After some research it seems AMD has still some issues with encode quality, idle power consumption and possibly drivers and support in Unraid, with the stock kernel.
Thinking about going with a combination of Intel Xeon E-2468 and Intel Arc A380 instead.
Encode performance has been a mixed bag on AMD, you need to crank up the bit rate to get on par vs QuickSync esp for a TV DVR box as sports ball leads to heavy motion blur.
Also nice that error injection works, that makes it easier to practically test if ECC works correctly, on AM4 I never saw this feature working. On AM4 the most reliable way to check if ECC is functioning correctly is overclocking the ECC memory to its limit without increasing the memory voltage to force memory errors which then get corrected and this action is subsequently logged in the memory testing program.
Stuff that also just works after swapping the 8700G out for a 7800X3D:
IOMMU Options
PCIe Bifurcation Options
Still sucks that on AM5 ASUS seems to be the only reliable option left regarding ECC memory support, ASRock manufacturs many of its motherboards with the necessary additional traces for the memory modules but then messes up the BIOSes regularly
On AM4 ASRock, Gigabyte as well as ASUS made motherboards with ECC generally working, only MSI was the negative exception.
As mentioned earlier the 8700G gets returned, feel a bit dirty… but on the motherboard I have which functions completely fine including prime95 with AVX-512 with the 7800X3D it is actually unusable in Windows with the 8700G, not even thinking about ECC, PCIe Bifurcation or IOMMU options.
Hopefullly by the time the PRO 8750G or whatever it is going to be called is going to be released, BIOS and driver updates will have improved the situation.
Tested AM4 APUs pretty early after their initial releases (3200G, 4750G and 5750G) and the situation there was never this bad.
Since the return window closed I am now again without the 8700G and will revisit this thread as soon as I get my hands on a PRO 8700G/8750G.
If anyone has an 8700G feel free to post your experiences with newer BIOS versions and your motherboard of choice here, I couldn’t test the latest BIOS 1905 from ASUS that disabled the on a desktop useless STAPM feature, maybe there are already other bug fixes in that release that aren’t explicitly mentioned.
Thank you for doing this. I initially am going to get a consumer 8700G without ECC since I’m not running my AM5 system as a server at first, but I am planning to move to the Pro 8700G with ECC down the line when the system does become a VFIO and Server machine.
Not a bad plan, but seems to me a 7700 is an even better plan. If the APU chip on the 7700 is too basic you can always get a $99 Intel Arc A310 or an Arc A380 or RX 570 for cheap.
There is very little reason to buy an 8700G for servers especially since they do not support ECC.
I would like a better iGPU in case I want to run my hypervisor in a graphical desktop environment. Hence 8700G.
Also I do not have another slot to use because it’s a mATX board. It has to have a good iGPU. The x4 slot will be using a Mellanox card for 10G copper Ethernet via SFP+.
Do you have Asrock boards on hand? I’m running a B650M PG Riptide and would like to know if it works or not. If you could get that exact board it would help my mATX Node 804 build a ton.
If no dice, I’ll just run the non Pro 8700G and trust ZFS to deal with errors.
Unfortunately don’t have ASRock AM5 motherboards, only a single ASUS ProArt X670E-CREATOR WIFI. The next planned motherboard purchase is currently set around Zen 5 and the new X870E etc. motherboard generation, I hope these will be better than the first-gen AM5 motherboards regarding board layout and features.
ASRock lost my trust with AM5 when they at first removed ECC support from their spec pages (it then returned).
Additionially the mentioned ASUS ProArt X670E motherboard is to this day the only AM5 motherboard that ticks all my requirement boxes.
Do you already have your AM5 ASRock board running? You mentioned “getting” a 8700G, does that mean you are currently running a Ryzen 7000 CPU?
With that ECC memory testing to check if the BIOS isn’t messed up regarding ECC should be pretty easy.
I would prefer to see how much they’ve progressed on the front before risking a Ryzen Pro purchase. mATX is a safer bet because that form factor is also shared with the Asrock Rack boards.
I currently have no CPU or RAM installed because I’m in a wait and see pattern. I bought my mobo because it’s the only mATX board with decent layer count and PCI-E slot layout. If I were to wait for B850, they must put out a 8 layer PCB mATX board with DDR5-7200 support (for Hynix A-die) and the same excellent PCI-E layout for me to consider replacing this board. Gigabyte lost my trust when they nuked so many PCI-E slots.
And unfortunately, I don’t have to look far to know my trust with ASUS is declining.
I’m likely not going to look at a Strix Point APU because combining Zen5 and Zen5c is going to add too much uncertainty to my core pinning plans for VFIO.
Not as interested in having both Zen5 and Zen5c cores together, cause there’s a risk it might have that E-core latency penalty. Even less interested if it’s two separate dies.
Also I looked at the ProArt X670E. Two things are dealbreakers: No Optical SPDIF out, and only a x2 bottom slot. I have many capture cards that need x4 bandwidth so this is a instant dealbreaker.
All I need to know is if my B650M board has the traces to support registered DIMMs.
Many here are saying Asrock does not support ECC. But they do, just not with their consumer mobos. I have tested Asrock rack b650d4u and it supports ECC with Ryzen 7000 series. Now ordered 8700G PRO and will test it with Kingston server premier 5600 mhz, should work. Asrock has multiple am5 server motherboards, very interesting ones with x300 chipsets also. I can recommend. Also I have Asus B650E-i and runs fine with ECC but lack server features like IPMI which is a must, and which Asrock rack boards have.
Since to my knowledge there hasn’t been any public AGESA version where ECC works with the non-PRO 8700G does it matter if that limitation is baked into the retail silicon or if it’s a blacklist in the AGESA, checking what features a specific SKU is allowed to use?
But my gut says that for debugging purposes during hardware development alone such “soft features” are to be controlled by software, not hardware.
The common method for disabling features is some of both, as it’s blowing a disable fuse on the die. APU ECC’s possibly just an AGESA thing but usually fuses there to prevent people from upgrading their hardware with a software mod. It’s also possible that binning means dies that ship with a feature disabled don’t have it enabled because it didn’t yield. The PRO APU features don’t seem like they’d take much area, though (hence the economics of fusing off), suggesting defects probably aren’t a large cause of disables.
About the only recent-ish exception I can think of is early Alder Lake where Intel hadn’t yet started fusing off p-cores’ AVX-512 and the option to enable it hadn’t yet been removed from all the BIOSes.