Several authors have contributed to my knowledge of BogoMips. In this place, I would like to thank them highly.
Quoted from the Internet, origin unknown but brought to the attention by Eric S Raymond, esr@thyrsus.com, and Geoff Mackenzie, freon@dialstart.net, there is an humourously illustrative definition of BogoMips as ''the number of million times per second a processor can do absolutely nothing.''
On a more precise basis, from mail from Lars Wirzenius, wirzeniu@kruuna.Helsinki.FI, dated 9 September 1993, explaining BogoMips, with additional detailed information by Alessandro Rubini, rubini@morgana.systemy.it, and by howto-author Wim van Dorst:
`MIPS is short for Millions of Instructions Per Second. It is a measure for the computation speed of a program. Like most such measures, it is more often abused than used properly (it is very difficult to justly compare MIPS for different kinds of computers).
BogoMips are Linus's own invention. The linux kernel version 0.99.11 (dated 11 July 1993) needed a timing loop (the time is too short and/or needs to be too exact for a non-busy-loop method of waiting), which must be calibrated to the processor speed of the machine. Hence, the kernel measures at boot time how fast a certain kind of busy loop runs on a computer. "Bogo" comes from "bogus", i.e, something which is a fake. Hence, the BogoMips value gives some indication of the processor speed, but it is way too unscientific to be called anything but BogoMips.
The reasons (there are two) it is printed during boot-up is that a) it is slightly useful for debugging and for checking that the computers caches and turbo button work, and b) Linus loves to chuckle when he sees confused people on the news.'
BogoMips are being determined in
/usr/src/linux/init/main.c (simple C
algorithm, with a nice example of floating point arithmetic
within the fully integer kernel), and the pertaining
kernel variable loops_per_sec is
used in several drivers for more serious purpose. The
actual delay function udelay() is in
assembler, and therefore each port has its own definition
in /include/asm/delay.h. The
loops_per_sec variable and the
udelay() function are used in numerous drivers,
see:
cd /usr/src/linux #or where else source is located
find . -name '*.[hcS]' -exec fgrep loops_per_sec {} /dev/null \;
find . -name '*.[hcS]' -exec fgrep udelay {} /dev/null \;
The BogoMips calculation loop for the non Intel CPUs is similar but not the same, because it is programmed in another assembler language. BogoMips is however the only portable way over the various CPUs (Intel-type and non Intel-type) for getting an indication of the CPU speed. Even CPU clock speed is not available on all CPUs.
From a initiative by Ian Jackson, ijackson@nyx.cs.du.edu, and Przemek Klosowski, then with just three or four entries, much updated and expanded by howto-author Wim van Dorst for current data, as listed below:
As a very approximate guide, the BogoMips can be calculated by:
System BogoMips Comparison Intel 8088 clock * 0.004 0.02 Intel/AMD 386SX clock * 0.14 0.8 Intel/AMD 386DX clock * 0.18 1 (definition) Motorola 68030 clock * 0.25 1.4 Cyrix/IBM 486 clock * 0.34 1.8 Intel Pentium clock * 0.40 2.2 Intel 486 clock * 0.50 2.8 AMD 5x86 clock * 0.50 2.8 Mips R4000/R4400 clock * 0.50 2.8 Motorola 68040 clock * 0.67 3.7 PowerPC 603 clock * 0.67 3.7 Intel StrongArm clock * 0.66 3.7 Nexgen Nx586 clock * 0.75 4.2 PowerPC 601 clock * 0.84 4.7 Alpha 21064/21064A clock * 0.99 5.5 Alpha 21066/21066A clock * 0.99 5.5 Alpha 21164/21164A clock * 0.99 5.5 Intel Pentium Pro clock * 0.99 5.5 Cyrix 5x86/6x86 clock * 1.00 5.6 Intel Pentium II/III clock * 1.00 5.6 AMD K7/Athlon clock * 1.00 5.6 Intel Celeron clock * 1.00 5.6 Intel Itanium clock * 1.00 5.6 Mips R4600 clock * 1.00 5.6 Intel Itanium 2 clock * 1.49 8.3 Alpha 21264 clock * 1.99 11.1 Centaur VIA clock * 1.99 11.1 AMD K5/K6/K6-2/K6-III clock * 2.00 11.1 AMD Athlon XP/Athlon 64 clock * 2.00 11.1 AMD Duron/Opteron clock * 2.00 11.1 UltraSparc II clock * 2.00 11.1 Pentium MMX clock * 2.00 11.1 Pentium 4 clock * 2.00 11.1 Centaur C6-2 clock * 2.00 11.1 PowerPC 604/604e/750 clock * 2.00 11.1 Motorola 68060 clock * 2.01 11.2 Intel Xeon (hyperthreading) clock * 3.97 22.1 Hitachi SH-4 not enough data (yet) IBM S390 not enough data (yet) Intel ARM not enough data (yet)
Note that the BogoMips calculation loop does not take full advantage of the parallelism of various processors, such as the Intel Pentium and the Alpha 21164. Also read the section on 'New BogoMips algorithm?' below, since for some of these CPUs recent kernels may give different data.
There are three methods to determine the current BogoMips, viz.
looking in /proc/cpuinfo, e.g., with `cat /proc/cpuinfo`. This method is highly preferred above the alternatives.
looking in the syslog output to see what was printed there during booting. If you're lucky the information may still be on the booting virtual console (if necessary, switch to it with the Alt-F1 key combination), otherwise you can retrieve the information explicitly with dmesg or syslogk. This alternative gives accurate information but is more work.
using the standalone bogomips program. This is only recommended for non-Linux system, for reasons pointed out below.
A non-determinative alternative, which is also applicable for non-Linux systems such as Crays, Palm PDAs, DOS, and similar, may be a standalone BogoMips program. Some versions are currently available, viz., by Darrick Wong, djwong@thibs.menloschool.org, and by Philip Snowdon, philips@users.sf.net. Jeff Tranter, jeff_tranter@mitel.com was the original author. From his readme file:
`Tired of rebooting your system so you can see how many BogoMips it's running at today? [...] "Bogomips" is a standalone program that displays your system performance using one of the world's most recognized benchmarks. It uses the same code that is used in the Linux kernel while booting, but runs as a user program. [...] Version 1.3 of BogoMips is now portable and should run on any system that supports an ANSI C compiler and library.'
Note that due to system load values calculated with a standalone program on Linux systems may be quite different from registered in the list below. Intrinsically a standalone program cannot give precisely similar information to the boot sequence BogoMips, since system load will compete with the program run by an ordinary user. Therefore only boot sequence BogoMips ratings are listed below.
Be aware that Jeff's original file on
sunsite.unc.edu, named
/pub/Linux/system/status/bogo-1.2.tar.gz,
internally designated version 1.3,
is rather outdated. Therefore, check out Darrick's program on
http://thibs.menloschool.org/~djwong/programs/bogomips/,
if still reachable. He even has a MSW*nd*ws binary version. And
very interestingly, Phil Snowdon has developed the PalmOS version,
to be found at http://bogomips.sourceforge.net.
From Linus Torvalds, torvalds@cc.helsinki.fi, explaining about the variation one may see in the BogoMips rating, in c.o.l.development, at 28 April 1994
`The BogoMips calculation loop is "quantizised", so you're most likely to get the exact same number all the time. You usually will get different numbers only if the speed is just on the "edge", when small variations (different time for interrupt ticks etc) will make it jump from one value to the other.'
If a kernel is not compiled specifically for the pertaining CPU, also some (even large) variations of the BogoMips ratings can occur, mainly due to erroneous alignment. This problem apparently only occurs on the various x86 CPUs (Intel and clones). Fortunately it can easily be solved: recompile the kernel specifically for your CPU.
No, the BogoMips algorithm, contrarily to popular believe, did NOT change over the various kernel versions. Furthermore, it is intrinsically the same for all CPUs of all makes.
What did change from kernel version 2.2.13 to 2.2.14 is the CPU state setting just before the BogoMips calculation. This affects the BogoMips rating for all Intel and AMD Pentium variations, resulting in approximately 2*clock, if they were not 2*clock already.
The entries based on these upgraded kernel versions for CPUs where this change was actually relevant for (Pentium II/III, Celeron, Athlon) are marked in the list below with an initial * (star symbol).
Suggested by various questions on the net and private mail, e.g., by Lily, lbliao@alumni.caltech.edu, and by Pierre Frenkiel, frenkiel@cdfap2.in2p3.fr. In March 1995 they asked:
`When I boot Linux I get the message:
Calibrating delay loop.. ok - 23.96 BogoMips failedWhere/why has the calibration delay loop failed?'
It didn't fail. If it had failed the text would have been
Calibrating delay loop.. failed
What likely did fail was a driver for some gadget which you may not have in your machine. Just after calculating the BogoMips rating all device drivers are initiated. First the SCSI devices, then Net devices, etc. Any failure is duly reported. Noteworthy is the AHA152x driver. Other effects of failing drivers (and not of failing BogoMips calculations) are systems crashes, long waits, and complete system locks, somewhere close to (just before or just after) outputting the ok - xx.xx BogoMips text.
Since Linux 1.2 many error messages have improved, so upgrade to at least that version to find out which particular driver it is that is failing. Also recompile the kernel to only include the drivers that are actually needed for your hardware configuration.
Cyrix 486-like CPUs need cache enabling software, sometimes referred to as BogoBoost software. Cyrix 5x86 and 6x86 CPUs may have their BogoMips improved drastically by branch-prediction (BIOS option). Note that the performance improvement may be marginal. There are several packages available for adjusting Cyrix CPUs, such as the BogoBoost patch, cx5x86mod, and set6x86, all from the normal archives, in obvious places. It is reported the Cyrix 6x86 CPUs may give better performance when the kernel is compiled with 486-optimization, instead Pentium-optimization.
NexGen 386-enhanced CPUs, marked as Nx586, are listed as 386-like, since the fact that they are performing like Pentium machines is not relevant to BogoMips.
AMD 5x86, also denoted as AMD 486DX5, are quadrupled 486/33 machines. They are fully in line with other 486 CPUs. The AMD K5 and the K6 are Pentium-like CPUs, with their own BogoMips multipliers.
Let me add that there are only two reasons for paying attention to the BogoMips rating that is presented on booting Linux:
To see whether it is in the proper range for the particular processor, its clock frequency, and the potentially present cache. Many CPUs are prone to faulty setups of
memory cache setting (write-back is wrong for BogoMips, often reported lower than 5; write-through is ok)
turbo-buttons (should be ON)
BIOS-software emulated fake cache (change it for real cache)
similar cache and clock related things, sometimes also BIOS-software related
To see whether your system is faster than mine. Of course this is completely wrong, unreliable, ill-founded, and utterly useless, but all benchmarks suffer from this same problem. So why not use it? This inherent stupidity has never before stopped people from using benchmarks, has it? :-)
Note of the author: this remark was made somewhere in the mid nineties when comparing high BogoMips rates for some low-capacity CPUs with high clockrates (e.g., 486) to rates for CPUs with lower clock rates but much more capacity (e.g., Pentiums).
More serious uses for real benchmarking are addressed in the Linux Benchmarking Howto by Andre D. Balsa.