[GUIDE] Successful Wi-Fi 7 (802.11be) on Framework 13 AMD with Qualcomm QCNCM865 and Arch Linux

I go ahold of some of these yesterday and all I can say is, they’re just barely functional on Linux right now.

On kernel 6.8.2 on Arch, you need to do what @GreyXor says in the original post just to get the card to function, however I ran into several issues:

  1. Large loads, like a speed test, will cause the card to crash and become unavailable until the machine is completely powered off. Whatever happens to the card also causes the machine to be unable to reboot (i.e. Linux will do its reboot procedure and appear to reboot the machine, but then it just hangs with a blank screen, similarly to what happens when you try to power on the laptop with the BE200 installed).
  2. Bluetooth worked until something disconnects or you try to turn off bluetooth, at which point Bluetooth becomes non-functional until powered off.
  3. 6ghz networks can be seen but will not connect.

I tried using Kvalo’s ath kernel by modifying the linux-git aur package, and while it compiled just fine I was greeted with a blank screen upon reboot. I didn’t spend any more time looking into using that kernel.

Then I tried linux-mainline from the aur, which I believe has most of the patches already from Kvalo, and it worked better, but still had some problems:

  1. While more stable, speed tests were only around 200Mbit on 5ghz when I can get 800+ with an AX210.
  2. 6ghz WPA3 networks can be seen but disconnect pretty much immediately after connecting.
  3. 6ghz WPA3 Enterprise networks cannot be seen at all and will not connect even if trying to connect to a “hidden” network.

So, as of right now my AX210 wifi cards are going back in my machines until either these become more stable, a workaround for getting the BE200 to work on AMD is discovered (though I hear it’s still lacking proper Linux support, so there’s that), the BE202 variant becomes available (It’s supposed to be A+E keyed which will likely solve the problem), or some other viable alternative comes around for Linux that doesn’t require an out-of-tree driver (so, no Broadcom and likely no Realtek adapters).

Edit: Forgot to mention: on linux-mainline 6.9-rc1, it’s no longer necessary to pull down the regdb.bin and boards.bin files from git.

6 Likes

Bravo! Exactly what I needed. Worked for me with the same Wi-Fi card on a fresh install of Endeavour OS. Thank you.

Hello,
yes you can use the aur package linux-mainline and changing the source to:
“$_srcname::git+https://git.kernel.org/pub/scm/linux/kernel/git/kvalo/ath.git#branch=master-pending”

After 2 weeks of use on my framework and a couple weeks on my Alienware I can confirm it works great and has none of the issues I had seen with previous Qualcomm cars I have used. (I have yet to use the included card) note I run win 11 pro

1 Like

Does this card work on Fedora 40 (newer kernels) out of the box now? Or do we still have to do stuff to make it work?

I didn’t try it in Fedora 40 exactly. But I try this WiFi card on 39 with 6.9 kernel preview.

It was a little better, but still unusable. WiFi dies on hard load, Bluetooth can’t connect to my headphones.

Does it need an adapter for Ryzen 7 5700U and currently has a WiFi6 card right now. I know with computers with Intel processors would just use an E key adapter. Not exactly positive if I need anything else for my mini PC other than the QCNCM865

Crap, it sounds like the experience I currently have with the Mediatek RZ616.

I guess I’ll just get the AX210 for now. It’s a out $20 on Amazon. A reliable connection for $20 is better than a flakey connection with the other WIFI cards (Qualcomm, Mediatek, and the BE200 won’t even work…). It’s like we don’t really have a choice but to get the AX210.

Update on my previous analysis:

The 6.10 kernel appears to have a bunch of fixes merged in and the Qualcomm card is actually functional now. I can connect to WPA3 and even WPA3 Enterprise networks and get download speeds of around 800Mbit/s, however specifically on AMD machines the card just… stops working after the download part of a local speed test. The card just refuses to respond after that; it’s still visible on the PCI bus, but I’m unable to power it off, unbind it, etc. to try and restart it. It just… stops. It does however on Intel machines just fine, but on those you can just use the new BE200 intel chips which I imagine would function much better.

I’ve wasted enough time now trying to get some sort of WiFi 7 on these machines on Linux.

2 Likes

Do you think it’s just something that will be fixed with the next kernel update?

This is the only viable WIFI7 card for Linux users on AMD it seems (I’m on a FW16 and Fedora myself).

I’ve been keeping up with patches that show up on patchwork.kernel.org, however as of yesterday it’s still crashing on upload. I’m applying those patches on top of kvalo’s master branch which I’ve merged on top of Linus Torvalds’ master branch to get the latest changes from everywhere.

I should probably submit a bug report but I’m tired of swapping stuff for now so I likely won’t be submitting one for a while.

One note is that, for an Intel framework motherboard, the QCNCM865 functions just fine (for wifi, anyway) and does not crash, so it’s possible it’s something AMD-specific that’s causing it.

I have some BE202 and MT7925 cards on order and will try those when I get them sometime over the next couple weeks. The BE202 will likely have the same problem as the BE200; I’m hoping that the MT7925 cards will work however given the problems people have with the 6E versions I’m not going to hold my breath for a bug-free experience.

2 Likes

I still have two issue under 6.10.0-rc1-1-mainline.

  • I cannot connect to any Audio Bluetooth devices (but other bt, like mouse is working good)
  • Something(driver?) crashing when I upload too fast(or too much?)

apart from these two problems, it works very well, with excellent speeds and latencies.

I think it’s too much uploading. If I use LibreSpeed on my own network, it crashes nearly instantly when the upload test starts, whereas if I go and use Speedtest.net, which is going to get much slower upload speeds, I can make it through probably one test before it crashes midway through the second one.

I’ve tried enabling debugging by setting ath12k.debug_mask=0xffffffff and looking at the output, but as I mentioned and as you’ve noticed, the output from the ath12k just stops completely. Then maybe 10-30 seconds or later I start getting wmi and other timeouts from ath12k.

I thought maybe it had something to do with power management differences between Intel and AMD (since it seems to not crash on Intel machines), but turning power saving off via iw, disabling aspm, and changing the device’s power PCIe power setting to “on” instead of “auto” does not help.

My guess now is that there’s a spinlock or something in the kernel that’s ending up in a deadlock, or that the firmware in the card is just crashing and then stops responding. I’m leaning toward the crashing part since, if I had reset the machine, the card generally won’t come back up on the first reboot. I have to reboot for a second time to get the card to start responding again.

P.S. I should also mention that I’m doing most of my testing on a BeeLink GTR7 Pro, as it exhibits the same behavior as my Framework’s and I don’t want to be tearing those open all the time for testing. Once I find something that is working on the BeeLink I’ll confirm that on one of my Framework’s before I got putting them in everything.

P.P.S. I haven’t really tested bluetooth at all; I think I at one point connected a mouse which may have worked, however I can’t recall if it was this card or another. I don’t use bluetooth audio streaming either so I wouldn’t be able to confirm that. Bluetooth for me is a secondary concern as I really don’t use it at all and is completely turned off most of the time.

Edit: Actually now that I think about it, this is the same behavior I initially had with an XPS 13 9310 several years ago that came with a Qualcomm card and used the new-at-the-time ath10k driver. I can’t recall if I just dealt with it or if it got fixed at some point as I’m usually connected to ethernet anyway. I no longer have the machine though so I can’t check it to see if it still has the same behavior or not.

I got some BE202’s today, all E-key only, and they cause a boot loop just like the BE200’s. Externally they appear to be identical other than the model number, and in fact use the same FCC ID as the BE200.

1 Like

What about the NCM865’s predecessor, NFA765, a.k.a FastConnect 6900? A WiFi 6E card that is rumored to do 2882Mbps on 5GHz

Was hoping for WIFI 7.

Thanks for testing and reporting back. So it looks like the BE202 and MT7925 are 160 MHz whereas the BE200 and the QCNCM865 are 320 MHz. Hoping for a 320 Mhz WIFI 7 card that works out of the box with AMD and Fedora. I guess I’ll continue to wait. The Intel AX200 is serviceable for now. It’s not killing my 5 GHz radio like the RZ616 at least.

I’m not looking for a 6E card. I already have a AX200 (6E) to keep myself afloat. I’m just hoping to get a WIFI 7 320 MHz card as the next upgrade. I just got the AX200 to keep from killing my 5GHz AP radio. For some reason, the RZ616 (6E) keeps knocking the AP radio offline which causes other endpoints to lose connection until I reboot the AP. It only happens when there’s sustained heavy traffic with the RZ616 (speedtest.net, file transfers to my NAS, etc), etc.

I got a handful of MT7925 (aka RZ717) cards today; On Linux, they only get ~200Mbit up/down connecting to the same network I can get 600-900Mbit from the AX210 cards. So, even though they’re technically better and Wi-Fi 7, I’ll stick to the AX210 cards for now.

Didn’t test Bluetooth…

@GreyXor Are you still having the crashing when uploading issue? I haven’t tried this in a couple weeks.

Yes, in 6.10.0-rc3. But I should test ath12k master from git

1 Like

I finally replaced my AP. Just to recap since I’m sure people don’t follow me and my issues - the old one was an Asus RT-AC3200 and the RZ616 kept killing my AP’s 5 GHz radio which knocks out all connectivity for all of my clients on the same band when the RZ616 is doing heavy transfers. I replaced the RZ616 with the AX210, and speeds was more or less the same, but more importantly, it was not killing my AP.

I now saved enough to get a TPLINK EAP773 AP (WIFI7), and the AX210 is performing very well. I was pretty impressed. however, I wanted to see how the RZ616 behaves and performs. Well, no knocking out my AP - good. BUT… in Fedora, I had a hard time connecting and STAY connected (once it does connect). When it DID finally stay connected (after multiple tries), I was only getting 200 Mbps down and 100 Mps up. In Windows 11, it connected quickly, BUT speeds were slower than the AX210. I was getting 800 down 700 up. With the RZ616, I get 690 down and 330 up.

I’m going back to the AX210. And based on my experience with mediatek, I will NOT be getting a WIFI7 card from them. So I’m kinda stuck if I want a WIFI7 card. I can only hope that either the BE200 or the Qualcomm card gets better support on both Linux and Windows with the AMD platform.

Color me disillusioned regarding this WIFI7/AMD situation.

1 Like