+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Inference Engines (https://www.datasculpting.ai/inference/forumdisplay.php?fid=9)
+--- Forum: Other Engines (https://www.datasculpting.ai/inference/forumdisplay.php?fid=39)
+--- Thread: prefill bottlenecks on longer prompts on A6000 (/showthread.php?tid=466)
Pages:
1
2
prefill bottlenecks on longer prompts on A6000 - northquill - 09-24-2026
This keeps coming up in conversation so I figured I'd start a thread.
The more I work with local setups, the more it seems like HBM constraints decides everything downstream.
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.
Re: prefill bottlenecks on longer prompts on A6000 - mintstream9 - 09-24-2026
Quote:This keeps coming up in conversation so I figured I'd start a thread.
Good point. That matches what I saw on A6000. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
Re: prefill bottlenecks on longer prompts on A6000 - thetatinker77 - 09-25-2026
A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at continuous batching.
Re: prefill bottlenecks on longer prompts on A6000 - kernelrunner - 09-28-2026
That matches what I was seeing. I ended up in about the same place using PyTorch. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.
If you post your exact settings, people here can probably narrow it down pretty fast.
Re: prefill bottlenecks on longer prompts on A6000 - vectorcrate98 - 09-28-2026
My first guess would be this: Before rebuilding anything, I'd test one clean baseline with Llama and compare from there.
Re: prefill bottlenecks on longer prompts on A6000 - kernelarch - 09-28-2026
My first guess would be this: On M3 Max, the easiest win is often to back off the aggressive settings and see whether the problem survives. The logs were somehow calmer than I was.
Re: prefill bottlenecks on longer prompts on A6000 - byteyard - 10-01-2026
I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Phi and compare from there.
Re: prefill bottlenecks on longer prompts on A6000 - ggufvault - 10-02-2026
Appreciate the detail. The pattern around MoE serving has been pretty similar for me. The biggest change for me was simplifying the stack instead of adding one more optimization layer.
Re: prefill bottlenecks on longer prompts on A6000 - oxidegrove40 - 10-02-2026
Quote:Appreciate the detail. The pattern around MoE serving has been pretty similar for me. The biggest change for me was simplifying the stack instead of a
A couple things I'd check first. On 4090, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece.
Re: prefill bottlenecks on longer prompts on A6000 - bytelab - 10-03-2026
Need a little more detail here. Is this local-only or are you routing anything externally?