and from the github page:
ROCm for Linux:
While ROCm is available for Linux on the releases page, recommend against using it directly. There are simpler and more effective methods for achieving compatibility. The easiest approach is using the
HSA_OVERRIDE_GFX_VERSIONenvironment variable. For example, settingexport HSA_OVERRIDE_GFX_VERSION=11.0.0can override your system’s default and provide support for gfx1103 and similar GPUs.
Seems that despite the new 9070 cards being in warehouses for months now, they haven’t yet finished the work required to support those with ROCM either…
That’s concerning.
I need to correct myself here. According to the link given by James3 it seems all RX6xxx and RX7xxx cards are supported by ROCm now.
This is simply not correct. There is an official support list and none of these GPUs are on it.
Linux: Linux support matrices by ROCm version — Use ROCm on Radeon GPUs
Windows w/ WSL: WSL support matrices by ROCm version — Use ROCm on Radeon GPUs
What you see bundled in the code or on fedora is work-in-progress. Thankfully this actually works pretty good.
But do not mistake experimental code for a stable and performance tuned version of ROCm. This only exists for the GPUs listed above.
As I said above, kudos to AMD for releasing the partial support. And it’s also praiseworthy they’ve not claimed “support” for unfinished code. But it’s bad they’ve not actually finished the job.
BTW, if you want a peek at what AMD seems to be thinking for longer term (~2028?) plans see here. Very cool, but safe to assume it wont help anybody for a couple generations of product yet.
That’s Interesting.
You are right, my apologies.
I installed pytorch with ROCm-6.3.3 from https://download.pytorch.org/whl/nightly/rocm6.3.3 and at least gfx-1101 is included.
My RX7800XT is supported out-of-the box.
Glad your 7800 XT is usable at least to some degree for you now. 
But, at the risk of being too pedantic, your 7800XT is “working” but is still a work-in-progress. Does that make sense? i.e. The code is not finished and thus not listed on the supported matrix. This what was discussed above.
The partial implementation is good to see, but until AMD is able to add it to their “supported” list you know there are still some bugs and skeletons that need quashing. So it’s a fair criticism of AMD that more than a year after 7800 XT was released they have still not done the boring work of finishing and validating the ROCm support for this GPU. Same for the FW16 GPU options.
The lack of support is undoubtly a problem for AMD.
Sadly I think it is intended to sell the big cards and the dedicated AI cards.
The announcement of this event that should make clear where AMD and ROCm are at. I’m looking forward to getting some clarity
Oh, and this too
But the comment about Strix Halo support I think is Linux related as the official doc notes support for Windows for AI Max+ Pro 395
gfx1100 gfx1101 gfx1102 gfx1103 gfx1150 gfx1151 gfx1152
Normally fedora 42 has already enabled the card with release 6.3.
For now the only probleme is that on gfx1103 (the iGPU of the FW16) the MES firmware is not fully stable. Have to wait to see how it is for the MAX. 
Also worth noting: gfx1151: I have pytorch, pytorch-vision, and hipBLASLt working well enough to run GPT2 · ROCm/TheRock · Discussion #244 · GitHub
But it’s sad, that this is coming from the community, rather than from AMD themselves.