Thats strange... which version exactly?  Can you give more data (settings)? I have to admit I did test more with V56 but for a V64 thats far too low even without the improvements of 0.9.4.

Only data I got of a direct comparison with trm on Win was by a user using 0.9.2. There indeed trm was faster but also used more energy.
With 0.9.4 on Linux I am pretty sure that flipped

Edit: slept over that msg and now wonder if you can tell me what driver version and CPU you use. All my Win tests refer to Adrenalin 19.10.1 and its one card on a rather strong CPU. Saw on Linux that the driver may make a difference here, e.g. two cards, Xeon server GPU: 19.30 driver is slower then ROCm 2.10, 6 card rig Celeron: 19.30 is quicker then ROCm 2.10.

So maybe driver and or CPU may be the issue on Windows, too.