Our group is trying to put a few CUDA-capable GPUs we have lying around the lab into action, and we would like to get any advice and/or suggestions on what we're thinking of doing as far as a setup is concerned. Specifically, we would like to make sure we aren't doing something outrageously silly... like obvious hardware mismatches or incompatibilities. So what we have to play with is the following:
GPUs:
2x GTX 470
5x GTX 480
1x Tesla C2070
CPUs:
2x Core-2 Quad Q8200
We have one workstation outfitted as follows:
Case: CoolerMaster NVIDIA CM 690 I
Motherboard: EVGA nForce 780i SLI
CPU: Core-2 Quad 8200
Memory: 4 GB DDR2 800MHz RAM
GPU1: GTX 470
GPU2: GTX 470
This works very well when not doing CPU-intensive jobs; however, when the CPU load is 100% on each (active) core for more than about 10-15 minutes, it causes the system to poweroff. We are suspecting this is an overheating issue, based on monitoring under heavy load (CPU temperatures as reported by sensors climb to dangerous levels very quickly). We are contemplating underclocking, undervolting, deactivating cores, etc., but are open to any means to increase the reliability/robustness of the system under load. Perhaps we need better cooling? If anybody has any suggestions on this unrelated issue, we would be very much obliged.
We are envisioning three additional machines outfitted as follows:
Case: CoolerMaster NVIDIA CM 690 II
Motherboard: GIGABYTE G1.Sniper Intel X58 LGA1366 Motherboard
CPU: Intel Core i7-950
Memory: 8GB DDR2 800MHz
HDD: 160GB SATA 7200RPM
PS: XION XON 1000P14F 1000W
GPU1: GTX 480
GPU2: GTX 480
Case: CoolerMaster NVIDIA CM 690 II
Motherboard: GIGABYTE G1.Sniper Intel X58 LGA1366 Motherboard
CPU: Intel Core i7-950
Memory: 6GB DDR3 1333MHz
HDD: 160GB SATA 7200RPM
PS: XION XON 1000P14F 1000W
GPU1: GTX 480
GPU2: GTX 480
Case: CoolerMaster NVIDIA CM 690 II
Motherboard: GIGABYTE G1.Sniper Intel X58 LGA1366 Motherboard
CPU: Intel Core i7-950
Memory: 6GB DDR2 1333MHz
HDD: 160GB SATA 7200RPM
PS: XION XON 1000P14F 1000W
GPU1: GTX 480
GPU2: Tesla C2070
Any feedback - really, please tell us anything that comes to mind - is appreciated. Does anybody suspect that we will face significant overheating/power/performance issues that diminish the value of this investment? We are still very new to building/configuring with GPUs, and would like for this project to be a success. Thanks for any help.
EDIT:
One very useful comment has pointed out that ventilation in the case is important. I think this is spot-on. Here's what the case offers:
Cooling System:
* Front: 140 x 25 mm Green LED fan x 1 / 1200 RPM / 19 dBA
* Top: 140 x 25 mm fan x 1 / 1200 RPM / 19 dBA (supports 120 / 140 mm fan x 2)
* Right Side Panel: 80 x 15mm x 1 (optional)
* Rear: 120mm included
* VGA Holder: 80x15mm fan x1 (optional)
* Oversized front and top mesh design for superb ventilation.
* Accommodates 120 x 240 mm radiator inside the top or bottom.
* Dust-control filters for all meshed areas.
* Air cooling support for up to 9 fans (with support for up to 3 x 140mm fans).
* External SATA X-dock.
* Front green LED fan on/off switch.
* Rear retaining holes for water cooling kit.
* Includes 1.8” & 2.5” HHD and SSD adapter.
* Cable management and CPU retaining hole for easy maintenance.
* Includes VGA card bracket (supports triple GPU card).
