You can actually use `fwupdtool install-blob` to update the UEFI if the board supports and uses UEFI capsule update. I did that with my Asus B550-I and it just worked, it reboots to ASUS EZ-Flash and does the update automatically.
All Asus would need to do is publish updates on LVFS.
I've updated my Thinkpad through the GNOME Software Store at some point, it was listed as a "system update" that automatically used fwupdmgr to install the latest UEFI image.
I don't know if the Legion line has the same level of support, but you may just be able to ru "fwupdmgr update" and get prompted to reboot for an UEFI/Intel ME/SSD firmware update.
Ooh that's very interesting. I had to reverse engineering the firmware update process for my work laptop[0], I'm guessing the update process there doesn't use capsule updates though...
They have two accounts on that service, and have uploaded firmware, but the one article I can find (Phoronix) indicates in 2020 it was only for a single board. Would indeed be nice if they'd expand that out to more boards.
I have an ASUS motherboard and recently went through the process of updating the BIOS. One of the things worth nothing is that it is not safe to just get the latest version and try to run EZ-flash. Some of the updates will update EZ-flash as well and those become required for installing later updates.
The download page actually has release notes which specify when a certain BIOS is required, but you basically have to read back through all of them to figure out which ones to install.
The annoying thing about the release notes is whether buying new or used, the BIOS is usually wildly out of date, so you end up in an unwinnable scenario.
I updated my asus mainboard recently too. It was straightforward: Copy the new firmware image into the efi partition somewhere. Reboot. Select the efi fat32 partition and the firmware blob therein in ez flash. Update.
Why format an usb stick when there is already a perfectly good fat32 partition lying around?
Does anyone actually use exFAT? I know it's a simple cross platform file system and all, but every phone, tablet, laptop and desktop OS supports NTFS these days. Especially for read only stuff.
Guess I was lucky then, I flashed a very similar new asus motherboard from this same generation for AMD ryzen 7000 cpu's, I used the first random usb stick lying around, wrote the bios update to it from linux, possibly had to rename the file to something specific (manually. Who needs a tool for that? Also why doesn't asus just give it the correct name in the zip file?), and flashing worked first try. May have been a usb 2.0 stick though.
Rebooting after that does take a while though (minutes) since it needs to re-learn the RAM. And this is a painful process indeed, especially when I originally just had the mobo since I had probably not seated the RAM to the fullest extent (DDR 4 seems to require pressing it very well even after it already clearly clicked in, after pushing it harder it worked well), and the only feedback from the mobo was very similar LED codes as when it "learns" the RAM, and sometimes even half working...
I recently updated an MSI Z690-A to the latest BIOS in preparation for installing a 13th generation Intel. I downloaded the updated BIOS file, put it on a thumb drive, renamed it to "MSI.ROM," put the thumb drive into the "Flash BIOS" port, had the motherboard connected to ATX and CPU1 power only, no CPU, or RAM, or any other hardware installed, flipped on power, pressed the "Flash BIOS" button and everything worked fine. This board "learned" the RAM sticks very quickly, maybe because the board never had RAM to begin with. It took seconds.
I was originally going to pick up an Asus Strix Z690-A but read about a variety of problems with that board, especially with regards to RAM. Many anecdotes I had seen online had similar complaints about need to really press hard to get the RAM sticks properly installed. I had no issues with the MSI. My previous build had an Asus mobo and it had strange issues, too.
Yeah I’ll give a +1 to MSI, their BIOS firmware updates have been nothing but smooth for me and I haven’t had to think much about which USB sticks I’ve used. I’m about 80% sure I’ve used a USB 3 stick at one point but can’t be certain
IF you read the special notes about a not-too-big flash drive and FAT and all that.
IIRC I also had to try several USB ports in back and something else hacky to get the flash going. It was still half black incantation and performative dance learned from the internet, and about 5% directions provided by MSI on their website or within BIOS.
I've used Gigabyte and Asus/ASRock in the past, and I think a Biostar ... and I'll be going back to those in the future.
MSI is not much better than the other choices when it comes to buggy firmware. Just grep on a z690 how many AE_ALREADY_EXISTS errors you get on a fresh kernel boot ( which indicates at the very minimum a total lack of attention to detail). It did not improve on the z790 despite the fact it was widely reported to them.
MSI? The same thing the author describes happened to me with a new MSI Mortar a while ago. It turns out the BIOS didn't recognize the new Ryzen CPU I bought because it was too new and an update was required. I'd never experienced that before so it took me several hours to realize that and find the "support newer CPU revisions" update buried in one of MSI's support pages. Then I found the flashing process wasn't working and there were no errors to indicate it was really the USB stick I was using that was incompatible. I went through 5 USB sticks before I could flash the thing successfully and have confidence that my brand-new motherboard wasn't a paperweight straight out of the box. Can't imagine what I would have done if I ran out of USB sticks earlier.
Firmware will typically configure the SPI controller such that flash can't be written directly from the OS, both as a self-protection mechanism (there's been malware in the past that deliberately bricked boards by overwriting the firmware) and as a security boundary (if you can modify flash directly, you can replace the set of keys trusted by secure boot).
Both single core and especially multicore in Geekbench and Cinebench R23 are 10-30% faster than the Mac Studio (though using about 100-300% more power (total consumption measured at the wall).
So it'll be nice for times when I need the power, but I definitely won't let this system stay on 24x7.
I updated my x399 board a few weeks ago using the built-in updater from the BIOS screen... all I had to do was plug a network cable in and select online update.
> So I popped everything together, made sure it benchmarked as expected, tested an Intel Arc A750 in preparation for testing on a Raspberry Pi, and then decided the 7900x's egregious power consumption wasn't to my liking the thing idled over 90W, and would eat up over 285W while compiling Linux!
> I wanted to enable AMD's 'Eco' mode, which limits the TDP from the stock 170W to either 105W or 65W. And in my own benchmarking (which happened later on, but I'm including the data here because it's kinda insane), the CPU got about 96% of its 170W performance when I limited it to 105W. And at that level, it idled at 45W and maxed out at 206W. Much better.
He may just have a well binned chip, they have to configure for the lowest performing chip otherwise yields would be down and prices would be through the roof.
Higher idle power consumption is more common on AMD CPUs (I think it's inherent to their design). I don't think intel CPUs consume nearly as much power when idle. Oddly enough, AMD GPUs also have the same problem under certain conditions (multiple monitors, or high refresh rates can result in 100w idle GPU power )
But I agree that the power consumption figures under load are a bit nuts, and intel/amd now seem to be eager to double max power use if it means they can squeeze 5% more performance out of the die.
Which number are you looking at? 18 W idle for a 5900X is very low. 30-40 W is typical for these. Your system is probably not configured correctly, is using slow RAM or you're looking at something other than package power.
GPU idle power seems quite high though, but as I understand most AMD GPUs struggle with power management with more than one monitor connected (no experience here, last AMD GPU was 2012).
18w idle is rather normal for what the ryzenmaster monitoring software will report (which I believe is PPT) . On Intel for PPT you will usually measure 0-5W -- AMD idle power consumption is really bad on desktops.
A 6700xt should idle at 6-7w with only one monitor. If it idles at 28w it means there is something it doesn't like about your monitors that forces it to keep the memory clocks at max. Try setting them at the same resolution, refresh rate, and depth/bpc).
I had a similar situation with two identical monitors except for some reason one of them was a slightly newer revision (same model!) and had an EDID that made it look as an 10bpc panel instead of 8bpc like the older revision. Setting both to 8bpc reduced the idle power consumption by 4x. If you are on windows, you may need some additional tool to be able to change the bpc (the AMD control panel applet works too).
When were we not in an unhinged race? GHz sells chips, gotta get as many as possible; it's just AMD and Intel have gotten really good at pumping the watts through the chips lately.
But you can buy CPUs with lower power limits, sometimes these are released later in the cycle though, or set the limits yourself. Or probably just leave it at defaults and not usually get up to the crazy high numbers anyway; most desktop use cases aren't going to scale usage so high, although there's plenty of examples that do.
Before Ryzen AMD was way behind for many years and Intel didn't have to compete. TDP was lower for a long time. The previous big run achieving speed by pushing power was in the Pentium 4 era if I remember correctly.
Yeah I was going to say, this reminds me of the Pentium 4 days when Intel ran into problems with both maxing out the clock speed and their huge instruction pipeline. I just want a machine that can perform well under pressure, but doesn't double as a space heater while idle...
I’m not sure why I’d care. An extra 100w amounts to a couple cents an hour for most people. Not relevant most of the time, and will shave time off almost any CPU task (as this is increasing the single-threaded speed, which applies to almost everything - sans memory or storage bottlenecks).
Sounds like a worthwhile trade off for most people to me, and more so if you do anything remotely CPU-intensive.
Sure, if you enjoy running a space heater in the summer that doubles as a jet engine regarding the sound scape for 4% more performance then more power to you. ;-)
This is of course, hyperbole. And geez, don't place your PC case with the vents blowing at you. Even at 100% CPU, noise should not be noticeably different if you’re using anything besides the stock cooler. Stock coolers are of course notorious for being bargain-basement tier in noise/performance.
I updated my Z390 ASUS board through the BIOS. I didn’t use the online updater because for some reason it pulled the 2nd newest version. So instead I just downloaded it off the website to a USB stick, booted into the BIOS, picked it, and done.
I once bricked a motherboard with a bad flash, probably around 2012. I ended up desoldering the SPI flash and reprogramming it from another computer. Worked. In hindsight, it was quicker to do that than to go through the "front door" with the manufacturer-provided utilities.
What was extra-cavalier about it was that I didn't even bother disassembling the computer to get the motherboard out of the case. Did it in-situ at awkward angles with the tower in my lap. I did take out the graphics card to give myself some space to work though.
Reminds me when I did something similar back around 2008. The motherboard was made by ECS and came with a "Top-Hat" -- essentially a second flash memory chip that you can snap on top of the board. No desoldering required! More details here: https://www.pcstats.com/articles/1835/3.html
Some motherboard makers (Gigabyte, I think?) would include a socket for the BIOS EEPROM chip, even in relatively mid-range motherboards. I thought that was really cool, but a bit strange. It makes sense on high-end overclocking motherboards where some customers might actually use it and where there's not much of a cost constraint. But it seemed a bit extravagant to spend ~$0.30 on something that almost nobody would ever use. Maybe their service department used it to fix some RMA'd motherboards more quickly.
It looks to have gone out of fashion, I can't find any modern motherboards that have it but I didn't look that hard. Probably because BIOS flashback has gotten more common and because the tall socket would interfere with all the SSD heatsinks they have now.
Maybe it's just that the last motherboard I got (MSI, and ... it was a newegg refurb), but the last PC I built was BY FAR the worse experience ever.
- RAM: only 2 of four slots worked, couldn't install in the "optimal" bank setting, just to get it to boot. Tried with two different stick brands, which various sites said would be compatible
- Ryzen 5600G needed a bios update to ... kinda ... get work. The HDMI does not work at all however
- BIOS update required undocumented help from the internet: need a FAT system on "not too big" of a flash drive.
- OK, so try a new minimalist video card ... doesn't get recognized, but an old video card does work.
- On the linux side, the kernel support for 5600G took another six months, the HDMI-over-displayport-adapter I DID get to work for 4k is now flickering and dropping signal... there are some support notes in arch linux about switching from y color space to others, or some EDID hack, or some thing with fooling DRM. None of this is configurable in text files of course, some systemd binary or direct bin file hacking....
I've never had to do even one of these problems building other PCs (this is probably my seventh or eighth box build in my lifetime).
Refurb often means previously broken and hopefully fixed; sometimes it just means returned, but the box wasn't nice enough to sell as open boxed. Looks like your board probably still had issues after refurbishment. It would probably have been better to return it when you first got it, although newegg return processing isn't always easy.
It's hard to be patient, but buying parts after they've been out long enough to get supported by the software you want to use also helps let other people find the problems with the rest of the package.
I did an AMD APU back in the A8/A10 days and that was the crankiest and most annoying build I did until that point.
It is so appealing to a non-hardcore gamer to get an AMD + graphics integrated chip, and it makes so much sense with all that excess silicon real estate, but the obvious lack of attention in the entire chain of software and hardware and firmware support is unacceptable.
The board (a B550) seemed to have gigantic amounts of support issues from what I can tell, because there were far too many people with similar issues, and they couldn't have been all refurbs. And obviously, a refurb/open box means there are lots of returns.
It was a bad decision on my part, but why is the board accepting an old PCI graphics card but not a "modern" one, even after the most recent BIOS (which is five months old at this point, so it looks like MSI is just abandoning this board from a support perspective, even though the advertised processor support was only enabled in one of the more recent BIOS releases)
I should have built it while I was close to a MicroCenter, but I'm not a person that likes to remount a CPU, so once the CPU is mounted, I'm pretty reticent to swap out a motherboard.
I'll just make it a containers / VM hosting box.
Finally, I just booted up my previous "fastest" box that I upgraded (problematically) to the latest Ubu and wow is it a lot slower than it was. I've had problems with snap update daemons and maybe all the snaps are eating available RAM with all their copies of libraries.
People, serial speed improvement is dead, it is the deadest of the dying "Moore's law" patterns.
The software industry has to start making its code more efficient. We need to loopback and start improving the stacks and execution environments to squeeze more from less. I'll know when we stop putting our heads up our you-know-wheres when OS releases make old machines faster, rather than bogging them down with Javascript Electron monstrosities and memory-is-cheap containers with complete OS's in them to run a web server or web browser.
Linux in particular probably needs to start repackaging libraries into stability-focused libraries. Sweep the libraries for parts that have not changed much in 5-10 years, and start moving those into "mature" libraries, and have "volatile" libraries. But then again the constant churn in GPU compute, vector extensions, etc means that just is not going to stabilize.
We might still be 50 years from where hardware "wringing" is done and the hardware interfaces stabilize for decades rather than years.
I bought my previous motherboard refurb, and found there was a capacitor barely hanging on near the first PCIe slot—someone had probably yanked a GPU and slammed it into that spot (there was also a little damage on the retention clip still).
After a few weeks the 2.5 Gbps Ethernet port stopped working, until I resoldered that capacitor.
Then a few months later it stopped working again, as one of the motherboard traces was also damaged and I couldn't repair it. Had to use a separate network card from that point on.
I didn't return the motherboard because of the hassle... it wasn't the end of the world using a different NIC.
Same thing on my asrock MB, nothing worked until I did the classic usb flashback. Its frustrating really, considering the delicate nature of bios updates.
I had to flash the BIOS on a Gigabyte board to support the Ryzen I bought for it. It was a crazy procedure where the CPU and RAM couldn't be installed, and like others here the only USB drive I could get to work was an ancient 4GB USB 2.0 unit. Mine was an Angry Birds novelty drive, not something I wanted to trust a firmware update to.
> tl;dr: Use an old-fashioned USB 2.0 flash drive, format it FAT32, download the firmware, make sure it's named correctly, and use the motherboard's 'BIOS Flashback' option after powering off the computer.
Also: Ensure the flash drive only has one partition. (Mine was previously a partitioned bootdrive, and formatting leaves the partitions intact!). And for good measure ensure the firmware is the only file, including hidden files. FWIW the manual does specify it must be USB 2.0 and 1+ GB (though for me it worked with a 256MB drive; the firmware was only 32MB).
I also had some trouble with BIOS flashing on my Gigabyte X670 motherboard. I can't remember the specifics but at one point my display output stopped working so I tried to update with BIOS flashback since the update was supposed to improve memory stability. The first attempt failed. The flashback light stopped and I waited several minutes. A second flashback attempt worked. Pretty nerve-racking because they tell you to never ever ever interrupt the flashing process.
The X670 platform was a bit under-baked, in my opinion. There were a lot of strange issues, especially with memory reliability. It took until a BIOS update in January for me to be able to reliably run at DDR5-6000 CL30 with a memory kit that was on the QVL and was marketed as EXPO AMD Compatible. Yes, it's AMD's first iteration with a new socket and DDR5 but meanwhile Intel has been running reliably at DDR5 7200, and AMD still can't do that. Leaves a bad taste in the mouth especially when $250 is now "low-end" for AMD motherboards.
The BIOSRenamer.exe tool included with ASUS BIOSes is only to be run if you need to use USB Flashback feature (USB Flashback expects a certain filename).
I just flashed the BIOS today on my ASUS board by simply unzipping and copying the file to the USB stick, then selecting the ROG-CROSSHAIR-VIII-HERO-ASUS-4402.CAP in EZ Update from the BIOS menu. All went well as usual.
Perhaps with the B650 board EZ Update doesn't like reading a BIOS file with the USB Flashback name?
75 comments
[ 4.6 ms ] story [ 220 ms ] threadThe ineptitude of firmware programmers continues to astound me.
Asus should upload their firmware to the LVFS <https://fwupd.org/> and make firmware updates as simple as 'fwupdmgr update'.
All Asus would need to do is publish updates on LVFS.
I don't know if the Legion line has the same level of support, but you may just be able to ru "fwupdmgr update" and get prompted to reboot for an UEFI/Intel ME/SSD firmware update.
[0] https://robots.org.uk/ToshibaX30X40FirmwareUpdate
The download page actually has release notes which specify when a certain BIOS is required, but you basically have to read back through all of them to figure out which ones to install.
I just assumed that the latest bios was good and the server happily bricked itsefl
Why format an usb stick when there is already a perfectly good fat32 partition lying around?
(Yet another example of the unfairness of the playing field for non-MS OSes).
Rebooting after that does take a while though (minutes) since it needs to re-learn the RAM. And this is a painful process indeed, especially when I originally just had the mobo since I had probably not seated the RAM to the fullest extent (DDR 4 seems to require pressing it very well even after it already clearly clicked in, after pushing it harder it worked well), and the only feedback from the mobo was very similar LED codes as when it "learns" the RAM, and sometimes even half working...
I was originally going to pick up an Asus Strix Z690-A but read about a variety of problems with that board, especially with regards to RAM. Many anecdotes I had seen online had similar complaints about need to really press hard to get the RAM sticks properly installed. I had no issues with the MSI. My previous build had an Asus mobo and it had strange issues, too.
IIRC I also had to try several USB ports in back and something else hacky to get the flash going. It was still half black incantation and performative dance learned from the internet, and about 5% directions provided by MSI on their website or within BIOS.
I've used Gigabyte and Asus/ASRock in the past, and I think a Biostar ... and I'll be going back to those in the future.
(1) https://www.flashrom.org/Flashrom
So it'll be nice for times when I need the power, but I definitely won't let this system stay on 24x7.
> So I popped everything together, made sure it benchmarked as expected, tested an Intel Arc A750 in preparation for testing on a Raspberry Pi, and then decided the 7900x's egregious power consumption wasn't to my liking the thing idled over 90W, and would eat up over 285W while compiling Linux!
> I wanted to enable AMD's 'Eco' mode, which limits the TDP from the stock 170W to either 105W or 65W. And in my own benchmarking (which happened later on, but I'm including the data here because it's kinda insane), the CPU got about 96% of its 170W performance when I limited it to 105W. And at that level, it idled at 45W and maxed out at 206W. Much better.
But I agree that the power consumption figures under load are a bit nuts, and intel/amd now seem to be eager to double max power use if it means they can squeeze 5% more performance out of the die.
With me using my computer, 2 x 27" QHD 144Hz monitors, my AMD GPU is using 28W, while the CPU is using 18W.
(Radeon RX 6700XT 12GB, Ryzen 9 5900X 12-core)
I know the Zen 4 / Ryzen 7000X use a lot more power, as they are designed to pull power and use all available thermal headroom all day!
GPU idle power seems quite high though, but as I understand most AMD GPUs struggle with power management with more than one monitor connected (no experience here, last AMD GPU was 2012).
CPU Power is 15-20W. SOC Power shows 17W.
The RAM is DDR4-3600 CAS16 running at 1800Mhz, which I don't believe is slow.
GPU is "relatively" idly but just having Ryzen Master or Adrenaline open and displaying these values means it is not truly idle.
VRAM clocks are 1990Mhz.
https://www.guru3d.com/articles_pages/amd_ryzen_9_7900_proce...
Both Intel 12900k and AMD 7900 are 78W idle.
So it's basically neck-and-neck on idle power because those CPUs are around the same performance.
Where the power usage is crazy different is multicore load, 320W for the Intel vs 158W for AMD.
edit: actually an AMD 7900X is 299W on MC load. A 7900X is 15% faster on MC than a 7900. So Intel and AMD seem quite similar on power usage.
On that note, the new-ish ATX12VO standard Intel has been pushing apparently allows for motherboards with much lower idle power consumption:
https://en.wikipedia.org/wiki/ATX#ATX12VO
https://hackaday.com/2021/06/07/intels-atx12vo-standard-a-st...
Sounds like a win if it becomes reasonably widely used.
But you can buy CPUs with lower power limits, sometimes these are released later in the cycle though, or set the limits yourself. Or probably just leave it at defaults and not usually get up to the crazy high numbers anyway; most desktop use cases aren't going to scale usage so high, although there's plenty of examples that do.
Sounds like a worthwhile trade off for most people to me, and more so if you do anything remotely CPU-intensive.
Maybe they should just leave a JTAG header somewhere on there.
It looks to have gone out of fashion, I can't find any modern motherboards that have it but I didn't look that hard. Probably because BIOS flashback has gotten more common and because the tall socket would interfere with all the SSD heatsinks they have now.
That's what they'll do if you RMA the board...
I don't think it has a socket, it just has two flash chips soldered to the board.
eg Gigabyte Z690 AORUS TACHYON https://www.gigabyte.com/Motherboard/Z690-AORUS-TACHYON-rev-...
- RAM: only 2 of four slots worked, couldn't install in the "optimal" bank setting, just to get it to boot. Tried with two different stick brands, which various sites said would be compatible
- Ryzen 5600G needed a bios update to ... kinda ... get work. The HDMI does not work at all however
- BIOS update required undocumented help from the internet: need a FAT system on "not too big" of a flash drive.
- OK, so try a new minimalist video card ... doesn't get recognized, but an old video card does work.
- On the linux side, the kernel support for 5600G took another six months, the HDMI-over-displayport-adapter I DID get to work for 4k is now flickering and dropping signal... there are some support notes in arch linux about switching from y color space to others, or some EDID hack, or some thing with fooling DRM. None of this is configurable in text files of course, some systemd binary or direct bin file hacking....
I've never had to do even one of these problems building other PCs (this is probably my seventh or eighth box build in my lifetime).
It's hard to be patient, but buying parts after they've been out long enough to get supported by the software you want to use also helps let other people find the problems with the rest of the package.
It is so appealing to a non-hardcore gamer to get an AMD + graphics integrated chip, and it makes so much sense with all that excess silicon real estate, but the obvious lack of attention in the entire chain of software and hardware and firmware support is unacceptable.
The board (a B550) seemed to have gigantic amounts of support issues from what I can tell, because there were far too many people with similar issues, and they couldn't have been all refurbs. And obviously, a refurb/open box means there are lots of returns.
It was a bad decision on my part, but why is the board accepting an old PCI graphics card but not a "modern" one, even after the most recent BIOS (which is five months old at this point, so it looks like MSI is just abandoning this board from a support perspective, even though the advertised processor support was only enabled in one of the more recent BIOS releases)
I should have built it while I was close to a MicroCenter, but I'm not a person that likes to remount a CPU, so once the CPU is mounted, I'm pretty reticent to swap out a motherboard.
I'll just make it a containers / VM hosting box.
Finally, I just booted up my previous "fastest" box that I upgraded (problematically) to the latest Ubu and wow is it a lot slower than it was. I've had problems with snap update daemons and maybe all the snaps are eating available RAM with all their copies of libraries.
People, serial speed improvement is dead, it is the deadest of the dying "Moore's law" patterns.
The software industry has to start making its code more efficient. We need to loopback and start improving the stacks and execution environments to squeeze more from less. I'll know when we stop putting our heads up our you-know-wheres when OS releases make old machines faster, rather than bogging them down with Javascript Electron monstrosities and memory-is-cheap containers with complete OS's in them to run a web server or web browser.
Linux in particular probably needs to start repackaging libraries into stability-focused libraries. Sweep the libraries for parts that have not changed much in 5-10 years, and start moving those into "mature" libraries, and have "volatile" libraries. But then again the constant churn in GPU compute, vector extensions, etc means that just is not going to stabilize.
We might still be 50 years from where hardware "wringing" is done and the hardware interfaces stabilize for decades rather than years.
After a few weeks the 2.5 Gbps Ethernet port stopped working, until I resoldered that capacitor.
Then a few months later it stopped working again, as one of the motherboard traces was also damaged and I couldn't repair it. Had to use a separate network card from that point on.
I didn't return the motherboard because of the hassle... it wasn't the end of the world using a different NIC.
Also: Ensure the flash drive only has one partition. (Mine was previously a partitioned bootdrive, and formatting leaves the partitions intact!). And for good measure ensure the firmware is the only file, including hidden files. FWIW the manual does specify it must be USB 2.0 and 1+ GB (though for me it worked with a 256MB drive; the firmware was only 32MB).
The X670 platform was a bit under-baked, in my opinion. There were a lot of strange issues, especially with memory reliability. It took until a BIOS update in January for me to be able to reliably run at DDR5-6000 CL30 with a memory kit that was on the QVL and was marketed as EXPO AMD Compatible. Yes, it's AMD's first iteration with a new socket and DDR5 but meanwhile Intel has been running reliably at DDR5 7200, and AMD still can't do that. Leaves a bad taste in the mouth especially when $250 is now "low-end" for AMD motherboards.
I just flashed the BIOS today on my ASUS board by simply unzipping and copying the file to the USB stick, then selecting the ROG-CROSSHAIR-VIII-HERO-ASUS-4402.CAP in EZ Update from the BIOS menu. All went well as usual.
Perhaps with the B650 board EZ Update doesn't like reading a BIOS file with the USB Flashback name?