RTX Ada 2000E uncorrected ECC fault simulation

Hi,

How can I force or simulate what happens when an Uncorrectable ECC Fault occurs?

I expect the workload on the GPU crashes, but I am trying to understand how far the ripple reaches when this happens. I would like to test it in a development environment to be sure my application keeps running, properly notices the fault has occurred, and takes appropriate recovery process

My application is in Rocky Linux 8.10, without any display. It is driver version 550.67

Thank you for your help
Irrogical

i tested ddr4 ecc by running it out of spec via overclocking, maybe you can try something similar by increasing the gpu ram speed

Burrito,

Did you get any official guidance from nVidia regarding this approach? Seems like some sort of error would soon occur, but how are you sure it’s ECC and not something else? Is it repeatable?

I like the idea, though it seems it might be more shotgun than rifle

Thanks
Irrogical