When I bought Samsung Chromebook friend started “nbench” on it. So I did same on my conference laptop. None of devices won…
Idea of testing cpu power was sitting somewhere at back of my head and finally I decided to just run one simple command available on nearly every GNU/Linux based system: “openssl speed”. Sure, on some systems it will use hardware accelerators, on others (or not) some options enabled to get more speed (like ARM assembly version which is not enabled in Debian/Ubuntu systems). But it is something what anyone can run at home.
Table may be hard to decipher so I also give it as Google Docs. It also has few more devices listed and whole tables (one below is for 8192 size samples).
Devices in table are:
- my Intel Core i7-2600K desktop
- my Intel U7300 (ultra low voltage) conference laptop
- Exynos5 Dual powered Samsung ARM Chromebook
- Exynos4 Dual powered Tizen development platform (got rid of it today)
- i.mx515 powered Efika MX Smartbook
- Beaglebone with AM335x cpu
- Sheevaplug (as only armv5te device which can compare with other entries)
Devices were running different versions of OpenSSL under different systems. It is listed in Google Docs document.
| CPU | Core i7 | U7300 | Exynos 5250 | Exynos 4210 | i.mx515 | AM335x | Feroceon 88FR131 |
|---|---|---|---|---|---|---|---|
| Architecture | x86-64 | x86-64 | armv7a (a15) | armv7a (a9) | armv7a (a8) | armv7a (a8) | armv5te |
| MHz | 3400 | 1300 | 1700 | 1000 | 800 | 720 | 1200 |
| OpenSSL version | 1.0.1c | 1.0.1c | 1.0.1c | 1.0.0f | 1.0.1a | 1.0.0i | 1.0.0d |
| md4 | 1111896 | 393198 | 328471 | 205906 | 143746 | 103068 | 119367 |
| md5 | 693969 | 249301 | 224040 | 148089 | 85401 | 53365 | 86518 |
| hmac(md5) | 686511 | 248859 | 225839 | 149153 | 86728 | 54981 | 87651 |
| sha1 | 721528 | 222770 | 147739 | 71233 | 49525 | 35446 | 38123 |
| rmd160 | 247453 | 93500 | 106935 | 57790 | 40188 | 26318 | 30803 |
| rc4 | 894615 | 225660 | 153949 | 86829 | 63770 | 29364 | 45036 |
| des cbc | 73703 | 27191 | 37811 | 21299 | 14966 | 8601 | 8829 |
| des ede3 | 28091 | 10578 | 14183 | 7806 | 5526 | 3005 | 3130 |
| seed cbc | 78204 | 31181 | 39002 | 24361 | 17650 | 11671 | 13087 |
| rc2 cbc | 44327 | 13839 | 23691 | 15494 | 10897 | 7393 | 10699 |
| blowfish cbc | 133455 | 52004 | 49471 | 37540 | 23536 | 15654 | 20584 |
| cast cbc | 118852 | 49162 | 55326 | 31738 | 22848 | 15298 | 20590 |
| aes-128 cbc | 127378 | 95955 | 65360 | 22386 | 16477 | 10876 | 11697 |
| aes-192 cbc | 106141 | 81002 | 55973 | 18653 | 13912 | 9221 | 9968 |
| aes-256 cbc | 90487 | 69148 | 48564 | 16419 | 12091 | 7981 | 8677 |
| camellia-128 | 187958 | 44403 | 58698 | 15447 | 23325 | 15507 | 14197 |
| camellia-192 | 141346 | 33180 | 45867 | 12090 | 18300 | 12261 | 11138 |
| camellia-256 | 141422 | 33272 | 45927 | 12050 | 18383 | 12247 | 11131 |
| sha256 | 216766 | 86791 | 64334 | 23427 | 18148 | 12022 | 13040 |
| sha512 | 336729 | 135935 | 31126 | 8877 | 5321 | 2484 | 3221 |
| whirlpool | 121211 | 47920 | 27820 | 4602 | 3840 | 2262 | 3085 |
| aes-128 ige | 122085 | 43018 | 63218 | 22126 | 15590 | 10469 | 11219 |
| aes-192 ige | 102133 | 36107 | 54269 | 18696 | 13355 | 8904 | 9647 |
| aes-256 ige | 87514 | 31001 | 47636 | 16307 | 11635 | 7735 | 8433 |
| ghash | 1938609 | 168034 | 35479 | 12136 |
Most interesting columns are U7300 and Exynos 5250 ones — 3 years old laptop which I bought for conferences compared to Chromebook. Looks like for next conferences/events I will rather go with Chromebook not UL30A. This will give me one or two hours of battery life less but it is much lighter device at same time. But have to test it first for few days to check is it comfortable enough for daily use.
your “wall of numbers” makes it hard to see anything, at least right align the numbers and drop the fractions
Done. That was also a reason why document with data is available.
This is really interesting, thanks for posting it. How does I/O compare across these systems? Memory/disk/network
This is hard to compare I/O in good way. Each device has other internal storage. Testing one SD card or USB drive would not test it too much. Similar with network — sheevaplug has 1GbE which can do 800-900Mbps which is far more then EfikaMX Smartbook can do over WiFi etc.
Which kernel were you using for the BeagleBone? Mainline 3.7-rc? The performance seems lower than it should be based on AM335x crypto benchmarks on the wiki.
The one it came with – 2.6.32 or something like that. I did not played with BeagleBone yet — just unpacked it to grab data from Cortex-A8 cpu.