Memory benchmarking
Linux Distributions appear to lack very simple tools for doing simple memory benchmarks of computers and VPS nodes. However, there are some tools which can be easily compiled and used to get a rough yet useful estimate of a computers memory performance.
STREAM[edit]
The The Department of Computer Science at the University of Virginia in the US has a very simple memory benchmarking tool called STREAM. It is written with "supercomputers" in mind but it works just fine on any computer.
STREAM's homepage has nothing but links benchmark results which are less interesting. However, the FAQ's getting started page has a link to the raw source-code and instructions for compiling it. You only need to download one file and compile it with GCC using standard options to use STREAM on a GNU/Linux machine. Download stream.c:
wget http://www.cs.virginia.edu/stream/FTP/Code/stream.c
(steam.c 5.10, copy in case it gets removed)
and compile it:
gcc -O stream.c -o stream
then run your fine stream binary:
./stream
Results[edit]
AMD A8-7600 Radeon R7 / DDR3 1600[edit]
Running ./stream produces the following on a AMD A8-7600 Radeon R7 with DDR3 memory running at 1600 MT/s (can be checked with dmidecode |grep MT):
Function Best Rate MB/s Avg time Min time Max time Copy: 6762.8 0.024147 0.023659 0.025088 Scale: 6749.1 0.024086 0.023707 0.025246 Add: 7896.6 0.030838 0.030393 0.032152 Triad: 7896.3 0.030804 0.030394 0.032172
AMD Opteron(tm) Processor 3365 / DDR3 1333[edit]
Function Best Rate MB/s Avg time Min time Max time Copy: 6238.6 0.026221 0.025647 0.028108 Scale: 6035.5 0.027099 0.026510 0.028166 Add: 7016.9 0.035991 0.034203 0.038751 Triad: 6783.5 0.036459 0.035380 0.039577
AMD Ryzen 5 2400G / DDR4 3000[edit]
Function Best Rate MB/s Avg time Min time Max time Copy: 15299.3 0.010567 0.010458 0.010715 Scale: 15714.2 0.010439 0.010182 0.010942 Add: 19301.9 0.012571 0.012434 0.012680 Triad: 19060.0 0.012695 0.012592 0.012823
AMD Ryzen 5 1600X / DDR4 2666[edit]
Function Best Rate MB/s Avg time Min time Max time Copy: 14815.0 0.011304 0.010800 0.012203 Scale: 15555.3 0.011094 0.010286 0.012718 Add: 18744.1 0.013326 0.012804 0.014540 Triad: 18873.8 0.013266 0.012716 0.014157
Intel(R) Core(TM) i7-5500U CPU @ 2.40GHz / DDR3 1600[edit]
Function Best Rate MB/s Avg time Min time Max time Copy: 13963.0 0.011535 0.011459 0.011761 Scale: 13701.0 0.012012 0.011678 0.012899 Add: 15690.4 0.015560 0.015296 0.016412 Triad: 15439.2 0.015704 0.015545 0.016123
Intel(R) Pentium(R) CPU N4200 @ 1.10GHz / DDR3 1600[edit]
Function Best Rate MB/s Avg time Min time Max time Copy: 6326.4 0.026008 0.025291 0.027652 Scale: 6189.1 0.028329 0.025852 0.033533 Add: 6569.6 0.036891 0.036532 0.037660 Triad: 6708.0 0.036325 0.035778 0.037086
Testing larger amounts[edit]
Compiling stream with -DSTREAM_ARRAY_SIZE=100000000 increases it's memory use from a default of just 0.2 GB to 2.2 GB - which is nothing on modern machines. It may be wise to call the binary something reflecting the array side used to compile it:
gcc -O -DSTREAM_ARRAY_SIZE=100000000 stream.c -o stream.100M
The results are very similar with the larger array size. The memory test does take longer.
sysbench[edit]
sysbench is a cli benchmarking suite which is available in most distributions under the package name sysbench. It comes with several built-in tests:
- fileio - File I/O test
- cpu - CPU performance test
- memory - Memory functions speed test
- threads - Threads subsystem performance test
- mutex - Mutex performance test
The memory test will, as the name implies, benchmark memory performance. It has several options which can be revealed by running sysbench memory help
- --memory-block-size=SIZE size of memory block for test [1K]
- --memory-total-size=SIZE total size of data to transfer [100G]
- --memory-scope=STRING memory access scope {global,local} [global]
- --memory-hugetlb[=on|off] allocate memory from HugeTLB pool [off]
- --memory-oper=STRING type of memory operations {read, write, none} [write]
- --memory-access-mode=STRING memory access mode {seq,rnd} [seq]
Running sysbench with the default options for a test is done by sysbench testname run. To run a write memory test, which is the default, type:
sysbench memory run
You may also want to check sysbench --memory-access-mode=rnd memory run to see the differences between sequential (the default test) and random memory writes.
AMD A8-7600 Radeon R7 / DDR3 1600 (4x4 GB)[edit]
- seq: 2767.71 MiB/sec
- rnd: 538.37 MiB/sec
AMD Opteron(tm) Processor 3365 / DDR3 1333[edit]
- seq: 2229.98 MiB/sec
AMD Ryzen 5 2400G / DDR4 2800 (2x16 GB)[edit]
- seq: 5979.86 MiB/sec
- rnd: 1872.83 MiB/sec
AMD Ryzen 5 1600X / DDR4 2666 (4x8 GB)[edit]
- seq: 6197.66 MiB/sec
- rnd: 1729.55 MiB/sec
Intel(R) Core(TM) i7-5500U CPU @ 2.40GHz / DDR3 1600[edit]
- seq: 4936.25 MiB/sec
- rnd: 1533.87 MiB/sec
Intel(R) Pentium(R) CPU N4200 @ 1.10GHz / DDR3 1600[edit]
- seq: 3400.94 MiB/sec
Takeaways[edit]
Memory performance depends just as much or more on the CPU than the memory type and speed. This is specially true when it comes to DDR 1600 on both older AMD processors and low-powered Intel processors like the N4200. You would think that the same type of memory sticks would perform similarly on two different machines but that is simply not the case. The CPU/platform (and perhaps other factors like motherboard) makes a huge difference.
Enable comment auto-refresher
Anonymous (409cac0fc2)
Permalink |