Multiple GPUs on a node?

[Abdulhadi Sahin]

Hi, I want to know if it is possible to connect more than one GPU to a single node. It’s giving an insufficient error, although the site I am using has 10 unused GPUs. It works just fine with a single GPU:

# Add GPU node
node.add_component(model=GPU_CHOICE, name=’gpu1′)
# node.add_component(model=GPU_CHOICE, name=’gpu2′)
# node.add_component(model=GPU_CHOICE, name=’gpu3′)

Thank you!

 

 

 

 

 

[Komal Thareja]

Hi Abdulhadi,

The GPU count you are referring to represents the total number of GPUs available at a site.

No single host at a site has more than 3 GPUs. In fact, only a few hosts are equipped with 3 GPUs. To check the per-host resource details, you can use the notebook: jupyter-examples-main/fabric_examples/fablib_api/sites_and_resources/list_all_resources.ipynb.

For convenience, the following code snippet can also be used:


from fabrictestbed_extensions.fablib.fablib import FablibManager as fablib_manager
fablib = fablib_manager()
fablib.show_config();
fields=['name', 'tesla_t4_capacity','rtx6000_capacity', 'a30_capacity', 'a40_capacity']
output_table = fablib.list_hosts(fields=fields)


Thanks,

Komal

[Abdulhadi Sahin]

Thank you for the information. I used the code suggested, which shows that the FIU site has 4 Tesla and 6 RTX600 GPUs. The FABRIC website confirms the same. However, when trying to add more than one GPU to a node, an error occurs. Is it possible to add more than one GPU to a single node? I assume several nodes can have GPUs added, but the specific goal here is to add multiple GPUs to a single node. Thank you!

[Komal Thareja]

Possibly you changed the call from list_hosts to list_sites. Please see the snippet below.

None of the hosts on FIU have more than 3 GPUs. Also, even 3 can be requested based on availability.

The screenshot only shows the full capacity. You can also check this from portal too.

FIU per host information can be seen here: https://portal.fabric-testbed.net/sites/FIU

Thanks,

Komal

[Author]

Posts