4090 vs 5090 for 70B quants on dual 3090

[turboyard94]

Thought I'd post a small recipe because a lot of threads ask similar things.

This isn't revolutionary, but the combination of SGLang and 7900 XTX felt surprisingly usable once I tuned around continuous batching.

Maybe useful as a starting point for newer folks.

I care more about stability than squeezing the last possible tokens/sec out of it.

[fusedhall]

Good point. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[paperrunner]

Maybe, but I had the opposite result. I get why that seems right, but on GB200 the bottleneck felt elsewhere. Not saying you're wrong, just that I wouldn't lock that in as the lesson yet.

[torchtrack24]

I'd start with the simpler stack.

[signaltrack44]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[latentpeak22]

worth checking the driver first.

[paperrunner]

I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Mistral and compare from there. Once you know the plain path works, then add the faster path back in piece by piece.

[embersmith]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at tool calling. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[turbogrid11]

I'd start with the simpler stack.

[deltastation]

I tested this tonight. My result was mixed: first run looked bad, second run looked normal.