Mac Pro 5,1 3x GPU's running slow

[Grafton]

Hello
So, I've been using maya forever and have just made the leap to redshift and GPU rendering. Purchased a Gigabyte Geforce GTX 1080ti gaming for my 2010 12 core mac pro 5,1 , 64 GB ram with High Sierra. Pleased with the render results I went ahead and purchased a Geforce GTX 1070ti. Everything's peachy. Found a great deal on another 1080ti (an msi aero), added an external PSU (EVGA 850GQ) and.... disaster. Well, not really, the performance is just terrible - screen lag, maya (and everything else) becomes unusably slow (not just rendering but interaction as well).
I've done some experimenting with different cards in different slots and the results are all over the place: geforce 1070ti in slot 1 gigabyte 1080ti in slot 2 = fine,
msi aero 1080ti in slot 1 with gigabyte 1080ti in slot 2 = fine,
gigabyte 1080ti with geforce 1070ti in slot 4 = lag and poor performance,
which made me think perhaps the problem is slot 4, however I was able to recreate the lag with two msi aero 1080 ti's in slots 1 and 2!

Thoughts anyone??

Yes, I have the latest Nvidia web driver (I think) 387.10.10.40.124

 

[bsbeamer]

Few things...

Only use Slot #1 and Slot #2 for GPU, they are two x16 slots on MP5,1. The other slots (#3 & #4) are x4 slots and not suitable for GPU.

Regardless of what reports may say, NVIDIA Web Drivers on Mac only work with multiple GPUs if the GPUs are IDENTICAL. Some people have worked around this, but results vary greatly. ESPECIALLY when troubleshooting, it's important to adhere to the matching GPU configuration.

The only MP5,1 that I've personally used with more than two GPUs was with an expansion chassis via x16 slot. Many of those have switching mechanisms built in to "manage" if they do not directly assign slot speeds within. Have not seen those configurations frequently reported on this forum.

Stick with 2 matching GPUs (at least matching series) or look into expansion chassis.

 

[Grafton]

Thanks for the input. Weirdly, when I have the two matching msi aero cards in slots 1 & 2, I'm still getting terrible performance but with the gigabyte and an aero it's fine.

 

[bsbeamer]

Just because cards are branded the same does NOT mean they are identical. Look for part numbers (P/N) and SKU's.

Also physically look at the GPU near the PCIe connection. If there is an NVIDIA logo on the two that are working together, that's a good sign. If there is not, it means the GPUs have been modified from reference in some fashion. Others will "work" but you're talking about such a special-use case scenario with that it's best to stick with reference designs.

If you know how to dump and modify ROMs, you could go that approach to force the ROMs to be identical, but that may be going down a rabbit hole for little to no real benefit.

(Note: The NVIDIA logo presence is an easier identifier on 10XX series GPUs, not always present on previous series/generations.)

 

[deppest]

Thanks for the input. Weirdly, when I have the two matching msi aero cards in slots 1 & 2, I'm still getting terrible performance but with the gigabyte and an aero it's fine.

[doublepost=1557322149][/doublepost]Sorry about the nil teply!

I disagree with previous poster regarding the general advice to use identical Gpus in multi gpu setups only and also re non-suitability of slots 3&4 for gpu purposes.

I am currently using dual gtx titan x maxwell on sierra in slots 1&3 with no issues, mainly for rendering. Previously, i had a 980ti (evga i think) along with the titan x, also in slots 1&3 without issues either. the x4 speed in slot 3 only shows in synthetic testing but is completely irrelevant for gpu rendering purposes.

 

[Grafton]

Just because cards are branded the same does NOT mean they are identical. Look for part numbers (P/N) and SKU's.

Also physically look at the GPU near the PCIe connection. If there is an NVIDIA logo on the two that are working together, that's a good sign. If there is not, it means the GPUs have been modified from reference in some fashion. Others will "work" but you're talking about such a special-use case scenario with that it's best to stick with reference designs.

If you know how to dump and modify ROMs, you could go that approach to force the ROMs to be identical, but that may be going down a rabbit hole for little to no real benefit.

(Note: The NVIDIA logo presence is an easier identifier on 10XX series GPUs, not always present on previous series/generations.)

Makes sense. I guess this isn't quite as straight-forward as I'd hoped (mixing & matching). I'll definitely look into getting some proper NVIDIA reference cards!

 

[bsbeamer]

I disagree with previous poster regarding the general advice to use identical Gpus in multi gpu setups only and also re non-suitability of slots 3&4 for gpu purposes.

As noted...

Regardless of what reports may say, NVIDIA Web Drivers on Mac only work with multiple GPUs if the GPUs are IDENTICAL. Some people have worked around this, but results vary greatly. ESPECIALLY when troubleshooting, it's important to adhere to the matching GPU configuration.

OP is troubleshooting an issue. Building out a render machine or eeking out performance beyond spec is not troubleshooting. Getting "stuff" work correctly first. OP never stated the use, purpose, or need for multiple GPUs. Also not assuming it's not JUST for rendering on this machine. Some people use multiple GPUs to drive multiple video displays and using them in PCIe x4 slots on MP5,1 is asking for trouble...

 

[Grafton]

[doublepost=1557322149][/doublepost]
Sorry about the nil teply!

I disagree with previous poster regarding the general advice to use identical Gpus in multi gpu setups only and also re non-suitability of slots 3&4 for gpu purposes.

I am currently using dual gtx titan x maxwell on sierra in slots 1&3 with no issues, mainly for rendering. Previously, i had a 980ti (evga i think) along with the titan x, also in slots 1&3 without issues either. the x4 speed in slot 3 only shows in synthetic testing but is completely irrelevant for gpu rendering purposes.

Good to know!
Perhaps card compatibility becomes a bigger issue when trying to run 3 in unison? I have also read that the x4 speed on slots 3 & 4 is not an issue for GPU rendering (specifically).

 

[h9826790]

Hello
So, I've been using maya forever and have just made the leap to redshift and GPU rendering. Purchased a Gigabyte Geforce GTX 1080ti gaming for my 2010 12 core mac pro 5,1 , 64 GB ram with High Sierra. Pleased with the render results I went ahead and purchased a Geforce GTX 1070ti. Everything's peachy. Found a great deal on another 1080ti (an msi aero), added an external PSU (EVGA 850GQ) and.... disaster. Well, not really, the performance is just terrible - screen lag, maya (and everything else) becomes unusably slow (not just rendering but interaction as well).
I've done some experimenting with different cards in different slots and the results are all over the place: geforce 1070ti in slot 1 gigabyte 1080ti in slot 2 = fine,
msi aero 1080ti in slot 1 with gigabyte 1080ti in slot 2 = fine,
gigabyte 1080ti with geforce 1070ti in slot 4 = lag and poor performance,
which made me think perhaps the problem is slot 4, however I was able to recreate the lag with two msi aero 1080 ti's in slots 1 and 2!

Thoughts anyone??

Yes, I have the latest Nvidia web driver (I think) 387.10.10.40.124

Avoid the V387 driver for multi GPU setup.

Disable SIP, then use the following command to force install the V378 driver. And see if it helps.

bash <(curl -s https://raw.githubusercontent.com/Benjamin-Dobell/nvidia-update/master/nvidia-update.sh) 378.10.10.10.25.106

 

[Grafton]

Avoid the V387 driver for multi GPU setup.

Disable SIP, then use the following command to force install the V378 driver. And see if it helps.

bash <(curl -s https://raw.githubusercontent.com/Benjamin-Dobell/nvidia-update/master/nvidia-update.sh) 378.10.10.10.25.106

Thanks. Unfortunately, redshift (the render software) states that it requires 387.128 or later . I'm not sure what the consequences would be if I downgraded?

 

[h9826790]

I'm not sure what the consequences would be if I downgraded?

Should be nothing more than “if doesn’t work, install the V387 driver again”

 

[Grafton]

I'll give it a shot

 

[Le_Cactus]

Hello
So, I've been using maya forever and have just made the leap to redshift and GPU rendering. Purchased a Gigabyte Geforce GTX 1080ti gaming for my 2010 12 core mac pro 5,1 , 64 GB ram with High Sierra. Pleased with the render results I went ahead and purchased a Geforce GTX 1070ti. Everything's peachy. Found a great deal on another 1080ti (an msi aero), added an external PSU (EVGA 850GQ) and.... disaster. Well, not really, the performance is just terrible - screen lag, maya (and everything else) becomes unusably slow (not just rendering but interaction as well).
I've done some experimenting with different cards in different slots and the results are all over the place: geforce 1070ti in slot 1 gigabyte 1080ti in slot 2 = fine,
msi aero 1080ti in slot 1 with gigabyte 1080ti in slot 2 = fine,
gigabyte 1080ti with geforce 1070ti in slot 4 = lag and poor performance,
which made me think perhaps the problem is slot 4, however I was able to recreate the lag with two msi aero 1080 ti's in slots 1 and 2!

Thoughts anyone??

Yes, I have the latest Nvidia web driver (I think) 387.10.10.40.124

Hi, try this webdriver (see link) with SIP disabled.
I've tested with two GTX 1070, it's the last version who works with multi-GPU from Nvidia.

https://forums.macrumors.com/thread...graphics-cards.1440150/page-189#post-26620848

 

[Grafton]

Avoid the V387 driver for multi GPU setup.

Disable SIP, then use the following command to force install the V378 driver. And see if it helps.

bash <(curl -s https://raw.githubusercontent.com/Benjamin-Dobell/nvidia-update/master/nvidia-update.sh) 378.10.10.10.25.106

Success!!! 3 GPU's running happily and maya/redshift seem to be ok. Thanks very much!!