can someone explain to me why server motherboards are so fucking awful (it could also just be supermicro.
-
can someone explain to me why server motherboards are so fucking awful (it could also just be supermicro. i've had two server motherboards and both were from supermicro)
the cheapest, most cost-reduced, aliexpress tier desktop motherboard easily outperforms both of those supermicro boards, combined, when it comes to things like "fucking booting like a normal PC"
@whitequark The one Dell server I operate takes even longer to boot than the Supermicro ones. Oh it’s POST’ing? Time to go and boil a cuppa
-
@whitequark The one Dell server I operate takes even longer to boot than the Supermicro ones. Oh it’s POST’ing? Time to go and boil a cuppa
@schrotthaufen how do u live like this
-
can someone explain to me why server motherboards are so fucking awful (it could also just be supermicro. i've had two server motherboards and both were from supermicro)
the cheapest, most cost-reduced, aliexpress tier desktop motherboard easily outperforms both of those supermicro boards, combined, when it comes to things like "fucking booting like a normal PC"
@whitequark uhh, okay? It won't be anything you don't already know though...
When you power the server on, it will wait for BMC to fully boot first (and as we know, that's a cursed, fully-fledge Linux distro usually booted from SPI flash on ARMv6 core (i.e: ASpeed AST2500)). So if BMC isn't up, x86 firmware will wait for BMC to finish booting and only then start going trough the POST.
Then there's a matter of rubbish from AMI riddled with many, many modules (that often aren't needed), as well as amount of hardware that needs to be initialized (memory controllers, qpi links on multi-socket systems, multiple rootports etc) and of course - DRAM training.
For example, OpenComputeProject's TiogaPass (Xeon SP Lewisburg (Skylake)) with 768GB of DDR4 (registered ECC) takes 1,5 minutes just to train the memory controller when running coreboot. DDR5 can easily take twice as long (my firewall with Intel N100 SoC has a single-stick of DDR5 SO-DIMM and takes 40 seconds to train the memory).
You can avoid going trough the POST if you just kexec a newer kernel but... yeah. -
@whitequark uhh, okay? It won't be anything you don't already know though...
When you power the server on, it will wait for BMC to fully boot first (and as we know, that's a cursed, fully-fledge Linux distro usually booted from SPI flash on ARMv6 core (i.e: ASpeed AST2500)). So if BMC isn't up, x86 firmware will wait for BMC to finish booting and only then start going trough the POST.
Then there's a matter of rubbish from AMI riddled with many, many modules (that often aren't needed), as well as amount of hardware that needs to be initialized (memory controllers, qpi links on multi-socket systems, multiple rootports etc) and of course - DRAM training.
For example, OpenComputeProject's TiogaPass (Xeon SP Lewisburg (Skylake)) with 768GB of DDR4 (registered ECC) takes 1,5 minutes just to train the memory controller when running coreboot. DDR5 can easily take twice as long (my firewall with Intel N100 SoC has a single-stick of DDR5 SO-DIMM and takes 40 seconds to train the memory).
You can avoid going trough the POST if you just kexec a newer kernel but... yeah.@elly i've sat here for ten minutes and the BMC that claims to be alive doesn't respond to arping
-
@grumpybozo long POST times are just DRAM being DRAM. i don't mind that too much. every motherboard is gonna have that problem to some extent
@whitequark @grumpybozo 19 years ago I worked on an IBM eServer that didn't let you turn it on for 5 minutes after connecting power, and then took 15 minutes to get through POST. Was probably the first server with UEFI that I encountered.
-
@whitequark The one Dell server I operate takes even longer to boot than the Supermicro ones. Oh it’s POST’ing? Time to go and boil a cuppa
@schrotthaufen @whitequark With just a little bit of tuning the CPU, RAM and PCI parameters you'll even have time to grow, harvest and roast your own beans for that!
-
@schrotthaufen how do u live like this
@whitequark Easy. Sleep till 9am, work from 10:30am - 5pm (incl. *excellent* dinner), clock out, don’t touch servers in my free time, and get a adequate monetary compensation
-
@whitequark @grumpybozo 19 years ago I worked on an IBM eServer that didn't let you turn it on for 5 minutes after connecting power, and then took 15 minutes to get through POST. Was probably the first server with UEFI that I encountered.
@jernej__s @whitequark @grumpybozo My first AMD EPYC-based servers came with a notice they may take up to 30 minutes to boot the first time. They are a little better now.
-
@schrotthaufen @whitequark With just a little bit of tuning the CPU, RAM and PCI parameters you'll even have time to grow, harvest and roast your own beans for that!
@scherzog @whitequark I could use the excess heat to roast the beans!! 🤯
-
@whitequark Easy. Sleep till 9am, work from 10:30am - 5pm (incl. *excellent* dinner), clock out, don’t touch servers in my free time, and get a adequate monetary compensation
@schrotthaufen i touch servers only in my free time
truly, if hell on earth exists this is it
-
@jernej__s @whitequark @grumpybozo My first AMD EPYC-based servers came with a notice they may take up to 30 minutes to boot the first time. They are a little better now.
@ocdtrekkie @jernej__s @grumpybozo jesus christ
-
@jernej__s @whitequark @grumpybozo My first AMD EPYC-based servers came with a notice they may take up to 30 minutes to boot the first time. They are a little better now.
@ocdtrekkie @whitequark @grumpybozo Some early DDR5 motherboards also came with stickers warning you that the first boot may take 10-20 minutes without showing anything on the screen.
-
at least the tax code is rational and easy to comprehend. comparatively
i like how half the responses to this are like "you haven't seen even half the ways in which these motherboards suck. you are like a little baby" and honestly i'm both impressed and scared
-
@ocdtrekkie @whitequark @grumpybozo Some early DDR5 motherboards also came with stickers warning you that the first boot may take 10-20 minutes without showing anything on the screen.
@jernej__s @whitequark @grumpybozo Yeah, these are early DDR5s.
-
@schrotthaufen i touch servers only in my free time
truly, if hell on earth exists this is it
@whitequark My deepest condolences
-
i like how half the responses to this are like "you haven't seen even half the ways in which these motherboards suck. you are like a little baby" and honestly i'm both impressed and scared
i feel like my next step is, once again, foregoing any remote administration capacity the manufacturer may have allegedly put into the board and just set an IP camera pointing at the screen
-
i feel like my next step is, once again, foregoing any remote administration capacity the manufacturer may have allegedly put into the board and just set an IP camera pointing at the screen
@whitequark Get a JetKVM and use that?
-
i feel like my next step is, once again, foregoing any remote administration capacity the manufacturer may have allegedly put into the board and just set an IP camera pointing at the screen
@whitequark do you happen to have access to a KVM like the nanokvm?
-
@whitequark do you happen to have access to a KVM like the nanokvm?
@cadey no (although it is dawning on me why that specific product category exists)
-
i like how half the responses to this are like "you haven't seen even half the ways in which these motherboards suck. you are like a little baby" and honestly i'm both impressed and scared
@whitequark Assuming the board is new enough and you manage to wake up IPMI, you might find out that using virtual media is locked behind a paywall (OTOH, this apparently only applies to HTML5 – if you use Java, virtual media supposedly works without needing to unlock it).
