+- Inference (https://www.datasculpting.ai/inference)
+-- Forum: Hardware (https://www.datasculpting.ai/inference/forumdisplay.php?fid=4)
+--- Forum: GPUs (https://www.datasculpting.ai/inference/forumdisplay.php?fid=28)
+--- Thread: VRAM bottleneck on consumer cards (/showthread.php?tid=1325)
VRAM bottleneck on consumer cards - signaltrack44 - 10-17-2026
Did a bit of testing and thought I'd post the result here.
I tried this with Mistral using TensorRT-LLM on 5090. The headline result was better than I expected, but HBM constraints still ended up being the limiting factor.
Nothing scientific here, just practical notes from a real setup.
My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.
Re: VRAM bottleneck on consumer cards - fusedhall - 10-17-2026
following this because I'm in the same boat.
Re: VRAM bottleneck on consumer cards - thetarunner - 10-17-2026
I think the bottleneck is somewhere else. I get why that seems right, but on B200 the bottleneck felt elsewhere. 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.