I am trying to add a Tesla P100 to my x99 workstation board. In the Tesla documents it says to remove other GPU then boot with the Tesla card. However my CPU has no on board graphics, so with only a Tesla plugged in I have no display output. I also watched TeamViewer to see if the computer would come online, it didn’t even though the motherboard signaled it finished the boot sequence.
I tried booting with both a Titan X and the Tesla P100, nothing happened.
I also tried installing Tesla drivers prior to plugging the card in, they would not install without the card.
I can’t find any webpages or documents explaining how to install the card on a motherboard with no on board graphics. Does anyone know how?
Thanks
The Tesla P100 is not really designed to be used in a workstation, or on a workstation motherboard. You may be able to find one that is compatible, but it would be by chance. There are no test reports issued for the product that list compatible workstation motherboards.
If you did get it to boot up, you’d next have to address the cooling issue. The card is not designed to keep itself cool like a typical workstation GPU. Instead, it requires that a lot of air be forced through the heatsink. For reasons like this and others, you generally won’t find guides that explain how to use a Tesla P100 in a workstation.
Instead its recommended that you purchase the Tesla P100 in a properly configured, certified OEM server.
Thanks for the reply txbob.
I have a rear mounted high speed ducted fan installed on the case like many OEM have used for Tesla and Xeon Phi in the past. So there is airflow, once/if I can get the card working I can see if that is a sufficient cooling solution when under load. Then decide how I will use the cards.
I will be getting a GPU server soon that I can use the card in if putting the cards in a workstation does not work.
So are you saying there will be no support for trying to use the PCI-E Tesla cards in a workstation?
Previous Tesla generations were used in workstations often, has the use cases shifted?
Side note: We had purchased the P100 prior to the more suited Quadro P6000 being available, this was due to a research budget cut off date which limited us on what card we could get.
Yes, that is what I am saying, for this particular Tesla P100 PCIE, at this time.
Regarding other tesla models, the only ones that were advertised for use in workstations were the “C” or “c” models, e.g. K40c, C2070/2075, K20c. There are no C/c models for the Tesla P100, currently, and your Tesla P100 does not have integrated fan.
These C/c models had an integrated fan, much like the “workstation” GPUs e.g. Quadro. Therefore they could keep themselves cool in a workstation environment. There were other differences as well, such as PCI BAR size differences (e.g. a K40c may have had a different PCI BAR size than a K40m), which could impact their ability to boot up in a particular system or not. This is largely a function of the specific motherboard BIOS regarding compatibility. An OEM certified system obviously has this kind of compatibility built into the BIOS. A random workstation motherboard BIOS may or may not be compatible for boot purposes.
YMMV.
Okay thank you for the information.
These retired enterprise GPUs could be powering innovation instead of sitting in recycling bins. While they may no longer meet enterprise performance needs, they’re more than capable of driving projects for entrepreneurs, makers, tinkerers, and non-profits operating on tight budgets.
The community has already proven we can handle the technical challenges - we’ve got cooling solutions and the expertise. All we need is for manufacturers to remove the artificial barriers preventing these GPUs from having a second life in workstations.
By unlocking these capabilities, we could put powerful computing in the hands of innovators who will use it to solve real problems. The hardware is capable. The community is ready. Dell, ASUS, and other manufacturers - it’s time to step up and enable this potential through BIOS support.