why do cpu 2 in lga 2011 run hotter than cpu 1?

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

EffrafaxOfWug

Radioactive Member
Feb 12, 2015
1,394
511
113
That all depends on how the OS schedules the work; it may be that I/O-intensive processes (which spend much of their time waiting) get scheduled on the CPU with the I/O, while CPU-intensive processes get scheduled on the other CPU.
Remember it's not just the work that the CPUs might be doing with what's attached to the PCIe controller, but also the PCIe controller itself. When this is resident on the CPU, simply shunting data back and forth over the PCIe bus - even if you're not doing anything computationally expensive with it - will still generate a fair amount of heat. Moving lots of bits is expensive.

Of course it'd be interesting to see if the temperature difference holds true for a system sitting completely idle, i.e. little to no IO at all.
 

WANg

Well-Known Member
Jun 10, 2018
1,307
970
113
46
New York, NY
Remember it's not just the work that the CPUs might be doing with what's attached to the PCIe controller, but also the PCIe controller itself. When this is resident on the CPU, simply shunting data back and forth over the PCIe bus - even if you're not doing anything computationally expensive with it - will still generate a fair amount of heat. Moving lots of bits is expensive.

Of course it'd be interesting to see if the temperature difference holds true for a system sitting completely idle, i.e. little to no IO at all.
Well, that's some of the reasons why server geeks in prop trading shops disable IRQ balancing, do interrupt pinning, over-provision cores to CPUS (running only 4 trading strategies on an 8 core CPU so Turboboost kicks it a little higher) , take advantage of NUMA RAM region locality and place specific models of specific network cards in specific PCIe slots - not everything in the machine is created equal even if they are side-by-side to each other, every little bit counts towards latency, and yes, the "uncore" components on CPUs also matter in heat generation. Sometimes these things are not immediately intuitive. Transient CPU load does not always equal heat generation, sometimes the peripheral components on the die can matter more.

And yes, you CAN test it out. Use numactl and stress to specify artificial workloads on the machine, and keep tabs on it via IPMI queries on the sensors.

Frankly, unless it shows at least a 15-20 degree celsius variance between the 2 sockets, I wouldn't even pay it any heed.
 

Dawg10

Associate
Dec 24, 2016
220
114
43
Given a fairly even air flow over both cpus, more heat means more energy is being consumed .

Perhaps the primary side of the board has seen some efficiency enhancements; updates that were deferred on the secondary side as a cost saving measure.

It's not that socket 2 is hotter; socket 1 is cooler.
 
  • Like
Reactions: laserpaddy

wildpig1234

Well-Known Member
Aug 22, 2016
2,227
478
83
49
Given a fairly even air flow over both cpus, more heat means more energy is being consumed .

Perhaps the primary side of the board has seen some efficiency enhancements; updates that were deferred on the secondary side as a cost saving measure.

It's not that socket 2 is hotter; socket 1 is cooler.
that's one relative way of looking at it ;)
 

mstone

Active Member
Mar 11, 2015
505
118
43
46
Remember it's not just the work that the CPUs might be doing with what's attached to the PCIe controller, but also the PCIe controller itself. When this is resident on the CPU, simply shunting data back and forth over the PCIe bus - even if you're not doing anything computationally expensive with it - will still generate a fair amount of heat. Moving lots of bits is expensive.
It's important to be specific about what "lots" means--a process that's CPU or memory bound is going to be touching a heck of a lot more bits than most servers will be moving over the PCIe bus.

Along those lines, on servers where CPU2 is hotter, I typically also see the DIMM slots for CPU2 running hotter--suggesting that they're hitting the RAM a bit more...
 

Samir

Post Liker and Deal Hunter Extraordinaire!
Jul 21, 2017
3,287
1,474
113
49
HSV and SFO
about 3-5 degree C by my acc... but it's significant because it's ALWAYS there.....
I've seen that much variation by differences in thermal paste installation. I'd double-check everything and even try swapping stuff like heatsinks and processors to see if the problem follows something or stays just with cpu socket 2.
 

wildpig1234

Well-Known Member
Aug 22, 2016
2,227
478
83
49
I've seen that much variation by differences in thermal paste installation. I'd double-check everything and even try swapping stuff like heatsinks and processors to see if the problem follows something or stays just with cpu socket 2.
It;s 's consistently the number 2 cpu... the difference is not much but it is consistently there and that difference is actually a little higher when cpus at 100%.. there is not a variation of once a while or anything like that...
 

Samir

Post Liker and Deal Hunter Extraordinaire!
Jul 21, 2017
3,287
1,474
113
49
HSV and SFO
It;s 's consistently the number 2 cpu... the difference is not much but it is consistently there and that difference is actually a little higher when cpus at 100%.. there is not a variation of once a while or anything like that...
I just used hwinfo portable to look at the cpu temps on my dl380 g5 with dual x5460s and I see the same variation of 5C on two of my cores between one of my cpus and the other. I'd just chalk it up to a difference in cooling pattern by design because now that I'm thinking about it, it would be quite expensive (design-wise) to get any dual cpu system to cool the exact same on each socket.
 

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
HPE DL380 have CPU’s side by side and with identical cooling ducts, you will have all fans installed with the 2nd cpu so can’t see it being a design issues.
That’s it today I am going to check some e5 v1/2 and e5 v3/v4 systems and see if what is being said is the case also in general with my systems.
 

wildpig1234

Well-Known Member
Aug 22, 2016
2,227
478
83
49
I just used hwinfo portable to look at the cpu temps on my dl380 g5 with dual x5460s and I see the same variation of 5C on two of my cores between one of my cpus and the other. I'd just chalk it up to a difference in cooling pattern by design because now that I'm thinking about it, it would be quite expensive (design-wise) to get any dual cpu system to cool the exact same on each socket.
well, check again and see which cpu is constantly hotter. i bet it will be cpu 2 consistently... :) like any time you check, cpu2 is always hotter..... so this is not just a random event..


HPE DL380 have CPU’s side by side and with identical cooling ducts, you will have all fans installed with the 2nd cpu so can’t see it being a design issues.
That’s it today I am going to check some e5 v1/2 and e5 v3/v4 systems and see if what is being said is the case also in general with my systems.

let me know if you find an exception ;)....it's the same for me in s2600, lenovo d30, asus z9, z10..... i haven't have amd old or new multiple cpu to tell... maybe someone with dual epyc can comment if they consistently see one cpu always hotter?