Hacker News new | ask | show | jobs
by stickydink 1930 days ago
(I'm not super informed in this space)

AWS WorkSpaces purports to have machines with meaty GPU's, are they not powerful enough?

4 comments

The problem with AWS Workspaces for us was the price was just too damn high.

Their Graphics Pro instance was okay spec: 16 vCPU, 122 GB memory, 8 GB VRAM. But they charge $1000/month for it.

Our $3000 workstations beat it in every benchmark, sometimes by 3-5x and have a useful lifetime of at least three years.

I'm sure for some people, that is a valid trade off, but for us, it was just too much.

Are you rendering 24/7....?
Most cloud GPUs are just thin-sliced datacenter GPUs which only have one redeaming quality: high RAM per PCIe slot.

The $100,000 GPU can be beaten by a $5,000 consumer space CPU for raw compute but is basically 3 to 4 times the amount of RAM.

There's a niche in the space for workspaces that support "low intensity 3D applications" with 99.99% uptime. Right now, all the major cloud vendors (AWS/Azure/GCP) are prohibitive with respect to pricing. I can spend $600 to buy a dedicated machine to run my GPU-bound workload 24 hours a day, or spend $600/month to have a dedicated GPU instance. Things like Paperspace don't cater to this particular workload, still. I've honestly thought about creating a new PaaS provider that attacks this niche. Seems doable.
I'm sure they are, I just need to make the leap. :)