7-Zip / Discussion / Open Discussion: 7-Zip: CPU performance test

Igor Pavlov - 2012-11-09

This message is for 7-Zip users.

I ask to perform CPU performance test with 7-Benchmark test suite:
http://sourceforge.net/projects/sevenmax/files/7-Benchmark/7bench1200.7z/download
Read readme.txt file for more instructions.

Also you can make additional tests with 7-Zip 9.30 alpha 64-bit (or 32-bit):

7z b -mmt1 > 7z.txt 7z b -mmt2 >> 7z.txt 7z b >> 7z.txt 7z b -mm=* -mmt1 > bench1.txt 7z b -mm=* > bench.txt

And include these *.txt files to results.7z archive.
Also you can add the log from CPU-Z program to results.7z.

The most interesting are results for new CPUs with Ivy Bridge, Bulldozer, Piledriver, Bobcat cores.
Please disable Intel "Turbo Boost" or AMD "Turbo CORE" feature in BIOS before test.

You can upload results.7z to some server and post the name of CPU and the link (in changed form) to results.7z file here at forum.
Or send email message to me via sourceforge message system.

Anti-spam filter at this forum doesn't allow links to external sites. You can change the link in forum message (remove dots from the link or something else).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Anonymous - 2012-11-09

Sorry, posts including links are not allowed, to prevent spam.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

gora2 - 2012-11-09

_ttp://gora.7zsfx.info/test/results.7z

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

virtualmanpl - 2012-11-13

sent on PM

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephan Busch - 2012-11-22

sent on PM

Kindly regards

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-23

squeezechart:

Turbo Boost was enabled in BIOS in your system.
So results were distorted.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

genda - 2012-11-26

sent on PM

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-27

kugend:
Please disable Turbo Core in BIOS.
And send all results as one archive.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

genda - 2012-11-28

ipavlov:
Limit P-State, non-turbo

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-28

kugend:
What is your CPU and OS version?
Does it have turbo core option in BIOS?
If yes, disable that turbo core option in BIOS and run test.bat.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-28

When the CPU frequency is OK (no turbo feature), RDTSC must return correct number of CPU cycles. It works so at least for Intel CPUs. I don't know about new AMD CPUs.

memlat p6 1
must show correct L1 cache latency in CPU cycles in first column.
if that value is inorrect, probably there is problem with CPU frequency.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-28

correction:
memlat 1 p6
must show correct L1 cache latency.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

genda - 2012-11-28

memlat p6 1
same result

CPU frequency (cpu-z report text)
TSC Stock frequency
APERF Real frequency
MPERF Stock frequency

ipavlov:
"show results in ns" damage my CPU?

1st run "memlat32 1 p16" (4-K   5.33   2.68……)
?th run "memlat32 1 p16" (4-K   5.33   2.68……)
?th run "memlat32 1 p16 n" (4-K   0.00   0.00……Cache latency = 0.00 ns = 0.00 cycles)
After run "memlat32 1 p16" (4-K   0.00   0.00……Cache latency = 0.00 ns = 0.00 cycles)
………………???
Run "memlat32/64 1 p16" again (4-K   0.00   0.00……Cache latency = 0.00 ns = 0.00 cycles)

Only parameter "p" have program, cold/warm boot also not restored :'(

4K→56K all 0.00
      4K   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00

64K→512K Ladder-like 00.00
    64-K 10.27   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00   0.00
   320-K 10.26   5.15   3.44   2.58   2.07   0.00   0.00   0.00   0.00   0.00   0.00   0.00
   512-K 10.28   5.17   3.45   2.58   2.07   1.73   1.48   1.30   0.00   0.00   0.00   0.00

BW- 32 B 21628 43105……
BW- 64 B 43256 86210……
BW-128 B 86512 172421……

I need contact my product warranty? >_< or AMD technical support…

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-29

kugend :

1) Can you answer my questions?
What is your CPU and OS version?
Does it have "turbo core" option in BIOS?
If yes, disable that "turbo core" option in BIOS and run test.bat.

2) You don't need to call
memlat 1 p16

you must call
memlat 1 p6

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

genda - 2012-11-29

Typos
inorrect:p16 correction:p6

My numbers come back @@?
Before "1→8 p2→p16" 00.00

After 1x hours shutdown
Now    "1→8 p2→p8"   have num, "p9→p16" 4K→56K gradual change to all of 00.00

1)8150, win7, yes, no need, test.bat takes a long time

Disable Turbo for vista/7/8?
Windows Registry Editor Version 5.00

"Attributes"=dword:00000000

Goto Power Options, find the option "Processor performance boost policy", change the options to <50 or 0.

2)According to the proportion of downsize, no difference
memlat64 1 p6

2.7G
4-K   5.33   2.68   1.81   1.35   1.07
3.6G
4-K   4.00   2.00   1.36   1.01   0.81

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-29

kugend:
1) I need "turbo core" disbaled (it's simpler to do via BIOS).
2) I need all results files from test.bat. (it's about 30-40 minutes for your CPU).

Why do you post 1.4 GHz / 2.7 GHz results, if stock frequency for your CPU is 3.6G?
Can you set CPU frequency to 3.6 GHz and call test.bat?

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

genda - 2012-11-29

ipavlov:
Why do you post 1.4 GHz / 2.7 GHz results?
1.4G is P-state 4, 2.7G is P-state 2
2.7G has about twice as big as that 1.4G, easy to compare results

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-11-29

All these P-State things are not so interesting,
I still need main results for stock CPU frequency (3.6 GHz for 8150).
So disable "Turbo Core", run test.bat and upload the archive with results to some server.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

onetime22 - 2012-12-02

ipavlov:
FYI the CPU frequency calculated by pipelen and memlat seems to be off. They both agree it's 3.2 GHz for my Phenom II 1090T.
While that is correct for stock clocks the local CPU P-stats are modified with K10STAT, having a P0 of 3.6 GHz (CPU-Z reports 3.6 GHz alright).
So it would be misleading to create a benchmark on this computer with 3.6 GHz while the logs show 3.2 GHz. Sure, you tell everybody to run stock clocks but IMHO the programs should report the actual CPU frequency nevertheless.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

onetime22 - 2012-12-02

addition because editing is not possible:

- BIOS config:
- - Cool'n'Quiet is off
- - Turbo core is off because they would interfere with K10STAT

- K10STAT config:
- - P0 state is not turbo-P-state anymore but just a normal P-state, so CPU throttles through P3 (800 MHz) -> P0 (3.6 GHz)

- 7bench:
- ran the Win7 x64 tests (autoselected by test.bat)
- - locked CPU at P0-state so no P-state-switching occurs (so why 3.2 GHz??)
- - I presume pipeline lengths and latency is miscalculated because of wrong detected CPU freq?

Is there a way I can contribute?

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Igor Pavlov - 2012-12-02

7bench uses RDTSC instruction to calculate the number of CPU cycles.
But RDTSC instruction works under stock frequency.
Future versions of 7bench will look real CPU frequency.

With current version of 7bench you must set CPU to stock frequency (the frequency of RDTSC instruction).

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

acm_fan - 2016-11-14

Hello! Can you make (an optional) function for holding position of the window for this test (CPU Benchmark)? It will be very helpful. Window opened in a center of display every time when it called. I mean, window must be open in a position that was used at the last time. Thanks! Example:
http://cdn.overclock.net/4/42/421f04e8_14456073282_abc23794da_z.jpeg

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

7-Zip: CPU performance test

A free file archiver for extremely high compression

Forums

Help

7-Zip: CPU performance test

7-Zip: CPU performance test

A free file archiver for extremely high compression

Forums

Help

7-Zip: CPU performance test document.SUBSCRIPTION_OPTIONS = { "thing": "topic", "subscribed": false, "url": "subscribe", "icon": { "css": "fa fa-envelope-o" } };

7-Zip: CPU performance test