Xeon D-1518 1u fan cover

scline

Member
Apr 7, 2016
80
25
18
35
I wanted better airflow in the 1u supermicro server I recently purchased from Wiredzone (link). Created a 3d printed cover to segment one of the 40mm fans directly over the CPU heatsink. Thought I would share for others in the same boat.

This is how it looked stock


This is after adding the printed parts




Next week ill see if it can be a bit cleaner/less bulky, parts are below for those wanting to do the same (^_^)
3D design Xeon D-1500 Fan Cover (part 1 of 2)
3D design Xeon D-1500 Fan Cover (part 2 of 2)
 
  • Like
Reactions: Patrick

Patrick

Administrator
Staff member
Dec 21, 2010
12,399
5,579
113
I wonder if the single mounting bracket puts a fan in the same spot as the dual fan mounting bracket. If so, I need this!

Also, maybe this should go in the new DIY forum: DIY and Makers Spot - thoughts @Patriot
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,399
5,579
113
Are you going to redesign it to include the memory as well? also interesting choice on the M2 850 evo
If you look at a lot of current OEM solutions, they omit substantial cooling for the DIMMs.

Each DDR4 RDIMM is somewhere around 1-3W for reference.

I actually think focusing air on the CPU is a better option here.
 

Deslok

Well-Known Member
Jul 15, 2015
1,122
125
63
32
deslok.dyndns.org
If you look at a lot of current OEM solutions, they omit substantial cooling for the DIMMs.

Each DDR4 RDIMM is somewhere around 1-3W for reference.

I actually think focusing air on the CPU is a better option here.
Fair enough, I just took a peek at the shroud on a CS-24SC before i asked that and that's older(5400 series xeons with ddr2(they're on my list to replace with Xeon-D systems i don't need more compute just more ram in the hyper-v cluster)
 

Evan

Well-Known Member
Jan 6, 2016
3,347
595
113
That's really neat !

I notice that all the HP ddr4 that I have ever had shipped up to 32gb modules (I have none larger than that) not one has had a heat spreader. Compared to ddr3 that was often shipped with them.
The drop to 1.2volt seems to have reduced the power draw and heat significantly.

@Patrick statement of 1-3 watt seems right.
HP used to publish a list of power consumption for memory but don't see it any longer, maybe another vendor or memory maker does.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,399
5,579
113
I think I have a presentation from SK.Hynix on new DDR4.

Most of the manufacturers are also dropping process nodes for DDR4-2400 so I think that lowers power consumption as well.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,399
5,579
113
Nice find Evan. I should revise my budgeting accordingly.

BTW - Maybe I am tired but is it hard to reconcile these two graphs?
upload_2016-4-25_17-8-14.png
 

Evan

Well-Known Member
Jan 6, 2016
3,347
595
113
Yes those values in the 2 graphs seems to differ a bit... Maybe different vendors ram used for different tests or something like that. @Patrick Would you like me to ask HP see if they can answer ?
I also found lots of info from Samsung but talked a lot more about % than absolute values like HP was.

I also wonder what does idle memory mean ? OS booted and just idling ? If so then any concerns about power usage don't seem that great.
Answers my question about how much 128gb (4 x 32gb RDIMM 2Rx4) would using in a Xeon-D setup... About 25watts at load.
 

scline

Member
Apr 7, 2016
80
25
18
35
Finally got home to test the difference this cover does, without the CPU sits about 55C. With the cover in place it drops to 48C - the server is just running ESXi and pfSense (not processing traffic yet) so near-idle.

I did have a makeshift cardboard cover on the CPU and RAM initially but found it only lowered temperature on the CPU by ~2C, the ram does not appear to get super hot like my older DDR3 stuff.

I did try to take the heatsink off the CPU when I first received the server to apply some better paste but it felt stuck on there - wondering if they are using some poxy or thermal tape :(
 
  • Like
Reactions: Patrick

Evan

Well-Known Member
Jan 6, 2016
3,347
595
113
That's a nice reduction, the idea is great, just what the major guys do with servers (look at any recent HP for examples of ducts)
 

ATS

Member
Mar 9, 2015
96
32
18
47
Yes those values in the 2 graphs seems to differ a bit... Maybe different vendors ram used for different tests or something like that.
Could be as simple as one graph being based off of theoretical numbers (Idd etc numbers from data sheets) and another being based off of measurements.

I also wonder what does idle memory mean ? OS booted and just idling ? If so then any concerns about power usage don't seem that great.
Idle is likely just refresh or refresh + CAS whereas load is probably continuous interleaved PRE-RAS-CAS to multiple banks.

The other thing to note that the more dimms per channel the lower the average power per dimm due to the inherent usage restrictions. So going from 1 dimm to 2 dimms per channel doesn't double the power requirement, realistically its closer to 1 dimm Pmax + 1 dimm Pidle than 2x dimm Pmax.

Realistically, if dimm thermal flux is an issue on your system with DDR4, you've got bigger problems.
 

Deslok

Well-Known Member
Jul 15, 2015
1,122
125
63
32
deslok.dyndns.org
That's an impressive temperature drop, the shroud isn't creating a dead space above the ssd thermally and causing it to go up a few degrees is it?(not than an 850 evo should be thermally constrained but a 950 pro can be)