Dual CUDA Card Dilemma

pfunkyfize

Hi all, wondering if anyone has run into this hardware/(software ?) problem:

I recently bought a GTX 1060 6GB to go along with my GTX 980 TI 6GB to speed up Iray renders. The 980 alone zooms through renders when i was just doing CPU renders prior to getting any of these GPUs.

I originally placed the 1060 alongside the 980 inside the box but the heat was killing the 980. Before I took it out I did notice that the PC and DAZ recognized the second card but renders were only getting sent to the 980 but not to the new card. I then saw that GPU-Z monitoring tool was reporting that the new card was inactive because it was 'Limited by total power limit' . So then I solved the major problem first - heat dissipation. I bought a separate PSU, a PCIe riser and expansion cable and reinstalled the new card outside of the PC for better heat dynamics and a 400w power supply dedicated to that new card alone. Well i beat the heat problem and the PC and DAZ recognizes the card but I still get the same problem that the 1060 is not even being given the render to assist the 980. Even though GPU-Z says there's a power limit problem i am scratching my head on that because the card has its own power supply. The cards are not connected by an SLI bridge and I don't intend to play video games with 2 different cards by SLI, this was meant to be a rendering rig. CUDA totals of the two cards is 4096, More than the Titan X of 3,072 at a fraction of the price $500 for two cards compared to $1200 Titan X. I just need to figure out why there is no render flow to the new card. Anyone run into this at all? Here is the screenshot while I was rendering a simple scene. The GPU-Z is running in the bottom right and you can see on the left that DAZ acknowledges the existence of my new card. I haven't done anything beyond what you see here in configuring my card for Studio...

Is it possible to have bought a card that turns on and is recognized by the PC but the acceleration properties are broken?

Thanks!

fixmypcmike

The version of Iray that supports the 10 series cards is still in beta -- Daz Studio incorporating that version is in Private Beta testing now, so we don't know what bugs if any it has or when it might be ready for Public Beta.

pfunkyfize

Fixmypcmike said:

The version of Iray that supports the 10 series cards is still in beta -- Daz Studio incorporating that version is in Private Beta testing now, so we don't know what bugs if any it has or when it might be ready for Public Beta.

Ah ok, so I could return this card and grab a GTX card with 6GB  of Vram that is not the 10xx series and it should spin fine. Though I am having trouble finding a 6GB CUDA card other than another 980 TI on the market. Any suggestions? I was going to go with a 970 with 4GB but it seems like my scenes are bumping right up on that limit...

neo_xenophanes

Oh this is good to know 10 series cards aren't fully supported yet. I'm in the middle of trying to seclect new cards myself. This is some useful info

pfunkyfize

neo_xenophanes said:

Oh this is good to know 10 series cards aren't fully supported yet. I'm in the middle of trying to seclect new cards myself. This is some useful info

At least you didn't just jump in and drop some dosh on the 10xx like I did LOL. I am thinking a 970 4GB will do fine for me as a side card to the 980.

pfunkyfize

Thanks guys,

Consider this solved, I took out another mortgage and bought another 980 TI. 5632 CUDA is looking better than 4096 I suppose...

nicstt

pfunkyfize said:

Thanks guys,

Consider this solved, I took out another mortgage and bought another 980 TI. 5632 CUDA is looking better than 4096 I suppose...

Not if you're comparing different generations; are you comparing a 1080 and 980ti? If so, it isn't just about number of cores, but how good they are; the 1080 is expected to be better, and Nvidia claim it is, but until real-world comparrisons are made, we don't know.

Comparing Cuda in different generations is not a reliable method. The reason is there are other factors that influence a card's performance.

pfunkyfize

nicstt said:

Not if you're comparing different generations; are you comparing a 1080 and 980ti? If so, it isn't just about number of cores, but how good they are; the 1080 is expected to be better, and Nvidia claim it is, but until real-world comparrisons are made, we don't know.

Comparing Cuda in different generations is not a reliable method. The reason is there are other factors that influence a card's performance.

Nope not comparing a 1080, i got the 'lowly' 1060 but is still a generation 10 card that won't work with DAZ - yet. Since the CUDA is the unit of measurement that is driving my performance calculations that's all I have to go on. Coming from a 0 CUDA hardware setup to something that kicks at 2816 CUDA I only have the CUDA to measure by, not experiences of what GPU X compared to GPU Y measured in at overclocking/non-clocked/voltage etc. factors. All I know is that 2816 CUDA does an Iray render that is 5 GB in size, 2400 pixel width in half an hour on my PC( with a 100% Convergence limit). Either way, I sent it back and getting a second 980TI coming tomorrow. It will be interesting to see what 5632 CUDA that are the same make and model (and supported by DAZ no less!) does for the speed...

fastbike1

Cuda isn't the only parameter that matters. GPU clockspeed is also a contributor and will have a more significant impact when using two non-identical cards. Iray will throttle the faster card(s) to match the slowest one. Could become significant since all of the 10XX cards have faster clocks than the 9XX series.

nicstt

fastbike1 said:

Cuda isn't the only parameter that matters. GPU clockspeed is also a contributor and will have a more significant impact when using two non-identical cards. Iray will throttle the faster card(s) to match the slowest one. Could become significant since all of the 10XX cards have faster clocks than the 9XX series.

It will have some effect for sure, but how much we will see.

The biggest factor will be the generation the CUDA cores and other aspects come from. Not all CUDA cores are created equal. (Unless from the same generation, then (as far as I am aware) they always are.)

fastbike1

Well my 980ti doesn't have all that many more CUDA cores than the 780 that I had (2816 to 2304), but the clock speed difference is much greater (1320 to 900) and is pretty relfective of the difference in rendering times.

pfunkyfize

nicstt said:

 (Unless from the same generation, then (as far as I am aware) they always are.)

I just confirmed this, I got my two 980TI cards up and running and that 5GB 30 minute render was done in 14 minutes. I am more than pleased...

nicstt

fastbike1 said:

Well my 980ti doesn't have all that many more CUDA cores than the 780 that I had (2816 to 2304), but the clock speed difference is much greater (1320 to 900) and is pretty relfective of the difference in rendering times.

Have you tried reducing the clock speed on the 980ti to the same as the other then comparing? Would be interesting. It would tell you how much was due to clock speed, and how much to other factors. Of course, other factors may not add much (per se), but do allow for an increase in clock speed (which makes it tough to determine if it's purely a clock speed increase, or the architectural changes Nvidia introduced with the 900 series). They managed to do some pretty nifty stuff that allowed a generation increase whilst keeping the same fabrication die-size.)

kentphoto

• AMD Radeon RX 480 (4GB)  does anyone know if this card supports Iray. And would this be a good alternative to
• Nvidia's 980TI till the Nvidia 1070 card is Iray supported.
• Thanks!

fixmypcmike

AMD cards do not support Iray.

fastbike1

kentphoto "AMD Radeon RX 480 (4GB)  does anyone know if this card supports Iray. "

Generally frowned upon to spam multiple threads with the same question.