+- 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: RTX 5090 local inference on M3 Max (/showthread.php?tid=1469)
Re: RTX 5090 local inference on M3 Max - ggufwire22 - 10-18-2026
What worked for me was pretty simple. If you're on MLX, 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.
Re: RTX 5090 local inference on M3 Max - rasterhall - 10-18-2026
I tested this tonight. On Threadripper box with llama.cpp, I saw basically the same behavior. The surprise was that memory pressure showed up before throughput collapsed.
Re: RTX 5090 local inference on M3 Max - thetarunner - 10-18-2026
same here on my side.
Re: RTX 5090 local inference on M3 Max - fusedhall - 10-18-2026
Tried to reproduce it just now. The issue went away once I changed one setting and restarted the stack cleanly. I expected the model choice to matter more than it did.
Re: RTX 5090 local inference on M3 Max - torchshed - 10-19-2026
Good point. I ended up in about the same place using LM Studio. The biggest change for me was simplifying the stack instead of adding one more optimization layer. I keep promising myself this is the last rebuild.
Re: RTX 5090 local inference on M3 Max - dockkite69 - 10-19-2026
following this because I'm in the same boat.
Re: RTX 5090 local inference on M3 Max - signaltrack44 - 10-19-2026
I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.
RTX 5090 local inference on M3 Max - thetarunner - 10-19-2026
Spent some time on this today and wanted to compare notes.
Two paths both seem viable, but they break down differently once MoE serving becomes important.
Interested in tradeoffs more than fan loyalty.
This is mostly for local work rather than public serving.
If I had to summarize the lesson so far, it would be that clean setup and observability beat clever guesses.
Re: RTX 5090 local inference on M3 Max - batchwire - 10-19-2026
A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at tail latency.