We tested two ways to use a Nvidia compatible GPU card in a HP ProLiant DL 360 rackmount server for number crunching with Cuda.
The challange was to find:
- a low (not low profile) card with only 1 PCI slot high (with a cooler not higher than one slot)
- a card that fits in full-height and half-length slot
- a low power consuming card that take its power via PCI bus and do not need extra power plug
First we use a very cheap card (MSI GeForce GT 1030 with 384 cuda cores) that was working fine. We are using Linux and mathlab. This card can be used without modifications.
MSI GeForce GT 1030 (e.G.: https://www.amazon.de/dp/B07D9MQDP4/):
- 384 CUDA Cores
- 2GB 64-Bit DDR4
- PCIe 3.0 x16 (Uses x4)
- Maximum Power Consumption 30 W
With a grant from Nvidia to one of our research group we could use a NVIDIA Quadro P2000.
WARNING: You need to modifiy the hood cover and remove the the opener. Maybe you lose your warranty!
NVIDIA Quadro P2000 (e.G.: https://www.amazon.de/dp/B06XB7PFLD):
- CUDA Cores 1024
- GPU Memory 5 GB GDDR5
- System Interface PCIe 3.0 x16
- Maximum Power Consumption 75 W
Remove the opener by drilling out the opener rivets. Otherwise you could not close the top of the server. Seal the whole with duct tape or what you like if no server is above this one in the rack. The card should benefit of of the server ventilation.
Both solutions are working fine for cuda number crunching.
We did not use or test the display port!
QuickSpecs HP ProLiant DL360 Generation 8 (G8) (https://h20195.www2.hpe.com/v2/getdocument.aspx?docname=c04128242)
- PCIe 3.0 x16 expansion slot (Full-height / half-length)
- PCIe 3.0 x8 expansion slot (Low Profile)
Could work - not tested:
PNY NVIDIA RTX 2000E 16GB (https://www.pny.com/en-eu/rtx-2000e-ada-generation)
- CUDA Cores 2816
- GPU Memory 16GB GDDR6 with ECC
- System Interface PCI Express 4.0
- Maximum Power Consumption 50W
NVIDIA Quadro P2200 5GB GDDR5 PCIe 3.0 (e.G. https://www.amazon.de/dp/B07T5XCB2N)
- CUDA Cores 1280
- GPU Memory 5 GB GDDR5
- System Interface PCI Express 3.0
- Maximum Power Consumption 75 Watt
NVIDIA T1000 8GB GDDR6 (https://www.amazon.de/dp/B09SL2HV5L/)
- CUDA Cores 896
- GPU Memory 8 GB
- System Interface PCI Express 3.0
- Maximum Power Consumption 50 Watt
Inno3D GeForce GTX 1050Ti 1-Slot Ed. 4GB GDDR5 (e.G. https://www.amazon.de/dp/B07FNTWMS2)
- CUDA Cores 768
- GPU Memory 4 GB GDDR5
- System Interface PCI Express 3.0
- Maximum Power Consumption 75W
HP NVIDIA GeForce GT 730 2GB PCIe x8 GFX (e.G. https://www.amazon.de/HP-Nvidia-PCIe-Graphics-Card/dp/B06X92G36L)
- 384 CUDA Cores
- 2GB 64-Bit DDR4
- PCIe 2.0 x8
- Maximum Power Consumption 23W
(This card uses the x8 PCI-Slot and may be used parallel to a GeForce GT 1030 card - not tested)