Nvidia Titan

[vlado]

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

It's a bit hard to say when things get down to seconds; that's why ideally the benchmark should be run for at least some minutes.

Best regards,
Vlado

[alexyork]

30 seconds v 44 seconds is quite significant! But yes, it would be best to run a benchmark that takes at least a few minutes for a real comparison. Ideally something around 5 mins normally.

[Guest]

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.

Any chance of some stock tests? Everyone has there own overclocking methods so its difficult to benchmark against it. I hear the Titan is around 85% performance of a GTX 690.

[werticus]

Its a lot IMO!
Percentage wise its a big saving of time.

a 44 hr render on the 680 would take 30 hrs on the titan.

[alexyork]

47% faster. That's significant. A typical overnight 5 hour render might take 3.4 hours. Huge time saving! And that's with the current drivers, which I'm sure will be optimised soon, along with RT itself perhaps? Exciting results...

*Thanks for correcting my terrible maths!*

[alexyork]

Actually I've now only noticed that you were comparing with a 680 which is notoriously poor for CUDA rendering compared to a 580. So I'm wondering if the speed increase will be more like 15% or so over a 580. I don't suppose you have access to a 580 you can bench against? We'd all appreciate it!

[sandykoufax]

32% ? Isn't 47% ?

btw, here is some opencl benchmark score
http://www.expreview.com/topic/img/r...22/dpvs680.png
http://www.expreview.com/topic/img/r...2/dpvs7970.png

[alexyork]

You're right!! I told you my maths was terrible.... 47% it is. And that's HUGE. So I would anticipate probably around 30% speed increase over a 580. In that ballpark?

[sandykoufax]

So I would anticipate probably around 30% speed increase over a 580. In that ballpark?

Brilliant!!

[Bpositive]

Great work Brian.

The result seems a bit disappointing though, so it would be primarily for the double amount of memory I would do an upgrade. But hopefully the drivers will improve it's performance.


Btw. have you remembered to switch on the double precision in the 3d settings!

[tammo]

What will switching the Double precision do?
Still waiting for my card to arrive,
i hope they didn't lie about having it on stock,
wouldn't be the first Internet shop that gets massive amounts of preorders that way.

[Bpositive]

Yeah, that is the most annoying trick...!

It should enable Double precision / FP64 Mode:

"While the GK104 core did feature Double Precision support, it only included eight units per 192-core SMX, leading to FP64 operations per clock which ran just 1/24th the SP data rate. With TITAN NVIDIA has increased this to 1/3, allowing for 896 concurrent threads to be processed within a single GK110 GPU. In addition, when working in FP64 mode, TITAN will eliminate Boost but also operate at dynamically lower clock speeds."

You can read a bit more here: http://www.hardwarecanucks.com/forum...ing-act-2.html

[alexyork]

Sounds interesting!

[3LP]

Vlado, should we create a new benchmark scene to have a new ground base for next setups/releases?
It become more and more often the case that ppl have multi-gpu rigs and high end cards.

Or should we just use the one you posted and change settings? If so, it might be a good idea to post the file again with the new settings so no human mistake is done with the settings, and we will all re bench and create a new updated chart.

Thanks

Stan

[werticus]

Why would you want to enable Double precision / FP64 Mode? Vlado says that it makes no difference to vray RT..