|
ROCm version that finally behaves with MLX
|
09-02-2026, 03:22 PM
I tested this tonight. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed.
I'm happy to test one more variant if you want a second data point.
throughput matters, but tail latency matters too
|
|
« Next Oldest | Next Newest »
|
| Messages In This Thread |
|
ROCm version that finally behaves with MLX - by tensorworks - 09-02-2026, 12:36 AM
Re: ROCm version that finally behaves with MLX - by relaynest82 - 09-02-2026, 02:43 AM
Re: ROCm version that finally behaves with MLX - by rocmmixer - 09-04-2026, 12:48 AM
Re: ROCm version that finally behaves with MLX - by batchsize - 09-02-2026, 03:22 PM
Re: ROCm version that finally behaves with MLX - by dockworks23 - 09-03-2026, 04:06 PM
Re: ROCm version that finally behaves with MLX - by tinycontext - 09-04-2026, 03:05 AM
Re: ROCm version that finally behaves with MLX - by batchyard - 09-04-2026, 07:41 AM
Re: ROCm version that finally behaves with MLX - by tensorworks - 09-04-2026, 08:27 AM
Re: ROCm version that finally behaves with MLX - by rocmmixer - 09-04-2026, 08:37 PM
Re: ROCm version that finally behaves with MLX - by batchyard - 09-06-2026, 09:09 AM
Re: ROCm version that finally behaves with MLX - by thetaember38 - 09-06-2026, 09:43 PM
|
