AMD EPYC 7302p+ Supermicro H11SSL-i version 2

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

penk38

New Member
Feb 16, 2023
4
0
1
Hello. I just ordered H11ssl through tugm4470 - great experience, super fast responses and shipment.

I am curious what kind of PSUs you guys use. I'd prefer cheaper and more silent options obviously - are desktop PSUs enough? Right now my system needs about 600W, if I think ahead for upgrades (1yr+ but who knows), I should still fit into 1000W (2 GPUs, some disks).

Do the cheaper desktop PSUs have all the needed connectors/outlets for multiple GPUs etc? I don't really want to go server - don't need redundancy and prefer quieter setup.

EDIT: Oh and it needs to fit into an ATX case, of course. Thanks for reading, it is my first server build, so any help much appreciated.
 

RolloZ170

Well-Known Member
Apr 24, 2016
5,142
1,546
113
Do the cheaper desktop PSUs have all the needed connectors/outlets for multiple GPUs etc?
no, they are cheap and two EPS12V + X PCIE connectors are expensive.
some very cheap PSU have weak +5VSB power, remember you need some more Amps for the BMC.
 

lopgok

Active Member
Aug 14, 2017
215
148
43
I use desktop power supplies. They have one 8 pin EPS power. I buy a PCI-E to EPS power adapter, as I don't use high power graphics cards, and desktop power supplies have one or more PCI-E power cables.
 

penk38

New Member
Feb 16, 2023
4
0
1
2u and 4u server chassis, with 800-2000w psu ;)
Nice. Are thos supermicro branded PSUs? Would you be kind enough to give me product codes so I can check? Even if they don't fit my case, maybe it will nudge me in the right direction and I will find something similar suitable :)

some very cheap PSU have weak +5VSB power, remember you need some more Amps for the BMC.
Oh, I will admit I know very little about that - but I decided to build a server partly as a learning experience/challenge after all. Thank you though.

Do you suggest to completely avoid desktop PSUs? I've seen some youtube (commercial) build tutorials using for example Corsair HX 1kW PSU - using dual EPYC Milan built for enterprise clients for about 20k USD total. So I thought I could get away with something much less expensive (Corsair goes for around 250 USD here - EU).

Then there are deep learning guides/forums that recommend using even miners' PSUs - which I'd assume are even worse quality.

I use desktop power supplies. They have one 8 pin EPS power. I buy a PCI-E to EPS power adapter, as I don't use high power graphics cards, and desktop power supplies have one or more PCI-E power cables.
That's great to hear, thank you. Would you mind sharing which ones? I'll check the adapters - I assume the manufacturers list maximum allowed power for these adapters, right?

Right now I will use just one 3060 (170W), so it should be fine I guess.. but later on I'll add a pair of much stronger cards (imagine 3080/40xy).
 

lopgok

Active Member
Aug 14, 2017
215
148
43
I am using a Rosewill fortress 80+ platinum 450w ps.
I have a H11SSL and an EPYC 7302 processor.

As for the pcie to EPS adapters, I am pretty sure supermicro sells them.
The ones I use are cheap Chinese ones from ebay.
 

vvkvvk

Member
Feb 1, 2023
75
26
18
I've got my system running with Corsair RM series power supply, most of them come with two EPS12V cables and have 3 A available on 5 VSB.
 

ocfguy

Active Member
Oct 25, 2022
100
50
28
I’m running my system using a 1000W Super Flower 80+ Titanium PSU that I bought on sale. Figured that the increase in efficiency will pay for itself over the lifespan of the PSU. From my understanding it’s essentially the EVGA 1000 T2.
 

RolloZ170

Well-Known Member
Apr 24, 2016
5,142
1,546
113
would not give a cent for that AIDA64 mem/cache benchmrk.
have tested with skylake lga3647: first with 6 channels, then with 5 channels (3+2) unbalanced which gives 5/6 the speed of 6 channel balanced.
so one test is wrong, the hexa(6) channel balanced, or the 5 DIMMs unbalanced.
seems to me this bench only tests single channels and adds all results.
 

ano

Well-Known Member
Nov 7, 2022
633
259
63
sysbench on linux seems to be decent, and the 8 channel on 2nd gen vs 4 channel improvement is THERE
 

kaleruka

New Member
Jan 5, 2021
5
0
1
got intel 3647 - never got any problems with memory speeds...bought new platfrom with amd epic 2 becouse in specs was 200 gb per second.,..max what i achived - 120gb with 3200 memory speed on second generation epyc
 

penk38

New Member
Feb 16, 2023
4
0
1
Thanks everyone for the PSU recommendations, I just put the system together and ran some tests

TL;DR:
I am getting about half the expected performance I expected - running generic benchmarks e.g. passmark has median 40k, I am getting below 20k overall and roughly 40-60% on all the sub categories (integer math, compression, super bad on primes). I have H11ssl-i and 7551p.

Given it's my first server build, does this show issues in my configuration, can I use other tools to test or just ignore the site's results? More details below

My tasks, for the most part, can be described as multi-threaded sci calculations (vector/matrix operations) and data processing running in long batches. It would be hard to share my tasks and don't have access to other servers, so benchmarks are a way to approximate perf.

I know benchmarks are a bit frowned upon, but iven this is my very first server build, is this an indication something is wrong with my configuration? It is fresh build, no overclocking or optimization, the benchmark above ran on Windows. What should I look or are there better tools to assess performance?

Could you check with your cpu if you get similar results? Or is it possible that the site bundled in a 2-socket builds with the results. Given its -p variant, it really shouldnt.

Thank you for reading.
 

RolloZ170

Well-Known Member
Apr 24, 2016
5,142
1,546
113
I am getting about half the expected performance I expected
are you running same version of tests ? on top 41150 rate was made with V9 but later...
CPU Test Suite Average Results for AMD EPYC 7551P
From submitted results to PerformanceTest V10 as of 21st of March 2023
edit:
Naples if very high negative effected by bad memory channel population. run always with 8 same RDIMMs
 

FlorianZ

Active Member
Dec 10, 2019
173
220
43
The best benchmark would be testing your real workload.

Beyond what @RolloZ170 said about ideally running with all 8 channels populated with 2666 dimms, can you make sure that your cTDP is set correctly in the BIOS? If you have proper cooling, it should be safe to set this to 200W for Naples.
 

berbasor

New Member
Feb 14, 2019
13
5
3
Just picked up a 7302p+H11SSL-i combo, got a 7302 instead. Code still works for free Fedex shipping, and the combo came assembled in some pretty nice packaging, better than pretty much any board I've picked up that wasn't in retail packaging.

Also, CMOS battery is not present and BIOS is pretty old. That said, it booted without issue and I haven't had a single issue. Board looks brand new, but CPU is clearly used (tiny scratch on heatspreader, stamp in one corner)