Anyone Tried OVH public cloud GPU

bdspice

I saw ovh claimed there public cloud gpu is 1000x more fast then cpu. so i wonder what if i use gpu for my ffmpeg script which is taking much time on normal cpu vps. I am wondering if a 10000 seconds video encoding take 10-100 seconds on gpu instead of 5000seconds on a single core cpu.
anyone tried ovh gpu or any public gpu with nginx, php-fpm ? also ovh offer 200$ credit for 1month if upgrade account by pay some dollar.

bdspice

it seems no one tried ovh gpu yet here

quicksilver03

Some colleagues at work use OVH GPU instances, not to encode video though.

I had to set up one of those instances and it was quite annoying, OVH didn't even test the OS images with the model in question and it tooks days for them to tell me that I had to use the OpenStack API to add an undocumented tag to the image for the instance to boot with it.

Also, their own guide on how to install the NVIDIA drivers didn't work, I cannot remember the exact details but it took a few days of fiddling to get it to work.

Considering the time you risk to waste with OVH GPUs, I don't think it's worth to move your encode jobs to it.

rjbl

I tried the V100 with the Pytorch template, using trial credits. There is less involved setting up but its definitely overpriced with the template addons and in comparison to niche providers. The whole experience was typical OVH experience = buggy but without the cheap factor.

bdspice

@quicksilver03 said:
Some colleagues at work use OVH GPU instances, not to encode video though.

I had to set up one of those instances and it was quite annoying, OVH didn't even test the OS images with the model in question and it tooks days for them to tell me that I had to use the OpenStack API to add an undocumented tag to the image for the instance to boot with it.

Also, their own guide on how to install the NVIDIA drivers didn't work, I cannot remember the exact details but it took a few days of fiddling to get it to work.

Considering the time you risk to waste with OVH GPUs, I don't think it's worth to move your encode jobs to it.

i intend to use REHL os like centos 7 or almalinux or rockylinux 8. so i found this on there tutorial for NVIDIA drivers

    sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo
    sudo dnf clean all
    sudo dnf -y install cuda-toolkit-12-4 cuda
    sudo dnf -y module install nvidia-driver:latest-dkms
    uname -m && cat /etc/*release
    reboot
    nvidia-smi

bdspice

@rjbl said:
I tried the V100 with the Pytorch template, using trial credits. There is less involved setting up but its definitely overpriced with the template addons and in comparison to niche providers. The whole experience was typical OVH experience = buggy but without the cheap factor.

Thanks. now i am fear to try them. i think i should quit thinking of using gpu.