Darn, that’s sad to hear. Tools that can’t be trusted with their fundamental functions out of the box are not worth much debug time if better hardware exists…
If it were a few tweaks to the BIOS or kernel command line, no big deal. Been there, done that, if I understand the fix, I’m fine rolling with it.
But despite trying all the options I found, it still wasn’t working - and it hit a “I can’t trust this, even it it runs for a week…” level. I trust hardware until it gives me suitable reasons not to, and this counts.
So back it goes. I guess I’ll go with Intel. 
I found a similar SuperMicro server that can run a Xeon, fewer cores (6C/12T vs 12C/24T), less cache, but… insert handwaves here. SuperMicro and Intel are both names I expect to work decently as servers together.
And, as a bonus, I can skip the transcode GPU, if I want, because I can get one with onboard video hardware for QuickSync.
Costs are roughly the same.
Eh. Phone wins where online fails. Server chassis and CPU are going back for a refund.
That’s insane - I can understand hardware not working with Linux if it’s desktop - but server-grade stuff is all Linux all the time.
Something must have been defective; not sure if it’s worth trying another one.
How much of it is Ryzen, though? The AMD server stuff is mostly EPYC.
This is apparently a super common issue, and nobody understands all the details, near as I can tell.
https://www.google.com/search?q=ryzen+linux+c6&oq=ryzen+linux+c6
I’ll pull the box apart tonight, pack stuff up, and send it out tomorrow. Just, going to be a few weeks until I can even order new stuff, because I need to use the gift card credits I get for the new order.
Anyone want a barely used Quadro P2000?
Other advice: If you’re happy with Intel, consider just buying a used server. They’re cheap…
I’ll probably add used servers if I need more power in the future.
Used is great especially if you’re paying a flat rate for power (or are replacing space heaters anyway 
).
I’m surprised at the Rizen thing - almost seems like one of those “differentiation” bugs companies like to leave in to keep people buying server chips.
I’m pretty sure it’s not intentional… just is a thing.
Not fixing it after years makes it a thing.
Excellent. Refunds provided. Now to verify the Intel option and order it.
Also, I should probably poke at the 1U boxes that ended up in the back of my car from someone who was moving. I’ve no idea at all what’s in them…
Intel bits ordered.
RAM and drives and such will be reusable.
Anyone want a barely-powered-on Quadro P2000? 
Good looking box. Hopefully this one doesn’t have any surprises.
But I suppose the pressing question is- will it run blogger?
1 Like
Chiming in to provide some other info, I have exactly the same mobo, 1700X and 2x16GB UDIMM ECC 3200 RAM, and after Syonyk pointed out the Ryzen C-State issue, I put in the following mitigations
- Power Supply Idle Current to Typical
- Disabled C State C6/C7, left the rest enabled
This is for FreeBSD 12.2 latest, and after 12 days now (I’m going for a full 14 days to say it’s solved) things are looking real good. Really not sure why Syonyk had continuous issues while mine doesn’t, no clue. It’s a really, really weird issue. Seriously, I hope AMD fixes this, it really shouldn’t be happening for *nix at all.
It shouldn’t. SuperMicro makes good stuff, as does Intel. ASRock and AMD… are both more prone to issues, though I was hoping that legacy knowledge from 2 decades ago didn’t still apply. Apparently it does.
I keep meaning to look at web hosted editing options for Jekyll or such, to provide some limited blogging-as-a-service hosting.
Well, it’s not been 14 full days uptime, looks like, at least for me, the UEFI mitigations have fixed the halting issue! Yippee!
Excellent!
Most of my new parts arrived today, so that’ll be fun to mess with soon.
Welp. I remain a dumbshit.
This particular board doesn’t support the onboard video, meaning no QuickSync despite a Xeon that supports it. I hadn’t realized (because I’m clearly too stupid to spec a modern computer) that not only does the CPU have to have an integrated GPU, but the board has to also support that in order for the GPU to be usable for internal video transcode work.
Crystal clear once you know exactly what to look for.
X11SCA-F | Motherboards | Products | Supermicro (not my board)
Graphics
ASPEED AST2500 BMC
Intel UHD P630 graphic(For Xeon E-21xxG/E-22xxG series only)
X11SCM-F | Motherboards | Products | Supermicro (my board)
Graphics
ASPEED AST2500 BMC
They both have the same chipset, but apparently without actual output hardware, the GPU doesn’t come online and isn’t accessible. Crystal clear, apparently, to everyone but me.
I now get to decide if I care about a boot mirror or hardware transcode more, because I only have space for one.
//EDIT: Poking around the board, it’s got dual NVMe onboard, I just couldn’t find one of them under some airflow routing shields. So I can shove dual NVMe and the Quadro in. Whatever…
That almost seems something you could force override in BIOS or via a kernel hack - trick the chipset into thinking a VGA port is available.
BIOS, maybe. Kernel hack, probably not.
And I’ve seen some that implies it’s actually a hardware difference - that pins to power the GPU simply aren’t powered on mainboards that don’t have GPU support.
So weird - I wonder if there’s a license cost associated with powering those pins.