consumer GPU serving economics

[stackarch]

This keeps coming up in conversation so I figured I'd start a thread.

One thing I keep noticing is that people underestimate long context until they actually try to run this stuff for more than one afternoon.

Wondering if this is just where the whole space is heading now.

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

[signalcraft14]

Interesting, I hadn't tried that. That matches what I saw on M3 Max. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[bytestack]

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at KV cache pressure.

[turboscope5]

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at KV cache pressure.

worth checking the driver first.

[localtinker]

My first guess would be this: If you're on llama.cpp, I'd confirm the version first and then reduce variables until the behavior is obvious. Not glamorous advice, but repeatable baselines save a lot of time.

[cudapeak]

I tested this tonight. The issue went away once I changed one setting and restarted the stack cleanly.

[rocmhouse90]

Maybe, but I had the opposite result. I've seen people blame HBM constraints when the real issue was much more mundane. Could be a workload difference, but I wouldn't generalize too fast from one setup.

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

[stackarch]

I don't think that's the whole story. I've seen people blame prefill/decode split when the real issue was much more mundane. Worth testing the simpler explanation before adopting the more interesting one.

[turboscope5]

Appreciate the detail. That matches what I saw on mini PC. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[vramport]

A couple things I'd check first. If you're on vLLM, I'd confirm the version first and then reduce variables until the behavior is obvious. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.