regardless, that means that the 'best' you can get right now is less than half the FPS you'd really be able to get out of the card

You are making no sense at all. Reviewers are benchmarking the demo. At least on two different resolutions. On both Turing cards as well as Pascal and Volta. This is by definition showing relative performance.
What do you mean "half the FPS you'd really be able to get out of the card"? They run a benchmark and get FPS reading which is what you can get out of it, no?

 

[Vayra86]

You are making no sense at all. Reviewers are benchmarking the demo. At least on two different resolutions. On both Turing cards as well as Pascal and Volta. This is by definition showing relative performance.
What do you mean "half the FPS you'd really be able to get out of the card"? They run a benchmark and get FPS reading which is what you can get out of it, no?

It tells you nothing because RT is a separate performance metric and when it comes to non-Turing cards, they use their regular hardware to push that forward, which they lose directly for all the rest. Cards without RT cores suffer heavily from doing a workload they aren't built for, and cards with RT cores lose a significant part of the die to facilitate that performance.

If 2080ti had used that die size for raw performance entirely, it would have been a whole lot faster.

Either way its a comparison that makes no sense whatsoever and says nothing about relative performance between these cards. It only serves to support and/or showcase the claims of Turing being much faster than Pascal... at doing pointless tasks.

Its like buying a 4K TV so you can upscale your 1080p content for the next decade because nobody offers anything in 4K - and then saying you can see 4x as many pixels on screen so its 4x better.

 

[RejZoR]

'Cinematic' quality... or in fact slideshow marketed to great heights by the film industry because anything better wasn't feasible or too expensive to implement.

Fixed

Now I'm split between being a "NVIDIA hater" as charged in some other thread and showing a little enthusiasm as being a fanboy apparently...

 

[Vayra86]

Now I'm split between being a "NVIDIA hater" as charged in some other thread and showing a little enthusiasm as being a fanboy apparently...

Does your vocab consist of only fanboy and hater? Damn

I've never said you were either, btw...