OTOY Forums

That sounds about right—Kepler on the whole, is showing around a 50% increase over fermi (5xx series).
I figured there was a reason nvidia wasn't concerned with launching the Titan alongside the Tesla line. Hyper-Q is showing to be pretty important for dense simulation work. Thanks for the info!

Mikel

Hey Brasco,

Thanks for all your input here - it makes the buying decision a lot easier.

So with the new build, in essence a stock fan O/C Titan is running at about 2x the speed of the 580 (3.8 MS/s v 7.51Ms/s)and at least twice the VRAM too. That's seriously worth considering.

darkline wrote:Hey Brasco,

Thanks for all your input here - it makes the buying decision a lot easier.

So with the new build, in essence a stock fan O/C Titan is running at about 2x the speed of the 580 (3.8 MS/s v 7.51Ms/s)and at least twice the VRAM too. That's seriously worth considering.

No problem, yes the Titan hit's at least 7.51 on that OC with stock cooling, I've actaully got it higher at the moment (pushing 7.95Ms/s at 1149 core), hasn't hit a wall yet.

Yes, it has twice the VRAM and twice the performance, but it's at least twice the price! The reason it was a no brainer for me, is that I intend to add more of them later on, and the scaling is fantastic. Also, you don't often see a 100% increase in performance with a program after buying whole new system, let alone from just a GPU upgrade.

If anyone was to ask me whether to buy a Titan over another consumer card, I'd say if you're mostly simulating using OpenCL (hopefully we'll see a fixed driver soon) and gaming, go for a 7xxx, and there even in some CUDA programs where the cores aren't saturated it's not perfect, but as you're here on the Octane forum, I think you'd do just fine with one (or four!)

If you want me to run any benchmarks other than Octane, let me know.

cheers
brasc

Really impressive numbers you have there, brasco.
I was curious how is Titan doing in OpenGL, Cinebench 11.5 to be more precise.
ATI cards are usually smoking in OpenGL, compared to the nVidia gaming cards.

Many thanks for posting the benchmark results.

Okay I just installed my two Titans and this is my first conclusion:

- looks ace and a great piece of hardware.
- very silent. On full load they are as loud as my two 580 on idle (subjective feeling).
- card 1 gets 15 degree hotter from below card (used pci slot 1+2). I think I will plug in 1+3 or so. Need to check manual of mainboard. For some reasons no pci 8x if I have tow cards but only 1x (same with old cards)
- very fast


Renderspeed deceases over time. My test-scene went from 24,4 to 22,2 samples after 25min rendering. If the card gets a bit cooler samples go up. If hotter then it goes down again. Will check what I do about it.

Can it take more textures than the GTX680 ?

mbetke wrote:Okay I just installed my two Titans and this is my first conclusion:
...
Renderspeed deceases over time. My test-scene went from 24,4 to 22,2 samples after 25min rendering. If the card gets a bit cooler samples go up. If hotter then it goes down again. Will check what I do about it.

so You get them in the end =) child won =p

as far as I read, nVidia puts Titans with 'thermal target' (thrue voltage & speed) ..so cards adjust their speed depending on temperature. to get best out of the it seems WC block from EK looks good solution, or at least good spacing without just sufficient airflow is needed =)

brasco wrote:Hi guys for interest here's a couple of results from my Titan:

1.10 - 6.53 Ms/s

1.11 - 7.51 Ms/s

1.11 with both Titan and my older 580 together hits 11.31 Ms/s.

System Info:

• Windows 7 x64 Professional
  i7-3930K 4.4Ghz
  32GB RAM
  EVGA GTX Titan
  EVGA GTX 580 Classified 855Mhz Core

Pretty happy with those results, what I'm really impressed with, is the improvement in 1.11, Good job!

Definitely will add another Titan when I can afford it, it's the quietest card I've owned in years.
If anyone want's me to test out anything else let me know.

cheers
brasc

I try to compare Brasco titan with single 690

1.10 - 6.03 Ms/s

1.11 - 6.57 Ms/s

dual gtx 690:

1.10 - 12.08 Ms/s

1.11 - 13.10 Ms/s

pinochio wrote:Really impressive numbers you have there, brasco.
I was curious how is Titan doing in OpenGL, Cinebench 11.5 to be more precise.
ATI cards are usually smoking in OpenGL, compared to the nVidia gaming cards.

Many thanks for posting the benchmark results.

Hi, no problem.

Cinebench OpenGL score is bad compared to the AMD cards, as usual, it's almost exactly the same as my GTX580 ~56, most likely that's due to the Cinebench OpenGL test being massively CPU dependent (when I OC it scales linearly). The OpenGL in Cinebench 11.5 is also very old now, they use a newer OpenGL in R14. Best cards for C4D viewport are AMD are the moment, but the Nvidia cards are above what I'd say is an acceptable theshold, and Cinema4D struggles with object counts way before you hit a polycount wall (I'm a user of over 10 years).

mbetke wrote:Renderspeed deceases over time. My test-scene went from 24,4 to 22,2 samples after 25min rendering. If the card gets a bit cooler samples go up. If hotter then it goes down again. Will check what I do about it.

I guessed with several keeping the temp target high might be a solution, I've upped the temp target and tuned the fan profile to keep temps in check, I think I'll put them under water when I get the next one, or Phase Change for +1.6Ghz...?

Zay wrote:Can it take more textures than the GTX680 ?

I'm not 100% sure on this, with 6Gb buffer vs 4Gb, my guess is, yes it can. Not sure how memory is allocated in Octane.

cheers
brasc

so You get them in the end =) child won =p

Yes child vs. brain 1:0

I got EVGA PRecision tools now and will play with it. Just moving the fan to max while rendering is the brute force solution hehe.
Gave Battlefield 3 a run last evening and it plays super-smooth with only 40% GPU load on one of my 24" screens. And even didn't try SLI for the game.
Just a side-info for the gamers among us.