AMD EPYC 7401P Linux Benchmarks and Review – Something Special

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

bitrot

Member
Aug 7, 2017
95
25
8
Truly EPYC benchmark results, a beast of a CPU indeed at a very competitive price!

I don’t need all that horsepower for my home server and hence decided to go the more power efficient Xeon Silver route. But if energy prices weren’t as ridiculous over here as they are and if those EPYCs were actually available to buy (I build my servers myself, so Dell etc. don’t count for me), I would have definitely considered an EPYC 7401P or 7351P. Unbeatable performance for the price.
 

MiniKnight

Well-Known Member
Mar 30, 2012
3,072
973
113
NYC
How's this against Threadripper 1950X that's almost the same price? 16 faster cores have gotta be better than 24 slower ones.
 

Edu

Member
Aug 8, 2017
55
8
8
33
Great article. The power measurements make more sense this time. 7401P uses slightly more than double the amount of power as a Silver 4116, but also has double the amount of cores.
 

bitrot

Member
Aug 7, 2017
95
25
8
That’s the problem, in addition to the power consumption (for me, at least) - EPYC CPUs are simply not available and motherboards, especially single socket ones, aren’t either. Supermicro hasn’t even listed their single socket motherboard properly on their website yet, it’s still “coming soon”.

Intel availability is far better over here in Europe, although the more ‘exotic’ motherboards like the Supermicro X11SPH-nCTF aren’t widely available yet either. Mine was sent directly to me by Supermicro Netherlands.
 

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Most of the EPYC sales now are in systems. Most servers are sold fully assembled.
 
  • Like
Reactions: eva2000

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
That sucks. That makes the barrier to entry higher than say an Intel setup for the DIYer. Oh well. I’ll keep sticking to watching the used intel market then on eBay and just look on with envy at all the Zen coverage.
Don’t worry i am sure it will change soon enough, just have to wait till there is more supply.
I am surprised there is not more already but I guess AMD has to be careful not to over produce in case it was not as successful as it looks like it will be.

Amazing chips, the nailed the performance this generation for sure, shame about the power consumption which I am rather surprised about given its on a new process node I did not expect as bad as it seems to be.
Originally I was critical of the likley backward step intel made with the new high TDP chipsets but seems not to be an issue at all with scalable being better than e5 v4
 

Edu

Member
Aug 8, 2017
55
8
8
33
Amazing chips, the nailed the performance this generation for sure, shame about the power consumption which I am rather surprised about given its on a new process node I did not expect as bad as it seems to be.
Power seems quite reasonable to me. It consumes twice the power of a Silver 4116 system, but also delivers twice the performance using twice the number of cores.
 
  • Like
Reactions: gigatexal

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
Power seems quite reasonable to me. It consumes twice the power of a Silver 4116 system, but also delivers twice the performance using twice the number of cores.
It uses 50% more power than a dual 4114/4116 system. (Not single)
That’s in Patrick’s 70% type load test, real world depending on workload could be completely different who knows, performance costs power sure, the point was that it’s not a nothing arguement.

In one of the other reviews or similar systems the difference was 800-1000kwh per year + cooling so the system would cost $300 or more extra to run per year, of course again that’s loaded not idle.
Then again you same up front 3 or 4 years operating costs :)
 

Edu

Member
Aug 8, 2017
55
8
8
33
I notice that all the benchmarks that you run are NUMA aware. In theory non-NUMA aware applications should run better on Xeon than Epyc, because of the fact the Xeon has just one monolithic die. However, no one has ever tested that theory with real world benchmarks. Are there any plans to run non-NUMA aware applications, for instance database applications?
 
  • Like
Reactions: Pepe

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
I notice that all the benchmarks that you run are NUMA aware. In theory non-NUMA aware applications should run better on Xeon than Epyc, because of the fact the Xeon has just one monolithic die. However, no one has ever tested that theory with real world benchmarks. Are there any plans to run non-NUMA aware applications, for instance database applications?
Sure some DB like mongodb have no NUMA awareness, on NUMA machines the only thing you can do is interleave memory, but Oracle on Linux although installing by default in non-NUMA mode can be enabled and it does have an effect on at least some DB operations.
 

KioskAdmin

Active Member
Jan 20, 2015
156
32
28
53
I notice that all the benchmarks that you run are NUMA aware. In theory non-NUMA aware applications should run better on Xeon than Epyc, because of the fact the Xeon has just one monolithic die. However, no one has ever tested that theory with real world benchmarks. Are there any plans to run non-NUMA aware applications, for instance database applications?
Example? I don't think much of that is really numa aware.