TR Forums

Hey All -

What do GPU reviewers mean by a cards' "compute performance?" They always differentiate between a card's gaming performance vs its compute performance, as if different parts of the architecture are involved in one or the other. But that doesn't seem likely. Is it all about the cards' OpenGL or OpenCL capability? Or is it strictly about the software you're running? Games vs everything else?

Ultimately I'm trying to find out whether the apps I run will benefit from the Titan vs the 780 (Adobe CS6, Cinema 4D, Painter12). The reviews say the 780 is a gaming card, period, because NV crippled the double-precision floating-point capability - "crippled" compared to the Titan that is.

I've gotten by for almost 3 years with a 470 so I'm sure any new high-end card will be a nice upgrade. But I'd like to buy one that is designed for "compute" instead of "games." If only I could figure out what the difference is!

Thanks you guys. Hopefully this isn't a "can of worms" question and you can shed a little light for me. Thanks again!

Its either OpenCL or CUDA performance. OpenCL is something everyone can run, CUDA is Nvidia-only. Its how well the gpu can be used for general purpose computing (the first GP in GPGPU).

Think of it as everything a video card can do that is not graphic displaying-related.

If only they made some sort of GPGPU ability back in the 90's with Windows 95. Heck yeah, gotta crunch some numbers on my Diamond Stealth 3D 2000 with the S3 Virge chipset.

canoli wrote:

Ultimately I'm trying to find out whether the apps I run will benefit from the Titan vs the 780 (Adobe CS6, Cinema 4D, Painter12). The reviews say the 780 is a gaming card, period, because NV crippled the double-precision floating-point capability - "crippled" compared to the Titan that is.

That should be really of no concern for the applications you are running. Those apps utilize single precision which is not "crippled" on Nvidia consumer cards.

Deanjo wrote:

canoli wrote:

Ultimately I'm trying to find out whether the apps I run will benefit from the Titan vs the 780 (Adobe CS6, Cinema 4D, Painter12). The reviews say the 780 is a gaming card, period, because NV crippled the double-precision floating-point capability - "crippled" compared to the Titan that is.

That should be really of no concern for the applications you are running. Those apps utilize single precision which is not "crippled" on Nvidia consumer cards.

okay thanks. Then I would only ask - how do you find out which apps do and which apps don't take advantage of DP FP?

canoli wrote:

Deanjo wrote:

canoli wrote:

Ultimately I'm trying to find out whether the apps I run will benefit from the Titan vs the 780 (Adobe CS6, Cinema 4D, Painter12). The reviews say the 780 is a gaming card, period, because NV crippled the double-precision floating-point capability - "crippled" compared to the Titan that is.

That should be really of no concern for the applications you are running. Those apps utilize single precision which is not "crippled" on Nvidia consumer cards.

okay thanks. Then I would only ask - how do you find out which apps do and which apps don't take advantage of DP FP?

DP is pretty much limited to big iron applications like scientific and financial research.

For those that do real work and their work and time depend on a graphics cards with intensive, double-precision floating math, and whatnot, Nvidia, and AMD have their branded line of professional video cards.

Basically, most customers are enterprises that rely on these professional grade cards to do a lot of heavy lifting. Something that a consumer would rarely justify the expense for such card.

Just for another log on the Fire, but Intel ALSO makes cards. And depending on your specific needs, one of the 3 brands is faster. We are working on a sound "filter" application which uses GPGPU and currently the Intel cards are actually faster.

okay thanks you guys. I guess I've just been over-thinking this whole "compute performance" vs. gaming. So DP is strictly for industrial users, fluid syms and medical imaging etc. Well that's good to hear I guess, saves me $350! Now I can buy a 780 and feel good about it instead of wondering if I made a mistake.

keltor wrote:

Just for another log on the Fire, but Intel ALSO makes cards. And depending on your specific needs, one of the 3 brands is faster. We are working on a sound "filter" application which uses GPGPU and currently the Intel cards are actually faster.

I'm not doubting your findings however I have found after looking at the code for many claims that "x is faster then z" usually comes down to running of unoptimized code on one of the devices.

keltor wrote:

Just for another log on the Fire, but Intel ALSO makes cards. And depending on your specific needs, one of the 3 brands is faster. We are working on a sound "filter" application which uses GPGPU and currently the Intel cards are actually faster.

Just don't call Intel's card a GPU; they got creamed in that market so bad that they stopped trying :-p. I'm glad that they didn't let that get to them, though. The Xeon Phi is still a very interesting product that may very well mature into something Intel uses across their line.