throughput per watt reality check

[signalcraft14]

Wanted to compare notes with people doing similar things.

I'm mostly working with Mixtral in MLX on B200. The part I'm trying to understand is tokens/sec.

Would appreciate practical advice rather than theory.

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

[bytestack]

Tried to reproduce it just now. The issue went away once I changed one setting and restarted the stack cleanly.

[fastfield4]

Maybe, but I had the opposite result. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Worth testing the simpler explanation before adopting the more interesting one.

I'm happy to test one more variant if you want a second data point.

[bytestack]

I'd start with the simpler stack.

[cobalttinker]

same here on my side.

[edgekite]

My first guess would be this: On 4090, the easiest win is often to back off the aggressive settings and see whether the problem survives.

[relaysupply90]

A couple things I'd check first. 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.

[pixelcrate82]

Did a small comparison this afternoon. On B200 with SGLang, I saw basically the same behavior. I expected the model choice to matter more than it did.

[fusedhall56]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at small local stacks. Not glamorous advice, but repeatable baselines save a lot of time.

[deepseeked]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at small local stacks. Not glamo

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with FLUX and compare from there.