The chassis is the new rack, at least for customers who want cheap server nodes and none of that fancy schmancy management software that comes in blade servers, but who do want the density and shared power. So Super Micro, an early maker of half-width motherboards that allowed servers to be double- and quadruple-stuffed with …
Only if you are deaf.
This is Supermicro, remember? Every rig of theirs I have used over the years had fixed revs fans at 6k+ using brute force instead of proper airflow design. If something gets hot - put another spot fan there. Credit where credit due - the approach is cheap, cheerful and works. It does however produce the sound of a Concorde on take off. I would be seriously surprised if this rig is any different.
You should crack the bubbly if it is at around ~70db. It will probably be more. I would not want to work in a datacenter full of these.
I don't know about you, but I don't work in a datacenter. The datacenter is generally a separate room/building/area of the city (delete as per appropriate to client) from where I am located. I only have to go into the datacenter when a widget is broken. Something that thankfully doesn't happen all that often.
When it does, there is some wailing banshee alarm of ultimate screaming high-pitched doom making its presence felt in a manner that far surpasses the irritation of any Supermicro case I have ever encountered.
Give me my spinny fans of high-RPM victory. They prevent the drives from dying and prolong the periods of silence in between having to enter the accursed room of auditory assault alarms.
I do not work in a datacenter either (nowdays).
A variable RPM fan which is spinning only as fast as needed in a well designed case is likely to have half the failures (or even less) of something that is spun up to hypersonic speeds over its design period. Even if you have all of your nodes cranked 100% you still get uneven cooling within a rack, hot/cold parts of the datacenter, "capricious" racks and isles - you name it. So variable speed can take car of that and drop your failure rate (and increase MTBF) even in the fully loaded case. Doubly so if you have hot standby nodes and some form of routine duty cycle/replacement.
Similarly, I would rather have something where the designer has gone around the _WHOLE_ motherboard looking for hotspots doing the thermals instead of slapping the biggest fan possible and ignoring 70C+ in some nooks and crannies.
So yeah, viva la high-RPM victory (going hypersonic towards crash and burn is always fun).
I can't speak for these new machines, but the previous generation of Supermicro twin servers are much much quieter than the first few generations. We first bought a bunch when they originally came out (was around the Intel 5000 chipset time I think), and have bought a few more as time has gone by. All of these machines were insanely loud. Our most recent purchase was a 2u twin squared of the outgoing Xeon architecture, so that made for 48/96 cpu cores/hyperthreaded cores for the unit. It is a hell of a lot quieter. In fact you almost work with it sitting next to you.
We have 76 of the old SGI XE320 1RU dual-node Xeon 5000 Supermicro based compute nodes. SGI shipped them with 4 * Hitachi Deskstar 3.5" SATA drives in the front of each node. Originally those drives benchmarked as having poor (and very variable) sustained read and write rates, and there was some handwaving of our concerns off as "what did you expect from SATA?"
Eventually they fitted some extra rubber mounts around the fans, but it was when we got the OK to drop the BIOS fan profile from the "flat out" default setting to the highest variable speed fan profile that the disk drive transfer graphs stopped looking so tragic.
Even at the lower fan speeds it's still necessary to use earmuffs when working near that cluster... can't say I really like 40mm fans.
Supermicro do a great job of looking for thermal discrepancies in their boards.
As to “high speed versus variable speed” that’s an open question. You are saying that you trust $vendor to know the when and where of kicking up the fan speed? My experience with every Dell laptop ever (and a fair number of tier 1 servers) would seem to disagree with your assessment.
Fans are consumables. I’d rather replace the $0.10 fan periodically as part of regular maintenance on my $10,000 server than wreck my $10,000 server trying to save my $0.10 fan.
But that’s just me.
40mm fans. Dear god; ASUS 1U pizzabox nightmares. I have a set that have, what...12 of the blighters? Give it wings it'll take off.
That said, 6 years in, the fans are still going...and so are the servers...
4 two-socket servers in 2U. Already on the market, legendary global support, greater power efficiency, just as dense, no loss of density when opting for higher capacity drives:
That's what I was thinking too. We have an M1000e populated with 4 half's and 12 quarter's (8 x 6 core & 24 x 4 core) with plenty spare... It's (relatively) quiet and has DRAC... 10U but best of both worlds... Best piece of kit I've ever had the privilege of playing with...
Biting the hand that feeds IT © 1998–2017