Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Single Core benchmark
#1
I run Cinebench R15 Single Core test, but still see all cores at HWInfo maxing out at appr. same frequency.

Should not only one revving up with the rest low ? 
It also seems that having all running, the expected max for one core cannot be reached (although temps and powers are low).
That is on i7-8850h, Lenovo P52.
Reply
#2
Indeed, only one core should be maxing out under such load. But it depends on how the benchmark is written, if Windows schedules the workload to other idle threads, it can have such an effect. Check the utilization of other threads in HWiNFO or task manager.
Reply
#3
            Good suggestion  Big Grin
Please see attached files for details.

  1. While "single-core" benchmark is running, all cores come to speed.
  2. That is the case, even if "affinity" is defined on one core only.
  3. Full single-core turbo max is seen as a spike (and not in the core that is mainly assigned with the benchmark).
  4. The only difference when affinity is defined is that the core has much less C7 occupancy (but not with obvious benefits).
Please note that the attached chart is from a normal run (no affinity used), while HWINFO printouts are for normal run and affinity-controlled run.
  • Is there any way to get a substantial one-core turbo max?
  • Why all cores fire-up (just out of curiosity Huh)? It looks like, that the end result is always getting max frequencies for 2-4-6 cores running.
Reply
#4
It seems that affinity is not working as expected any more (W10 Pro 1809). Still OS allocates on its own.
The only program I found to provide a really controlled single-core test is OCCT.

I ran OCCT on single-thread and it gives a fully loaded (100%) thread. But still all other cores are up and running at full speed.
Max frequency is 4.2 GHz, instead of the expected 4.3 GHz.
Reply
#5
Some input would be greatly appreciated.
Reply
#6
(02-09-2019, 01:43 PM)abax2000 Wrote: Good suggestion  Big Grin
Please see attached files for details.

  1. While "single-core" benchmark is running, all cores come to speed.
  2. That is the case, even if "affinity" is defined on one core only.
  3. Full single-core turbo max is seen as a spike (and not in the core that is mainly assigned with the benchmark).
  4. The only difference when affinity is defined is that the core has much less C7 occupancy (but not with obvious benefits).
Please note that the attached chart is from a normal run (no affinity used), while HWINFO printouts are for normal run and affinity-controlled run.
  • Is there any way to get a substantial one-core turbo max?
  • Why all cores fire-up (just out of curiosity Huh)? It looks like, that the end result is always getting max frequencies for 2-4-6 cores running.

Often load is not kept on single core, but load is cycled from one core to another, so it should go like one core maxes out, then load is moved to another core, then that maxes out and so on.

With high performance power plan, all cores should be then at maximum possible all the times.

Then I have found out that 8th gen 6 core CPU is quite easily lowering turbo boost of 1 core when there are even 1% load on any of the other cores, while 6th gen did allow much more load on other cores until it was 'counted' and 2, 3 or 4 core max turbo was applied.

With 8th gen that happens really easily, so disabling all background processes have effect on time which maximum 1 core turbo is held during 1 core benchmark.

Some overclocking oriented motherboards feature multicore enhancement which changes a bit how this works, so load on other cores is not so easily dropping single core turbo, but I really wish it would work like it did with 6th gen 4 core CPU, but added cores seem to use different % limit that is allowed on cores until load is considered 2 core load or more.


So, if there is 3 cores that have load 1-5% and 1 core with 100% load, it is 4 core turbo for my 8th gen CPU and 1 core turbo for 6th gen CPU.

Maybe with your CPU something similar happens.

In such case, if your motherboard's bios has Multicore enhancement option, enabling that might help, disabling any not needed process might also help, but it is quite unlikely considering you have Lenovo.


However in practice in most situations, 8th gen Intel 1 core turbo is existing only in theory, under very perfect conditions from my experience, in all practical use, turbo is less than 2 core turbo speed, so benchmarking 1 core at full 1 core turbo speed gives only something to show off, in practice I see 4.6 to 4.3Ghz instead of 5Ghz of turbo boost clock speeds, but sure, it is possible to run Cinebench at 5ghz 1 core most of the time, which gives nice 221 score.

In practice, performance is better with all cores set to 4.6Ghz than with default, 1 core 5Ghz and 6 cores at 4.3Ghz, because CPU drops to 4.3Ghz even with heavy load only on 1 core so easily.
Reply
#7
Very similar to what I experience.
As you predict, no bios settings available.
Not possible to run Cinebench at 1 core max.
Strangely (even if on Better Performance), when loaded all core frequencies get high (4.1-4.2), although higher loads are jumping between few (see chart).
Maybe, with this mode of operation, some juice is lost in single-core applications (which are still some); and not only benchmarking "ego".
Reply
#8
(02-14-2019, 03:20 PM)abax2000 Wrote: Very similar to what I experience.
As you predict, no bios settings available.
Not possible to run Cinebench at 1 core max.
Strangely (even if on Better Performance), when loaded all core frequencies get high (4.1-4.2), although higher loads are jumping between few (see chart).
Maybe, with this mode of operation, some juice is lost in single-core applications (which are still some); and not only benchmarking "ego".

I think most that I do with computer benefits from higher single core speeds, there are then few things where more cores in addition to single core speed is great, but yeah, that is why I got 8086K, I can overclock it to perform when doing single core tasks more.

Of course you need to also count how much of clock speed you are loosing, something like 200Mhz is going to be less than 5% I believe, so what would be actually practical difference there?

Surely it is measurable and depending how much CPU does per tick does define how much more work higher speed would give you, but it can be quite small difference.

I can set 4 cores to 5Ghz so that clocks speeds don't change, it is improvement when doing single core tasks, but not really noticeable unless I really seek it out, I can't remember what score was in Cinebench, but it was not over 230 I think, however with stock settings test result is 208 or something like that as it is not holding turbo too well so compared to that there is clear improvement.

It is shame that laptops don't offer same customization options in Bios, that might help a bit.

I wonder how 9900K is as if it too drops to all core turbo speeds with even slight CPU load on other threads, then in single threaded apps it might be slower than 8086K, which is fastest together with 9900K in single core tests.

Still, I could use close to 30% higher single core performance, but there is nothing that gives such big boost. That would be close to 290 in Cinebench single core test.
Reply
#9
Totally agree that differences can be practically unimportant.
Still: 
  1. we are used to consider a cpu as a high precision piece of equipment, that is expected to work as expected.
  2. Testing and trying to figure out how "it" really works is always interesting  Rolleyes
  3. Abundance of cores and frequency is relatively new, so "plus or minus some, who cares" attitude not well engraved  Big Grin

[quote]Still, I could use close to 30% higher single core performance, but there is nothing that gives such big boost. That would be close to 290 in Cinebench single core test.[quote]
I hear you. But probably, we will have to wait for the next cpu generation.
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)