fit 4 platinum xeon in 1U?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

LukeP

Member
Feb 12, 2017
183
21
18
44
I am limited to 1U and want to fit as much CPU as possible. Are there any 4 Socket motherboards around? I can only find the SM ones that come with a server.

if i were to find a mb, where would i find a compatible psu? The SM motherboards have 7 x 6 pin plugs. Do retail psus come with so many plugs?

note i would fill it with ES or QS plats to save money so I am not interested buying expensive super-servers!

Thanks!
 
Last edited:

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
You mean a quad board ?
There is certainly systems that out 2 x dual boards in 1U. Not sure about a quad process or board that works for 1u...
 

LukeP

Member
Feb 12, 2017
183
21
18
44
2 x dual boards is no good for my use. if they can make 2 x dual you would think they could make 4 1U. its even easier.
 

alex_stief

Well-Known Member
May 31, 2016
884
312
63
38
Quad-socket boards are generally very hard to find on the retail market. For LGA3467, I have not seen any.
I think your only option for now is buying a complete server.
 
  • Like
Reactions: Tha_14

William

Well-Known Member
May 7, 2015
789
252
63
66
I think you will find that the heat output of 4x processors will overwhelm a 1U case.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,518
5,819
113
We are working on a review of the SYS-2049U-TR4.

It is a proprietary motherboard where the PSUs attach directly to the motherboard. It may be possible, but very difficult. The SYS-2049U-TR4 is very cool though!

On the ES/QS CPUs, ES (QL) will likely not work. Maybe QS (QM/QN) will.
 
  • Like
Reactions: William

LukeP

Member
Feb 12, 2017
183
21
18
44
its simple, because where i want to stick it 1U costs $1200/month and 2U costs $2400/month

i cant get by with 2 systems. nano/microseconds matter.

gpus are a problem because of the slow kernel launch latency and pcie transfer latency

so.... big cpus with tight NUMA/QPI links are what i need.
 
Last edited:

LukeP

Member
Feb 12, 2017
183
21
18
44
4 broadwells is also not as good as 2 xeon scalable. the scalable xeons have incredibly beasty FMAs. dual AVX-512

4 times the Zen+, so a pair of epics is also not as good as two scalables.

thats why i want 4 x skylake-x! .... awesome.

i think intel understands the way the world is headed better than amd.
 

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
no, not latency bound... . vectorised maths. ie for machine learning.
Did you see the intel Xeon d-2100 AVX-512 performance even if single FMA, if you could squash a lot of them in (maybe not 1u but the supermicro blades that are verticalnin 3u I think) then odd as it may seem it may make sense if workload is easily distributed which maybe you case appears not)
 

aero

Active Member
Apr 27, 2016
347
86
28
54
If nanoseconds matter then are you sure you want a quad socket box? Qpi bus will jack up your latency. Best to stick to a single numa node for each workset...