Power usage of AMD Epyc 7282 vs 7302P

Jellyfish

New Member
Apr 3, 2021
3
0
1
Hi everyone,

For my new build I've settled on going with an AMD EPYC 7002 series processor. I nearly bought the 7282 for having a low power rating with great performance. But then I noticed a strange thing about the benchmarks in the STH reviews and now I'm totally confused. The 7302P appears to be way lower on all power measures, even though it has a higher TDP advertised!

The 7282 is officially rated at a 120W TDP, and the benchmark on STH was reported as follows:
  • Idle Power (Performance Mode): 113W
  • STH 70% Load: 224W
  • STH 100% Load: 262W
  • Maximum Observed Power (Performance Mode): 319W
(From: https://www.servethehome.com/amd-epyc-7282-benchmarks-and-review/)

On the other hand, the 7302P is officially rated at 155W TDP, but then the STH benchmark looks like this:
  • Idle Power (Performance Mode): 99W
  • STH 70% Load: 169W
  • STH 100% Load: 198W
  • Maximum Observed Power (Performance Mode): 221W
(From: https://www.servethehome.com/amd-epyc-7302p-review-a-category-killer/)

Who's wrong? AMD? STH? Or are both right in different ways and I'm somehow misunderstanding something?

Any help will be much appreciated!
 
Last edited:

Evan

Well-Known Member
Jan 6, 2016
3,252
559
113
I think the systems they were installed in have some effect, the Tyan being a 1u hence using a bit more power for fans, also the P has all the chip to chip communication logic disables so a bit less power.
but that power consumption difference is huge ! Not sure what is going on.
 

hmw

Active Member
Apr 29, 2019
308
96
28
Depends on what the Tyan board had - 1.2kW dual PSUs, 100GbE & 25GbE Mellanox cards, Intel Optane drives - yeah that will suck up power like crazy. I wish STH did CPU power figures - like by measuring CPU power lines or maybe RAPL counters in Linux...

I have a 7302p in a Tyan S8030 and can tell you that the latter STH figures for the 7302p are actually spot on ...
 
  • Like
Reactions: Evan

Jellyfish

New Member
Apr 3, 2021
3
0
1
I've had a closer look now at some other STH reviews, and it's just getting weirder:

The 7352 has a 155W TDP advertised just like the 7302P, and these were its results on the exact same setup as the 7282 was tested on:
  • Idle Power (Performance Mode): 99W
  • STH 70% Load: 181W
  • STH 100% Load: 203W
  • Maximum Observed Power (Performance Mode): 232W
(From https://www.servethehome.com/amd-epyc-7352-benchmarks-and-review/)

If we look at the 7402P which is rated at 180W TDP, we see a roughly proportionate increase in power consumption figures as one would expect (tested on the same system as the 7302P):
  • Idle Power (Performance Mode): 99W
  • STH 70% Load: 185W
  • STH 100% Load: 212W
  • Maximum Observed Power (Performance Mode): 242W
(From https://www.servethehome.com/amd-epyc-7402p-review-24-cores-of-performance-and-value/)

Strangely, the 7302P and 7532 were tested before the 7282, so when testing the 7282, why didn't it stand out as that the power consumption was so much higher, despite a lower TDP rating?

From the results of the 7352 I'm taking that it's not the setup that caused the higher power usage. So what is it? Was the tested 7282 faulty? Was there an issue with how the test was run? Or does the 7282 truly use nearly 100W more at peak than the 7302P and 7532?

It's good to hear that the figures for the 7302P are in the right ballpark. Does anyone know more about the 7282? Prices changed over night and now the 7282 is significantly cheaper than the 7302P again, but I don't want to buy just on price if it will cost more in power usage in the long run and I get (slightly) less performance on top of that.
 

uldise

New Member
Jul 2, 2020
22
7
3
i have 7282 in Supermicro H12SLL-C, and measured a power with just 2x32GB RAM and one sata boot disk with Proxmox booted (no VMs running) :
- at idle it consume about 80W;
- give it a stress at 32 cores, and then it consume about 150W;

hope it helps.
 

meefishli

New Member
Apr 11, 2021
3
1
3
I second this: 7282 in Supermicro H12SSL-i 4x32GB, Seasonic PX-850, m.2 nvme running Proxmox.
- idle no VMs: less than 60W
- 4xHDD, Connect-X3, few VMs active, idling just about 90W
 
Last edited:

Jellyfish

New Member
Apr 3, 2021
3
0
1
That's great to hear, in that case I'll definitely go with the 7282! I've added a comment to the 7282 review in case someone else is confused.
 

meefishli

New Member
Apr 11, 2021
3
1
3
I was pleasantly surprised by the low power usage.

Some notes:
  • BMC power is about 6W (server switched off)
  • m.2 datacenter nvmes don't have a low power mode, so they consume about 5W when idle, whereas comsumer nvmes are less then 0.1W
  • It seems that the board and CPU can disable unused PCIe lanes, and thus reduce the power consumption.
  • Platinium PSU is very efficient.
  • I don't now anything about the "performance mode" mentioned. I used the BIOS defaults for most testing. Linux kernel is 5.4.106-pve, Linux CPU governor is performance. (proxmox default)
  • CPU Cooler/fan SNK-P0064AP4 The standard Supermicro fan control via ipmi works.
  • 4x low power PC Fans using a fan controller
 
  • Like
Reactions: Evan