Just a quick update regarding my rendering tests.
Downloaded a trial version of bunkspeed (now SolidWorks Visualize) and tried to run a couple of tests. What I liked about this trial software is it gave you an option to select either CPU, GPU or Hybrid (both) rendering. Unfortunately it did not have any option to select which GPU to use, (if you have multiple GPU's in your system). Initially tried to load my previous SolidWorks model but was only presented with a blank screen though I believe it may be due to bunkspeed being a 2015 version and the SolidWorks model being created in SolidWorks 2016. Went online and downloaded a couple of relatively complex earlier version SolidWorks models and tried these. Through some trial and error and educated guess work I was able to load a model and render a scene - great! Pressed the render button and after a couple of minutes my screen went completely blank/black and the computer was unresponsive (but the power was still on and the fans had spun up).
(Similar issue to when I had my previous Z600 and Quadro 4000 setup? Long story, but in brief it was down to the vanes on the Quadro 4000 card being blocked by PC fluff. Removed the fluff and the workstation was a lot quieter and no more black screen/shut downs. Obviously this issue was down to over heating, or more correctly, the lack of cooling!).
I was also wondering whether it was the GPU cards overheating or down to pulling to much power from the Z620 800W psu. Did a quick calculation;
Quadro K4200 = 108W
Tesla M2090 = 225W
2x E5-2670 @115W ea = 250W
Total = 583W
I'm effectively leaving only 217W to power everything else, e.g. motherboard, SATA III card, ASUS M.2 SM951, 64GB memory, all the DVD/SSD/HDD's and the enclosure fans. Did some more digging and can confirm the Z620 does support the maximus set-up (up to 225W power) so should be okay? One thing to note is that the Tesla M2090 card does have a 6-pin and a 8-pin PCIe power connectors but since I was already using one of the 6-pin power connectors on the Z620 for the Quadro K4200, I only had one spare 6-pin PCIe power connector. I purchased a 6-pin PCIe to 8-pin PCIe adapter cable from Maplin and used that to power the card. (NOTE: The Tesla data/spec sheet says it requires either 2x 6-pin connections or a single 8-pin connection for power so should be okay. Also, the HP site does state that adapters can be used provided the PSU can handle the power demand on the 12v power rails). NOTE: If any HP techs are reading this and know what the power rails are rated at then please let me know. I searched everywhere for a detailed spec of the HP Z620 800W psu without any luck? I note on other threads it is stated as 50W per rail and since the PCIe 6-pin socket has 3x 12v rails should be able to handle 150W (the PCIe lane provides and additional 75W giving 225W in total).
Posted by rgaik elsewhere on this forum;
Still wasn't 100% certain at this point if the black screen was software or hardware related so started looking for some way of monitoring the GPU temperatures in real time.
Found another really good GPU rendering benchmark program called Octanebench which is now my favourite GPU benchmark tool. It's a standalone program (can even run directly from a USB stick) but the main difference is it allows you to select which GPU to use for benchmarking and it is specifically optimized for CUDA processing. Tried to run a test last night rendering with both GPU's and the screen went blank again. Tried the Quadro K4200 by itself and it ran no problem. Running with only the Tesla M2090 went to the blank screen.
Thought I would give it another try this morning (with effectively a cold/cooler PC) and was able to run all the tests successfully;
Quadro K4200 score = 39.92
Tesla M2090 score = 46.84
Quadro K4200 + Tesla M2090 = 86.59
https://render.otoy.com/octanebench/results.php
Looking at the benchmark tables, this score is comparable to a single quadro K6000 benchmark which I'm very happy with. I obviously need to investigate the over heating or PSU issue but clearly the Tesla is more that doubling my rendering performance (dependant on the rendering program utilizing CUDA). Using the nvidia control panel I was able to monitor the GPU utilization in real time;
Planning on pulling the Tesla M2090 from the workstation today and give it a good inspection and clean. I will consider the problem resolved once I can get bunkspeed to run in Hybrid mode without the PC crashing.