Nvidia Titan

[alexyork]

Looks like they should be shipping around 28th so end of Feb early March. And at £699+VAT...

VERY tempted but I think I'll wait for Vlado or someone else to post some benchmarks against a 580 so we can really see what they're like. A request, Vlado - if you do benchmark, please also do try to max out the VRAM by using a massive scene with lots of proxies etc. as a separate test. Something really challenging that the 3GB 580 can't actually handle due to RAM constraints.

[Macker]

VERY tempted here too. The price point of decent spec Quadro cards always puts off the powers that be when it comes to spending money on something, but I feel the titan might just be enough to swing it in our favour. Would love to see it used in an actual production environment first though, with a proper scene, proxies, high res textures, HDRI etc.

Can you use SLI with these beasts?

[alexyork]

Yes you can SLI them as far as I can see. Really, imagine the power! I think 6GB should cover most complex scenes that have been properly optimised. You might need to strip some stuff out into passes but I reckon most high-end shots should fit if you're careful. The most I've pushed a real production scene to was around 10GB for one shot but we could have got away with smaller textures and reduced that considerably I'm sure.

I've never tested Vray RT at 5000 pixels. How does it handle high-res renders for print? It's not really much use to me otherwise. Might try some tests on it today with my 2GB 580.

[vlado]

I've never tested Vray RT at 5000 pixels. How does it handle high-res renders for print? It's not really much use to me otherwise. Might try some tests on it today with my 2GB 580.

It will handle it fine; we don't store any part of the final image on the GPU. Of course, you get more pixels to calculate, so naturally it's slower to get the same number of samples per pixel.

Best regards,
Vlado

[alexyork]

It will handle it fine; we don't store any part of the final image on the GPU. Of course, you get more pixels to calculate, so naturally it's slower to get the same number of samples per pixel.

Best regards,
Vlado

Sounds good to me. Time to do some testing...

[Guest]

I was made aware my times were off on the GTX 690 as I had them noted in seconds. I meant to note them as minutes and seconds. Anyway they have now been corrected

[Jiri.Matys]

I have GTX 580 and its faster in GPGPU (VRay RT) then GTX680. So I really wanna know about GTXTitan. Its a little bit expensive, but it can be a real killer, if it use all 2688 CUDA cores the same way as GTX 580 with 512 CUDA cores. Any news about it? I can see some shops are selling this beast. If someone have it, let us know the benchmarks please.

[dtolios]

I have GTX 580 and its faster in GPGPU (VRay RT) then GTX680. So I really wanna know about GTXTitan. Its a little bit expensive, but it can be a real killer, if it use all 2688 CUDA cores the same way as GTX 580 with 512 CUDA cores. Any news about it? I can see some shops are selling this beast. If someone have it, let us know the benchmarks please.

The Titan is still Kepler, which means its performance cannot be comparable to the 580's Fermi architecture core per core. We have to compare apples to apples, and CUDA core architecture has changed - probably forever. It will most likely be much faster than a 580, but nowhere near that the CUDA core count would suggest (i.e. more than 5 times faster).

[tammo]

I just ordered one that was listed as "in stock".
If that's not an error, I can try out the card tomorrow or the day after.
If you have some benchmarks that i can run for you, please send me some info or links how to do that.
I will post as soon as its running, hoping that the reseller didn't make an error about availability.

Here is an interesting quote from the latest Anandtech review, are they talking about apps like V-Rray RT?:

"Meanwhile in the world of GPGPU computing Titan stands alone.
Unfortunately we’re not able to run a complete cross-platform comparison due to Titan’s outstanding OpenCL issue,
but from what we have been able to run Titan is not only flat-out powerful,
but NVIDIA has seemingly delivered on their compute efficiency goals,
giving us a Kepler family part capable of getting far closer to its theoretical efficiency than GTX 680,
and closer than any other GPU before it.
We’ll of course be taking a further look at Titan in comparison to other GPUs once the OpenCL situation is resolved
in order to come to a better understanding of its relative strengths and weaknesses,
but for the first wave of Titan buyers I’m not sure that’s going to matter.

If you’re doing GPU computing, are invested in CUDA, and need a fast compute card,
then Titan is the compute card CUDA developers and researchers have been dreaming of."

[Alan Iglesias]

Sounds very interesting Tammo, thanks very much for getting in there and checking this out, assuming of course you get the card.

If you could at least run the standard benchmark (http://www.spot3d.com/vray/scenes/vr...h_interior.rar), I'm sure we'd all love to see it (make sure you render it using CUDA as I believe the file is still set to openCL as a default). I'm also dying to know how big of a file you can render with 6GB VRAM!

I recently rendered a 347MB/3.6M poly scene in 1.6GB of GPU VRAM, so I'm quite hopeful the 6GB will be useful for many folks out there.

Thanks again, and looking forward to your reports,

-Alan

[nikand]

Nice Tammo...I'm realy looking forward to your benchmarks. Fingers crossed, that Titan delivers!

[alexyork]

Looking forward to your results! If you could please run a comparison with your older card so we can compare directly, then maybe try to run a scene with as many proxies etc. as you can fit in the VRAM, that would be very useful! It would also be very good to hear about noise and heat levels, too.

[BKE]

I have compared an EVGA GTX 680 4gb Classified, overclocked with a slightly overclocked EVGA GTX Titan Superclocked. I used the test file referred to in this thread and ran the latest RT. Here are the results, at 1920x1152 sized image:
680 OpenCL = 1m 48.3s
680 CUDA = 0m 44.1s

Titan OpenCL = 1m 42.7s
Titan CUDA = 0m 29.9s

I am not an expert VRay user, by any means. I have appreciated all the help I have received in this forum, so I wanted to share my interesting results!
Also - the 680 GTX was tweaked for max performance, the Titan is not.

Thanks,
Brian

[vlado]

That's odd, the benchmark normally renders in a bit less than three minutes on a 680 at its original resolution. Did you change anyhing else besides the resolution?

Best regards,
Vlado

[sandykoufax]

44s vs 30s ? That's not so impressive. :-/