I thought GCP only supports Linux, but it seems not. Today I noticed there are Windows servers options.
So, (don’t call me an idiot!) I started testing GCP windows VM machine to run Rhinoceros. I have my own code that calls Cuda APIs from within a C# component. There are also heavy for loops that run parallel.
It is really easy. But 300$ free evaluation credit cannot be used for windows VM / GPU. I needed both. So I needed to start billing with no evaluation period, which is a bit scarely. (I haven’t received my first billing report, so the below report does not contain information about cost/prising.)
Not all machines support GPUs. You need to do a quick search to figure out which machine types support GPUs.
-
the default GPU quota is set to 0. you need to request a raise of quota. I just requested only 1 GPU and was approved immediately.
-
CPU quota is set to 24. I requested 16 vCPUs. Sadly, it was slower than my Laptop CPU (core i7 6 cores) for multi-threading.
-
local storage quota is set to 500gb. So I assigned 500gb all to my single VM.
-
I requested Tesla T4 for my GPU.
-
Rendered/Raytraced mode does not work properly so far over a RDP connection.
-
CUDA performance is a bit faster than my laptop (geforce 1650 mobile max-q).
I don’t know if anyone finds this information useful. I expected something 100x faster than my laptop, but it seems not. But there are faster GPUs, more CPUs you can assign. Thinking this is a good option for some heavy multi-threading purpose.