Update hardware-failure.md #1221
Conversation
Changed the gpu-burn example to use tc instead of d to create more of a workload when testing.
| @@ -156,7 +156,7 @@ make | |||
| And now we can run it like so (this example will run it for 60 minutes/1 hour): | |||
|
|
|||
| ```bash | |||
| ./gpu_burn -d 3600 | |||
| ./gpu_burn -tc 3600 | |||
There was a problem hiding this comment.
How does this behave on old GPUs without tensor cores?
There was a problem hiding this comment.
This will crash on cards older than RTX 20X0
There was a problem hiding this comment.
Let's keep both commands then, and explain that -d must be used for older GPUs.
There was a problem hiding this comment.
I removed the double switch for GTX GPUs and added another prompt for RTX GPUs with the -tc switch.
There was a problem hiding this comment.
Just to add my 2¢
It is probably best to go with a multifaceted approach and do more variations for less time.
-d uses less power than using no argument other than time. While this may stress parts of the GPU more, it doesn't stress the power delivery components as much which could be where the failure is.
./gpu_burn 1800
./gpu_burn -d 1800
./gpu_burn -tc 1800
./gpu_burn -d -tc 1800
Adjusted for RTX GPUs to use -tc and removed the -d switch for GTX GPUS.
Changed the gpu-burn example to use tc instead of d to create more of a workload when testing.