i5-10500T vs i7-8700 - Can't even get close to median scores

[schelles77]

I recently moved some edge/IoT machines from an i5-10500T platform to an i7-8700 platform. Immediately after deploying some of the first machines I could just "feel" that they were slower, but couldn't determine why. After A LOT of PassMark testing I couldn't get anywhere about 10,800 on the i7-8700, where the median score on cpubenchmark is 13300+.

So, for comparison, I wondered what the "old" machines would test at - they scored 10,736 which is about as close to the median of 10,742 as you can get.

Why are these i7-8700's so much "slower" than the i5-10700T? If the median PassMark score is 13,300 for the i7-8700, why can't I even get above 11,000? I've enabled SpeedStep, Turbo Mode, I've used tuned-adm to set the profile to throughput-performance -- nothing has even moved the needle. They are just stuck at 10,800 -- the same PassMark score as an i5-10700T.

I'm at my wits end. I've spent hours researching, reading, making changes. I even downgraded to 18.04 to see if there was a diffence. Nada. I tried different versions of BIOS, nada.

I'm interested in any possible ideas to help troubleshoot this. Here are the PassMark results from the "old" i5 and the "new" i7:

[David (PassMark)]

Physics and Prime tests are especially low and these tests use more RAM.

So maybe it is a slow RAM issue?
e.g. RAM not in dual channel mode, or just slow RAM.

See also this page for general performance advice issues
https://www.passmark.com/support/per...erformance.php

[schelles77]

Physics and Prime tests are especially low and these tests use more RAM.

So maybe it is a slow RAM issue?
e.g. RAM not in dual channel mode, or just slow RAM.

See also this page for general performance advice issues
https://www.passmark.com/support/per...erformance.php

Thank you, David.

There was only 1 SODIMM installed, so we were single channel. We replaced it with 2x8GB Crucial memory and the score improved - 11505 now versus 10816, still miserable compared to a normal i7-8700. We have SpeedStep turned on in the BIOS, HyperThreading, but there's nothing else obvious that would explain this. Any other thoughts on why this particular model of PC could be so drastically bad compared to all the thousands of other results?

We're testing on Ubuntu 20.04.

The only thing left is a crappy SSD that is installed by the manufacturer, but I didn't think SSD performance would impact PassMark?

[David (PassMark)]

I assume the Primes and Physics results are still the most significant low ones?

Is the RAM in the correct slots to get dual channel enabled?
What's the specs for the RAM (clock speed & timing)?
Are you running a XMP profile?

Do you have any other background activity running on the machine?
Are CPU temperatures OK?

No, the SSD performance shouldn't impact the CPUMark.

[schelles77]

I assume the Primes and Physics results are still the most significant low ones?

Is the RAM in the correct slots to get dual channel enabled?
What's the specs for the RAM (clock speed & timing)?
Are you running a XMP profile?

Do you have any other background activity running on the machine?
Are CPU temperatures OK?

No, the SSD performance shouldn't impact the CPUMark.

David,

Thank you for taking the time to help with this, and so quickly. The overall issue we have is that the factory is testing the exact same hardware, only under Windows versus Ubuntu, and they are getting much different results:



We run the same tests, with the only difference being 2x8GB of RAM versus their 2x4GB and we get:

 PassMark PerformanceTest Linux

Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (x86_64)

6 cores @ 4600 MHz | 15.6 GiB RAM

Number of Processes: 12 | Test Iterations: 1 | Test Duration: Medium

--------------------------------------------------------------------------

CPU Mark: 11505

Integer Math 39474 Million Operations/s

Floating Point Math 24799 Million Operations/s

Prime Numbers 35 Million Primes/s

Sorting 19224 Thousand Strings/s

Encryption 4978 MB/s

Compression 147 MB/s

CPU Single Threaded 2784 Million Operations/s

Physics 747 Frames/s

Extended Instructions (SSE) 7529 Million Matrices/s

Our RAM is DDR4-2400, Crucial. There are only 2 slots, so that's good, too.

Temperatures are fine, there is nothing else running. We've tried clean, new installs of both Ubuntu 20.04 and 18.04.

coretemp-isa-0000

Adapter: ISA adapter

Package id 0: +35.0°C (high = +82.0°C, crit = +100.0°C)

Core 0: +35.0°C (high = +82.0°C, crit = +100.0°C)

Core 1: +34.0°C (high = +82.0°C, crit = +100.0°C)

Core 2: +33.0°C (high = +82.0°C, crit = +100.0°C)

Core 3: +34.0°C (high = +82.0°C, crit = +100.0°C)

Core 4: +34.0°C (high = +82.0°C, crit = +100.0°C)

Core 5: +32.0°C (high = +82.0°C, crit = +100.0°C)


pch_skylake-virtual-0

Adapter: Virtual device

temp1: +38.0°C

Yesterday we replaced the 1x8GB RAM with the 2x8GB and went from 10816 to 11505. Nothing else we've tried will get us above that, and the factory claims there's nothing wrong since they are testing close to average.

[schelles77]

And the plot thickens..

Out of curiosity I decided to install Windows 10 on the exact same hardware that was producing the PassMark of 11505, essentially the exact same hardware and now software that the manufacturer is using. And, to my surprise, I got a 91 percentile result - just like them:



So then I decided to run PT10:




So PT9 = 13858, PT10 (Windows) = 11373, PT10 (?) Linux = 11505.

I've read the threads about changes between 9 and 10, but I'm really confused as to what to believe now. Not sure if I have a problem and poor hardware (56th percentile PT10) or if it's working exceptionally (91st percentile PT9).

[David (PassMark)]

V9 and V10 use different benchmarks and are only very roughly comparable. So best not to even bother with a V9 to V10 comparison if you have other data available.

There is a more detailed description of the V9 to V10 differences here
https://www.passmark.com/support/per...10-results.php

pch_skylake-virtual-0
Adapter: Virtual device

Would you be happening to run all these tests in a VM, but not mention it?

(Windows) = 11373, PT10 (?) Linux = 11505

Good to see they are close, despite the very significant operating system differences.

Not sure if I have a problem and poor hardware (56th percentile PT10)
or if it's working exceptionally (91st percentile PT9).

PT9 was active between 2016 and 2019.
So in 2016 a i7-8700 machine (released in 2017) would have been blazing fast. 91th Percentile

PT10 was active from 2020. So a i7-8700 is far from impressive compared to current machines. 56th Percentile

[schelles77]

V9 and V10 use different benchmarks and are only very roughly comparable. So best not to even bother with a V9 to V10 comparison if you have other data available.

There is a more detailed description of the V9 to V10 differences here
https://www.passmark.com/support/per...10-results.php



Would you be happening to run all these tests in a VM, but not mention it?




Good to see they are close, despite the very significant operating system differences.



PT9 was active between 2016 and 2019.
So in 2016 a i7-8700 machine (released in 2017) would have been blazing fast. 91th Percentile

PT10 was active from 2020. So a i7-8700 is far from impressive compared to current machines. 56th Percentile

David,

Not virtual at all, this is a bare metal machine. I don't know what that virtual sensor is.

I read up on the differences between PT9 and PT10, but here's where I'm still struggling - maybe missing the bigger picture: If we use the most recent PT10 results, the machine is scoring in the 11,500 range on both Windows 10 and Linux, so they agree. But, when I check the scores for this processor (https://www.cpubenchmark.net/cpu.php....20GHz&id=3099) the average score is 13,078.

Is it reasonable for me to be asking the manufacturer why this configuration is 22% slower than the average i7-8700 system out in the wild?

[David (PassMark)]

Is it reasonable for me to be asking the manufacturer why this configuration is 22% slower than the average i7-8700 system out in the wild?

Yes, it is a reasonable question.

See also this page for the most common issues.
https://www.passmark.com/support/per...erformance.php

Here is the distribution of results in PT10 for this CPU.

So while the average is around 13,000 the median is 13,700.



What do the RAM benchmarks look like in PT10?

[schelles77]

Well, after much back and forth with the hardware vendor their response was:

"Repeated testing by our engineers,we found that the overall power design of the baseboard unable to make i7-8700 get the best efficiency.(and turbo model is not recommended)"

It seems that we proved that this hardware design is substandard and we were not getting our money's worth.

Thanks for the help and support, David!

[David (PassMark)]

Well it is nice to hear that the PerformanceTest software was able to help.

No so good that the motherboard is poorly designed. This is a pretty rare thing as motherboard requirements for particular CPUs / socket types are very well defined.
Is this some type of super tiny motherboard or super low power device?