Switching between two GPUs Using only one GPU at a time for development

[twinbee]

I have a 9600 GT GPU at the moment.

For development purposes, I’d like to purchase a Fermi GPU on top to sit aside the current 9600. However, I don’t want to use both at the same time, but rather each separately so that I can test performance on both of them individually. Obviously, I could physically swap them in and out, but that would be a pain.

So basically, I’m wondering if there’s a ‘software’ switch that would only recognize one of the GPUs at boot time?

[twinbee]

I have a 9600 GT GPU at the moment.

For development purposes, I’d like to purchase a Fermi GPU on top to sit aside the current 9600. However, I don’t want to use both at the same time, but rather each separately so that I can test performance on both of them individually. Obviously, I could physically swap them in and out, but that would be a pain.

So basically, I’m wondering if there’s a ‘software’ switch that would only recognize one of the GPUs at boot time?

[seibert]

For development purposes, I’d like to purchase a Fermi GPU on top to sit aside the current 9600. However, I don’t want to use both at the same time, but rather each separately so that I can test performance on both of them individually. Obviously, I could physically swap them in and out, but that would be a pain.

Is this for power reasons? OS compatibility?

So basically, I’m wondering if there’s a ‘software’ switch that would only recognize one of the GPUs at boot time?

On Linux, you can now hide a device from a user by changing the permissions on the /dev/nvidia[number] file. On Windows, I have no idea.

[seibert]

For development purposes, I’d like to purchase a Fermi GPU on top to sit aside the current 9600. However, I don’t want to use both at the same time, but rather each separately so that I can test performance on both of them individually. Obviously, I could physically swap them in and out, but that would be a pain.

Is this for power reasons? OS compatibility?

So basically, I’m wondering if there’s a ‘software’ switch that would only recognize one of the GPUs at boot time?

On Linux, you can now hide a device from a user by changing the permissions on the /dev/nvidia[number] file. On Windows, I have no idea.

[cbuchner1]

I have a 9600 GT GPU at the moment.

For development purposes, I’d like to purchase a Fermi GPU on top to sit aside the current 9600. However, I don’t want to use both at the same time, but rather each separately so that I can test performance on both of them individually. Obviously, I could physically swap them in and out, but that would be a pain.

Device Manager on Windows. Disable the card you don’t want available.

But anyway, you can tell your CUDA application on which of the available devices to run.

So messing with Device Manager is rather pointless.

[cbuchner1]

I have a 9600 GT GPU at the moment.

For development purposes, I’d like to purchase a Fermi GPU on top to sit aside the current 9600. However, I don’t want to use both at the same time, but rather each separately so that I can test performance on both of them individually. Obviously, I could physically swap them in and out, but that would be a pain.

Device Manager on Windows. Disable the card you don’t want available.

But anyway, you can tell your CUDA application on which of the available devices to run.

So messing with Device Manager is rather pointless.

[twinbee]

Thanks both!

Good to speak to you here cbuchner1 - finally managed to drag myself into learning some GPU stuff. I’m bracing myself!

Is this for power reasons? OS compatibility?

It’s so I can switch between them both when testing my code. I want to run code on them individually to see the differences.

So messing with Device Manager is rather pointless.

Not to mention seemingly dangerous if I disable the card and the monitor blacks out without the other card kicking in. I guess windows has some safety net in that case though (which I might need if I experiment with OpenCL or some other environments).

[twinbee]

Thanks both!

Good to speak to you here cbuchner1 - finally managed to drag myself into learning some GPU stuff. I’m bracing myself!

Is this for power reasons? OS compatibility?

It’s so I can switch between them both when testing my code. I want to run code on them individually to see the differences.

So messing with Device Manager is rather pointless.

Not to mention seemingly dangerous if I disable the card and the monitor blacks out without the other card kicking in. I guess windows has some safety net in that case though (which I might need if I experiment with OpenCL or some other environments).

[seibert]

It’s so I can switch between them both when testing my code. I want to run code on them individually to see the differences.

cudaSetDevice() is all you need, then.

[seibert]

It’s so I can switch between them both when testing my code. I want to run code on them individually to see the differences.

cudaSetDevice() is all you need, then.