NVIDIA
Looking for driver for P100 to run tensorflow in ESX 6.5
Hi, My company wanted to purchase P100 to run tensorflow on ESX6.5 hosts, but I was told by VMWare that the driver will not be ready until Dec this year. I had downloaded an eval driver 384.73-385.41 but there is confusion on the user guide as user said Linux Guest VM support: 64-bit Linux guest VMs are supported only on Q-series GRID vGPUs. So it still not able to support tensorflow as they need 64bit Linux. Not sure if this correct. Can someone advise when will the driver be ready. thank you.
Hi,

My company wanted to purchase P100 to run tensorflow on ESX6.5 hosts, but I was told by VMWare that the driver will not be ready until Dec this year. I had downloaded an eval driver 384.73-385.41 but there is confusion on the user guide as user said Linux Guest VM support: 64-bit Linux guest VMs are supported only on Q-series GRID vGPUs. So it still not able to support tensorflow as they need 64bit Linux.

Not sure if this correct.

Can someone advise when will the driver be ready. thank you.

#1
Posted 10/16/2017 03:26 PM   
Hi I’ve not done any work with TensorFlow before, so I may be misunderstanding what you’re asking ... You have a couple of different driver options ... 1 is to use it in Passthrough mode so that a single VM can use the entire P100. When used like this, you just need the standard Linux P100 driver, as ESXi won’t need a driver installed. With this method, there are no driver licensing costs and you can use multiple P100s within the same VM to enhance performance. 2 is to virtualise the P100 and share it between multiple VMs. When used like this, both ESXi and the VMs will need drivers installed. The latest GRID 5 drivers should support this. There are additional licensing costs involved here, and you can only have a single vGPU per VM, so there are performance limitations to be aware of as well. I don't know what driver VMware are referring to in your comment above, as the P100 already works with ESXi. Just out of interest, how many P100s are you planning to use in the server / system? Regards Ben
Hi

I’ve not done any work with TensorFlow before, so I may be misunderstanding what you’re asking ...

You have a couple of different driver options ...

1 is to use it in Passthrough mode so that a single VM can use the entire P100. When used like this, you just need the standard Linux P100 driver, as ESXi won’t need a driver installed. With this method, there are no driver licensing costs and you can use multiple P100s within the same VM to enhance performance.

2 is to virtualise the P100 and share it between multiple VMs. When used like this, both ESXi and the VMs will need drivers installed. The latest GRID 5 drivers should support this. There are additional licensing costs involved here, and you can only have a single vGPU per VM, so there are performance limitations to be aware of as well.

I don't know what driver VMware are referring to in your comment above, as the P100 already works with ESXi.

Just out of interest, how many P100s are you planning to use in the server / system?

Regards

Ben

#2
Posted 10/17/2017 07:50 PM   
Hi, I would also be interested in better understanding the use case here. Do you want to run P100 with vGPU or is it just to run it "virtual" with GPU Passthrough? Regards Simon
Hi,

I would also be interested in better understanding the use case here. Do you want to run P100 with vGPU or is it just to run it "virtual" with GPU Passthrough?

Regards

Simon

#3
Posted 10/18/2017 11:58 AM   
Scroll To Top

Add Reply