VRAM bottleneck on consumer cards for Llama

[paperrunner]

Posting this in case it saves someone else a little time.

If you were starting from scratch today for this use case, what would you choose first and what would you ignore?

Would rather learn from your scars than collect my own all at once.

For context, I'm not chasing leaderboard numbers. I mostly want the system to feel consistent day to day.

[cudapoint]

same here on my side.

[vrambandit]

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at tokens/sec. Once you know the plain path works, then add the faster path back in piece by piece.

[kernelarch]

Need a little more detail here. Is this local-only or are you routing anything externally?

[audiorunner74]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there. Once you know the plain path works, then add the faster path back in piece by piece.

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

[fusedhall]

What worked for me was pretty simple. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at token cost. Once you know the plain path works, then add the faster path back in piece by piece.

[turbogrid11]

Thanks, that helps. The pattern around router policy has been pretty similar for me.

[batchsize]

That matches what I was seeing. The pattern around HBM constraints has been pretty similar for me. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

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

[fusedhall]

same here on my side.

[turbogrid11]

Good point. The pattern around consumer GPUs has been pretty similar for me.