consumer GPU serving economics on 4090

[relaynest82]

Did a bit of testing and thought I'd post the result here.

After a few runs on Threadripper box, I ended up with decent performance, but the setup only really made sense once I adjusted around batch size.

Curious if anyone else landed in the same place.

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

[stackarch]

Thanks, that helps. That matches what I saw on dual 3090. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[macropilot]

That matches what I was seeing. I don't think your conclusion is far off.

[phasestream53]

following this because I'm in the same boat.

[tensorfield88]

Can you fill in a bit more context? Are you testing this on dual 3090 or something else?

[kernelghost]

I tested this tonight. On 4090 with LM Studio, I saw basically the same behavior. I expected the model choice to matter more than it did.

[torchkite]

I tested this tonight. On 4090 with LM Studio, I saw basically the same behavior. I expected the model choice to matter more than it did.

My first guess would be this: If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious. Once you know the plain path works, then add the faster path back in piece by piece.

If you post your exact settings, people here can probably narrow it down pretty fast.

[oxidemixer]

Can you fill in a bit more context? Are you testing this on 7900 XTX or something else?

[rocmpath]

Appreciate the detail. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[prompttinker44]

Appreciate the detail. I don't think your conclusion is far off. Worth rechecking your defaults though, because one bad assumption can make the rest o

following this because I'm in the same boat.