CPU benchmarking utility for Linux

Question

I am looking for a utility that will benchmark CPU performance under single and multi threaded instances. At present I have an old rig with a dual core CPU (E7500) at 3.6 Ghz and I am looking at replacing it with a quad core CPU (Q9400) at 3.2 Ghz. I want to see if I will notice a performance improvement with the extra 2 cores (albeit with a drop in core speed). I will clock the CPU's with the same FSB (400Mhz) and the cache size is the same per CPU (1.5MB) and for what its worth I have 4GB ram (with potential to upgrade to 6GB)

My son mainly uses the PC for playing TF2 (which I am trying to still get working under Linux) and I also use it for some video encoding (MP4 to DVD)

I am thinking that I could be better off with the quad core but any feedback would be appreciated.

Answer 1

Actually there is a a tool named sysbench.

You can install it with:

sudo apt-get install sysbench

To do CPU benchmarking you can use:

sysbench cpu run

This will run a single-threaded CPU benchmark. To use more than one thread, use:

sysbench --threads=16 cpu run

Answer 2

Alternatively, one can use stress-ng. It has a CPU stress test as one of the many stress tests built into the tool. The cpu stress test contains many different CPU stress methods covering integer, floating point, bit operations, mixed compute, prime computation, and a wide range of computations.

Install using:

sudo apt-get install stress-ng

To see the cpu related stress methods use:

stress-ng --cpu-method which

To benchmark, for example, matrix product for 60 seconds on 4 CPU threads, use:

stress-ng --cpu 4 --cpu-method matrixprod  --metrics-brief --perf -t 60
stress-ng: info:  [15876] dispatching hogs: 4 cpu
stress-ng: info:  [15876] successful run completed in 60.00s (1 min, 0.00 secs)
stress-ng: info:  [15876] stressor      bogo ops real time  usr time  sys time   bogo ops/s   bogo ops/s
stress-ng: info:  [15876]                          (secs)    (secs)    (secs)   (real time) (usr+sys time)
stress-ng: info:  [15876] cpu              71657     60.00    239.60      0.00      1194.25       299.07
stress-ng: info:  [15876] cpu:
stress-ng: info:  [15876]            885,244,279,148 CPU Cycles                    14.75 B/sec
stress-ng: info:  [15876]          1,289,303,858,968 Instructions                  21.49 B/sec (1.456 instr. per cycle)
stress-ng: info:  [15876]            201,499,961,692 Cache References               3.36 B/sec
stress-ng: info:  [15876]                    790,424 Cache Misses                  13.17 K/sec ( 0.00%)
stress-ng: info:  [15876]            157,689,508,544 Branch Instructions            2.63 B/sec
stress-ng: info:  [15876]              1,232,539,732 Branch Misses                 20.54 M/sec ( 0.78%)
stress-ng: info:  [15876]              5,755,605,036 Bus Cycles                    95.92 M/sec
stress-ng: info:  [15876]            817,296,440,876 Total Cycles                  13.62 B/sec
stress-ng: info:  [15876]                      8,532 Page Faults Minor            142.19 sec  
stress-ng: info:  [15876]                          0 Page Faults Major              0.00 sec  
stress-ng: info:  [15876]                        220 Context Switches               3.67 sec  
stress-ng: info:  [15876]                          0 CPU Migrations                 0.00 sec  
stress-ng: info:  [15876]                          0 Alignment Faults               0.00 sec  

Answer 3

Old question (with no selected answer yet 😱)

But I recently was looking for a tool available in multiple "distros" (Termux not really being a distro) including Ubuntu, and while the above mentioned packages are a common good choice, I read here: https://linuxhint.com/useful_linux_stress_test_benchmark_cpu_perf/ that 7-zip has a built-in benchmarking tool! And 7zip can be found in nearly every distros repository.

To run a single-thread benchmark: 7z b -mmt1

To run a multi-thread benchmark: 7z b

Results from my Pixel 2 phone:

7-Zip [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=utf8,Utf16=on,HugeFiles=on,64 bits,8 CPUs LE)

LE
CPU Freq:  1509  2234  2434  2447  2433  2406  2430  2425  2400

RAM size:    3657 MB,  # CPU hardware threads:   8
RAM usage:    435 MB,  # Benchmark threads:      1

                       Compressing  |                  Decompressing
Dict     Speed Usage    R/U Rating  |      Speed Usage    R/U Rating
         KiB/s     %   MIPS   MIPS  |      KiB/s     %   MIPS   MIPS

22:       1666    99   1631   1621  |      30427   100   2608   2598
23:       1602    99   1644   1633  |      29815   100   2589   2581
24:       1517    99   1644   1632  |      29441   100   2595   2585
25:       1397    99   1607   1596  |      28748   100   2567   2559
----------------------------------  | ------------------------------
Avr:              99   1632   1620  |              100   2590   2581
Tot:              99   2111   2100

Answer 4

phoronix-test-suite

sudo apt-get install phoronix-test-suite
phoronix-test-suite list-available-suites
# Chose one, and run it.
phoronix-test-suite run pts/cpu    

Benchmarks several real world CPU-heavy use cases like compression, encryption and databases.

Beware that pts/cpu and other benchmarks takes up a few gigabytes of disk space. This might imply that they have more realistic workloads.

Tested on Ubuntu 16.10.

Answer 5

If you don't have root permissions (e.g. managed web servers)

Measure single core:

-bash-4.2$ time cat </dev/urandom | head -c 1G | gzip >/dev/null

real    0m43.011s
user    0m42.608s
sys     0m8.819s

Measure all cores:

-bash-4.2$ time cat </dev/urandom | head -c 1G | pigz >/dev/null

real    0m13.870s
user    0m45.988s
sys     0m6.803s

If you already have pv but don't feel like installing special benchmarking software it's even more straightforward

a@a-virtual-machine:~$ pv </dev/urandom | gzip >/dev/null 
 149MiB 0:00:04 [37.5MiB/s]
a@a-virtual-machine:~$ pv </dev/urandom | pigz >/dev/null 
 348MiB 0:00:02 [ 173MiB/s]