I want to build a lightweight server and install CentOS. Does anyone have a recommendation for a suitable motherboard?
Thank you, Joe
19 thoughts on - Mother Board Recommendation
I have a number of Small Form Factor units standing on their sides as servers. I prefer Lenovo these days, HP SFF power supplies have been a problem the past few of those, blowing up after running non-stop for a couple months.
Of course now I am looking at ARM boards for servers. Redsleeve (and Fedora) for now but soon CentOS! I perfer the Cubieboard (2 and truck)
as ARMv7 units. Not hard to put together your own server with them. THe lower power draw and heat makes them so much better.
What will the role of the server be? How “lightweight”? How many users, what kinds of services, what (if any) performance requirements, etc? Room for future growth/expansion?
Budget?
Joseph Hesse wrote:
Not sure what you mean by lightweight, but I’ve been happy with Asus and Gigabyte motherboards. STAY AWAY from Supermicro: we have a bunch of servers with 64 cores – so we’re talking high end – which we bought from Penguin, and we’ve had a *bunch* m/b problems, as in, we’ve sent back a number of them for repairs, where they replaced the m/b. So, if high-end servers are like that, I do *not* trust their quality control.
mark, hoping nixpam will let this through
If it’s just for play… a refurbished Lenovo Core 2 Duo Intel V-Pro PC
from Microcenter is about $120. And includes a copy of Windows 7 Pro.
It’s my preferred way to buy Windows 7 and get a free PC. And I keep one around just for parts.
Thank you for your reply.
I am currently using, as a server, a workstation computer. It was built with a gigabyte motherboard and has 8G ram and a 1000G Sata 3 drive.
It is accessed with SSH and runs an ftp server, a web server and a samba server so my wife can back up her pc to it. It is running behind a router.
This computer more than meets my needs as a lightweight server except the hardware is dying and I want to replace it.
The question in my mind is: should I just buy another workstation class motherboard and duplicate what I already have or buy a motherboard which is intended to be used as a server?
Thank you, Joe
I won’t run a server that doesn’t have ECC main memory. most desktop hardware doesn’t support ECC.
at home, Im using a HP MicroServer N40L
for my public internet server (at a coloc site), I just picked up a HP
DL180 G6 with 24gb ram, and 2 x Xeon X5650 cpus (6 cores each at
2.66ghz) and SAS2 raid with 14 3.5″ SAS/SATA bays for $740 from ebay.
At work, we paid $6500 each for a similarly configured machines about 3
years ago. this is a 2U rackmount server, and far too noisy for anywhere but in a rack in a data center…
Joseph Hesse wrote:
You *really* don’t need a real server. Note that my idea of a “real”
server is rackmount, has two or four physical CPUs with anywhere from 8
(used to be 4) to 16 cores each, and 8G? Everything here has at least 64G
(used to be 32G, though there were some 10-yr-old compute nodes with less). The workstation you have – do you find it under a heavy load? Do you or your wife feel that it’s not responsive enough? If no, then just replace what you have. I rebuilt my home system early this year, with a Gigabyte m/b, 8G RAM. Got a little crazed, so it’s got 2 1TB h/ds as Linux software RAID 1, and for $30 I bought a hot-swap drive bay to fit in the case, and a 2TB drive to put in there for offline backups, like I do at work….
mark “really do gotta set up samba for my wife & the kid”
John R Pierce wrote:
Oh, right, *all* the servers here use ECC DIMMs. And you really, REALLY
don’t want to go there: a) price, b) n/s is not buffered is not registered, none of the above compatible in the same bank, and oh, yes, dual rank is *not* compatible with single rank or quad rank… I kid you not. I’ve had servers simply not boot by mixing two of those, and let’s not forget not fitting in the slot…., and c) see a).
mark, semi-guru on DIMMs….
ECC is such a horrible pain in the rear. If you don’t have things like
“SLA” in your casual vocabulary, pretty much any desktop board works find for CentOS6. For spare/personal/backups servers, I use whatever old hardware sits in the junk room.
Anything using ECC is such a pain to match up correctly that I tend to buy motherboard/RAM/CPU from a vendor as a package unit so it’s warranted to work together. Registered/Unregistered, CAS timing, single/double/quad ranked, never mind voltages, and making sure your CPU
supports it!
For all the promises of better uptimes, I’ve had far more trouble with mis-matched ECC than I’ve ever experienced in bad non-ECC RAM. Truly, this is a sorry showing for ECC.
Ben
As one other data point, I’ve had a few Supermicro motherboards, and have only had one bad one so far. (I’m less enamored of their IPMI
implementation.)
–keith
seriously? where did you EVER get this from?
any server thats storing data which is even remotely important, not having ECC means that soft bit errors go undetected, leading to VERY
hard to detect data corruption. and the more memory you have, the higher the frequency of soft bit errors.
sure, you need to get the correct spec memory, you can’t mix and match random parts…. no different than anything else, from spark plugs or tires for your car to clothes . go ahead, try and put volvo 940
spark plugs in your mercedes 300te, see how well /that/ works. and I
really can’t get into my son’s “M” t-shirts anymore, being that i’m kinda XL.
Am 03.06.2014 um 20:26 schrieb Lists :
It
All of those specs are listed in the motherboard manual. If you’re buying your RAM from a reseller that doesn’t give you the corresponding specs to match up against the mobo specs, stop buying from that vendor.
There are vendors that will match up your specific motherboard with the RAM that works in it, and will exchange the RAM for the right stuff if by some tiny chance they specified the wrong stuff. (e.g. Crucial)
At home I have two of these atom C2750 boards each with 32GB of ECC ram . They are extremely good, powerful and very cheap boards. They use very little power so you don’t need fans which is nice. And they have ipmi controllers.
The processors are 8 core and will easily outstrip a 5i or 7i.
Perhaps, perhaps not. Remember the old saw about simplicity and reliability? ECC ignores that saw completely, resulting in a complex, error prone hardware landscape fraught with incompatibilities.
With desktop systems, it’s pretty straight forward to get memory that
“just works”. Yes, with ECC, all the specs are in the Motherboard manual, but there are far more variables to consider. And, even when you buy exactly to spec, I’ve had issues with systems that worked fine for days at high loads and passed extensive Memtest, but locked up sporadically (every few weeks), at a tremendous cost to diagnose. The only way I’ve found to be confident that a purchase will work is to buy only hardware on a specific hardware compatibility list, and typically I
buy the hardware together. Most of our hardware is in this category.
I’ll consider my irreverent point of view, that this complexity is unnecessary and counter-productive, as somewhat unpopular and leave it at that.
-Ben
PS: As an aside, we’ve had extremely good results with SuperMicro – we shop for them almost exclusively. The only problems we’ve had was with onboard SAS controller on a specific model on EL6 – worked fine on EL5. We ended up disabling SAS and going with onboard SATA with no further issues.
Warren Young wrote:
Buried in some of them, and others, well, it tells you what it will take… and it *assumes* that you’re just building the system, and buying all the DIMMs as one batch, *not* that you’re replacing a failed DIMM. But you’ve got to match even things like cl2whatever. If it doesn’t have
*exactly* what’s on the other DIMMs, it won’t work.
This may be industry standard, and I understand that. I just think that it’s a poor showing that this is what you have to do to take advantage of technology meant to be more reliable than “consumer grade” stuff, which somehow manages to be quite reliable even if you mix and match.
So, the “better” one is, in practice, less reliable unless you follow extremely narrow guidelines, and not only pay more per unit, but buy more units than needed! Usually, the cheap one is flaky and unreliable, and the expensive one “just works”. See: shovels, auto parts, shoes, kitchen knives, mechanic tools, bread machines, and virtually every other product category on the market.
My complaining on an obscure list won’t change anything, but this is still a sorry state for ECC.
Ben
consumer non-ECC memory will just pass occasional bad bits under those same circumstances of mismatched timing or voltages or whatver.
I do suppose noone remembers the SDRAM (pre-DDR) where AMD systems would support 256MB single rank DIMMs but Intel systems only supported 128MB
single rank and the rather rare 256MB dual rank (oft called double sided because they usually were) but often had problems if you used 4 of these.
in fact, my previous Intel Core2Duo desktop was unreliable with 4 dimms, yet rock solid with 2, and this was regardless of the timing. 4 dimms would boot, pass memtest, etc, but the system would randomly lockup with no clues. ECC would have thrown a fault the instant there was a bad bit.
dynamic memory is much more complex than it superficially appears.
19 thoughts on - Mother Board Recommendation
I have a number of Small Form Factor units standing on their sides as servers. I prefer Lenovo these days, HP SFF power supplies have been a problem the past few of those, blowing up after running non-stop for a couple months.
Of course now I am looking at ARM boards for servers. Redsleeve (and Fedora) for now but soon CentOS! I perfer the Cubieboard (2 and truck)
as ARMv7 units. Not hard to put together your own server with them. THe lower power draw and heat makes them so much better.
What will the role of the server be? How “lightweight”? How many users, what kinds of services, what (if any) performance requirements, etc? Room for future growth/expansion?
Budget?
Joseph Hesse wrote:
Not sure what you mean by lightweight, but I’ve been happy with Asus and Gigabyte motherboards. STAY AWAY from Supermicro: we have a bunch of servers with 64 cores – so we’re talking high end – which we bought from Penguin, and we’ve had a *bunch* m/b problems, as in, we’ve sent back a number of them for repairs, where they replaced the m/b. So, if high-end servers are like that, I do *not* trust their quality control.
mark, hoping nixpam will let this through
If it’s just for play… a refurbished Lenovo Core 2 Duo Intel V-Pro PC
from Microcenter is about $120. And includes a copy of Windows 7 Pro.
It’s my preferred way to buy Windows 7 and get a free PC. And I keep one around just for parts.
Thank you for your reply.
I am currently using, as a server, a workstation computer. It was built with a gigabyte motherboard and has 8G ram and a 1000G Sata 3 drive.
It is accessed with SSH and runs an ftp server, a web server and a samba server so my wife can back up her pc to it. It is running behind a router.
This computer more than meets my needs as a lightweight server except the hardware is dying and I want to replace it.
The question in my mind is: should I just buy another workstation class motherboard and duplicate what I already have or buy a motherboard which is intended to be used as a server?
Thank you, Joe
I won’t run a server that doesn’t have ECC main memory. most desktop hardware doesn’t support ECC.
at home, Im using a HP MicroServer N40L
for my public internet server (at a coloc site), I just picked up a HP
DL180 G6 with 24gb ram, and 2 x Xeon X5650 cpus (6 cores each at
2.66ghz) and SAS2 raid with 14 3.5″ SAS/SATA bays for $740 from ebay.
At work, we paid $6500 each for a similarly configured machines about 3
years ago. this is a 2U rackmount server, and far too noisy for anywhere but in a rack in a data center…
Joseph Hesse wrote:
You *really* don’t need a real server. Note that my idea of a “real”
server is rackmount, has two or four physical CPUs with anywhere from 8
(used to be 4) to 16 cores each, and 8G? Everything here has at least 64G
(used to be 32G, though there were some 10-yr-old compute nodes with less). The workstation you have – do you find it under a heavy load? Do you or your wife feel that it’s not responsive enough? If no, then just replace what you have. I rebuilt my home system early this year, with a Gigabyte m/b, 8G RAM. Got a little crazed, so it’s got 2 1TB h/ds as Linux software RAID 1, and for $30 I bought a hot-swap drive bay to fit in the case, and a 2TB drive to put in there for offline backups, like I do at work….
mark “really do gotta set up samba for my wife & the kid”
John R Pierce wrote:
Oh, right, *all* the servers here use ECC DIMMs. And you really, REALLY
don’t want to go there: a) price, b) n/s is not buffered is not registered, none of the above compatible in the same bank, and oh, yes, dual rank is *not* compatible with single rank or quad rank… I kid you not. I’ve had servers simply not boot by mixing two of those, and let’s not forget not fitting in the slot…., and c) see a).
mark, semi-guru on DIMMs….
ECC is such a horrible pain in the rear. If you don’t have things like
“SLA” in your casual vocabulary, pretty much any desktop board works find for CentOS6. For spare/personal/backups servers, I use whatever old hardware sits in the junk room.
Anything using ECC is such a pain to match up correctly that I tend to buy motherboard/RAM/CPU from a vendor as a package unit so it’s warranted to work together. Registered/Unregistered, CAS timing, single/double/quad ranked, never mind voltages, and making sure your CPU
supports it!
For all the promises of better uptimes, I’ve had far more trouble with mis-matched ECC than I’ve ever experienced in bad non-ECC RAM. Truly, this is a sorry showing for ECC.
Ben
As one other data point, I’ve had a few Supermicro motherboards, and have only had one bad one so far. (I’m less enamored of their IPMI
implementation.)
–keith
seriously? where did you EVER get this from?
any server thats storing data which is even remotely important, not having ECC means that soft bit errors go undetected, leading to VERY
hard to detect data corruption. and the more memory you have, the higher the frequency of soft bit errors.
sure, you need to get the correct spec memory, you can’t mix and match random parts…. no different than anything else, from spark plugs or tires for your car to clothes . go ahead, try and put volvo 940
spark plugs in your mercedes 300te, see how well /that/ works. and I
really can’t get into my son’s “M” t-shirts anymore, being that i’m kinda XL.
Am 03.06.2014 um 20:26 schrieb Lists:
It
All of those specs are listed in the motherboard manual. If you’re buying your RAM from a reseller that doesn’t give you the corresponding specs to match up against the mobo specs, stop buying from that vendor.
There are vendors that will match up your specific motherboard with the RAM that works in it, and will exchange the RAM for the right stuff if by some tiny chance they specified the wrong stuff. (e.g. Crucial)
At home I have two of these atom C2750 boards each with 32GB of ECC ram . They are extremely good, powerful and very cheap boards. They use very little power so you don’t need fans which is nice. And they have ipmi controllers.
The processors are 8 core and will easily outstrip a 5i or 7i.
http://www.asrockrack.com/general/productdetail.asp?Model
Perhaps, perhaps not. Remember the old saw about simplicity and reliability? ECC ignores that saw completely, resulting in a complex, error prone hardware landscape fraught with incompatibilities.
With desktop systems, it’s pretty straight forward to get memory that
“just works”. Yes, with ECC, all the specs are in the Motherboard manual, but there are far more variables to consider. And, even when you buy exactly to spec, I’ve had issues with systems that worked fine for days at high loads and passed extensive Memtest, but locked up sporadically (every few weeks), at a tremendous cost to diagnose. The only way I’ve found to be confident that a purchase will work is to buy only hardware on a specific hardware compatibility list, and typically I
buy the hardware together. Most of our hardware is in this category.
I’ll consider my irreverent point of view, that this complexity is unnecessary and counter-productive, as somewhat unpopular and leave it at that.
-Ben
PS: As an aside, we’ve had extremely good results with SuperMicro – we shop for them almost exclusively. The only problems we’ve had was with onboard SAS controller on a specific model on EL6 – worked fine on EL5. We ended up disabling SAS and going with onboard SATA with no further issues.
Warren Young wrote:
Buried in some of them, and others, well, it tells you what it will take… and it *assumes* that you’re just building the system, and buying all the DIMMs as one batch, *not* that you’re replacing a failed DIMM. But you’ve got to match even things like cl2whatever. If it doesn’t have
*exactly* what’s on the other DIMMs, it won’t work.
mark
Am 03.06.2014 um 22:39 schrieb m.roth@5-cent.us:
That
This may be industry standard, and I understand that. I just think that it’s a poor showing that this is what you have to do to take advantage of technology meant to be more reliable than “consumer grade” stuff, which somehow manages to be quite reliable even if you mix and match.
So, the “better” one is, in practice, less reliable unless you follow extremely narrow guidelines, and not only pay more per unit, but buy more units than needed! Usually, the cheap one is flaky and unreliable, and the expensive one “just works”. See: shovels, auto parts, shoes, kitchen knives, mechanic tools, bread machines, and virtually every other product category on the market.
My complaining on an obscure list won’t change anything, but this is still a sorry state for ECC.
Ben
consumer non-ECC memory will just pass occasional bad bits under those same circumstances of mismatched timing or voltages or whatver.
I do suppose noone remembers the SDRAM (pre-DDR) where AMD systems would support 256MB single rank DIMMs but Intel systems only supported 128MB
single rank and the rather rare 256MB dual rank (oft called double sided because they usually were) but often had problems if you used 4 of these.
in fact, my previous Intel Core2Duo desktop was unreliable with 4 dimms, yet rock solid with 2, and this was regardless of the timing. 4 dimms would boot, pass memtest, etc, but the system would randomly lockup with no clues. ECC would have thrown a fault the instant there was a bad bit.
dynamic memory is much more complex than it superficially appears.