Programming Puzzles & Code Golf Stack Exchange is a question and answer site for programming puzzle enthusiasts and code golfers. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top


There are lots of utilities out there capable of creating a high CPU load to stress-test your processor(s). On Microsoft Windows, you can even use the on-board calculator.exe, enter a large number like 999999999, and press n! several times to make your CPU(s) work overtime.

But what’s in a solution if you didn’t create it yourself?

The mission

Your mission – if you choose to accept it – is to create the smallest CPU stress-test tool on the planet.


  1. must produce 100% CPU load until aborted
  2. must take a numeric input, representing the number seconds the stress-test should run
  3. must allow user interaction (keypress, closing terminal window, or something like that) which should enable a user to abort the stress-test and/or quit the program
  4. must target Microsoft Windows, Mac OSx, and/or Linux.
    (Even a hamster could stress a Comodore64… therefore, you must target a current operating system.)

Must not…

  1. must not use 3rd-party programs or tools which replace expected functionality.
    (Proposing shortcuts in the likes of system('cpuStressThing.exe') disqualifies your proposal.)


  1. may use any approach/algorithm/functionality to produce expected 100% CPU load
  2. may use any programming or scripting language
    (as long as it allows practical verification of its functionality by running it)

Winning Condition

Present the smallest sourcecode possible. The winner is the one presenting the most minimal (in size) sourcecode that complies to the above “must” and “must not” conditions. Now, make that baby burn…


Since the question came up in the comment area… you only need to target 1 CPU core. I'm definitely not expecting you to produce a multi-core solution. After all, this should be fun – not work.

share|improve this question
Is "100% of one core" enough, or do you mean "100% of a multi-core CPU"? – Tobia Feb 17 '14 at 20:43
@Tobia Yep, 1 core is enough. I've edited my question to specifically include that information. Thanks for pointing me to the fact that that wasn't all too clear. – e-sushi Feb 17 '14 at 21:12
do cryptocurrency miners count/ – TheDoctor Feb 17 '14 at 21:59
@TheDoctor If you can make it fit the conditions I described… be my guest. It would surely be interesting to see a cryptocurrency miner that is able to beat (for example) a 36 byte bash script in filesize. – e-sushi Feb 17 '14 at 22:07
The problem is that most miners are several thousand lines of code. – TheDoctor Feb 17 '14 at 22:18

27 Answers 27

up vote 26 down vote accepted

Bash and standard utilities, 36 31 22 29 28 26 bytes

yes :|sh&sleep $1;kill $!
share|improve this answer
That looks pretty gorgeous for a Bash code! That's really a nice answer! – Ismael Miguel Feb 17 '14 at 21:36
You don't need the : in do :; done. I've found do;done does the job - that'll pull you in 2 bytes. Also +1 for being nearly half the length of my bash solution (I made it overly complicated for no good reason as I forgot about $!). – Chris J Feb 17 '14 at 21:45
@ChrisJ - that doesn't work for me: bash: syntax error near unexpected token `;'. I've tried these bash versions: 3.00.15(1)-release (x86_64-redhat-linux-gnu), 3.2.48(1)-release (x86_64-apple-darwin12), 4.2.25(1)-release (x86_64-pc-linux-gnu) – Digital Trauma Feb 17 '14 at 21:50
@ChrisJ - I guess you have yourself a 34-byte ksh answer then ;-) – Digital Trauma Feb 17 '14 at 22:05
I'd put $1 in place of 10 there, just to make it into a script that "takes numeric input". – Tobia Feb 17 '14 at 22:45

Bash/iputils (Linux), 14 bytes

ping6 -fw$1 ::

Flood-pings the IPv6 null address, until the -w deadline timer expires

caveat - only consumes 55-60% CPU on my test VM

Edit: - I retract my caveat. While top reports the ping6 process only consumes 55-60% CPU, I see the total CPU idle percentage (2 core VM) approach zero. This presumably is because a good deal of the processing is going on in the kernel as it handles the packets.

Note - must be run as root. As @Tobia comments, this seems like a reasonable requirement for something that will hog the CPU. And the OP approved it in the comments.

share|improve this answer
+1. I would remove sudo and just state that the test must be run as root. It seems like a reasonable requirement for something that will hog the CPU. – Tobia Feb 17 '14 at 22:41
@Tobia - thanks - I think you're onto something ;-) – Digital Trauma Feb 17 '14 at 22:44
ping -6 :: -t --> only if this on windows made my cpu go wild... It's just the linux equivalent on windows, which only gives errors and doesn't even load the cpu at 1%! I'm using windows 8 pro x64 on a core2quad 2.63GHz. – Ismael Miguel Feb 18 '14 at 0:40
@IsmaelMiguel - yep - I don't have windows to hand to test this. Thats why I stated "Linux" in the title ;-) – Digital Trauma Feb 18 '14 at 0:47
I know, I'm just "giving" away this answer as a non-working one, for those who are thinking about trying the same in windows, as I did and failed. – Ismael Miguel Feb 18 '14 at 0:56

bash builtins only 20 bytes

ulimit -t $1;exec $0
share|improve this answer
@e-sushi try with a faster terminal, e.g. a text console or xterm or rxvt – Geoff Reedy Feb 19 '14 at 15:58

Elf32 standalone binary - 86 bytes

I bet this is the smallest correctly formed Elf format binary that can be made to perform this function. This will execute without any additional support on any linux based platform, or potentially even without an operating system.

Binary download:

Hex dump:

0000000: 7f45 4c46 0101 0100 0000 0000 0000 0000  .ELF............
0000010: 0200 0300 0100 0000 5480 0408 3400 0000  ........T...4...
0000020: 0000 0000 0000 0000 3400 2000 0100 0000  ........4. .....
0000030: 0000 0000 0100 0000 0000 0000 0080 0408  ................
0000040: 0080 0408 5600 0000 5600 0000 0500 0000  ....V...V.......
0000050: 0010 0000 75fe                           ....u.

This is done by building an asm file with a minimal Elf header of its own, and skipping the use of ld altogether.



              org     0x08048000

ehdr:                                                 ; Elf32_Ehdr
              db      0x7F, "ELF", 1, 1, 1, 0         ;   e_ident
times 8       db      0
              dw      2                               ;   e_type
              dw      3                               ;   e_machine
              dd      1                               ;   e_version
              dd      _start                          ;   e_entry
              dd      phdr - $$                       ;   e_phoff
              dd      0                               ;   e_shoff
              dd      0                               ;   e_flags
              dw      ehdrsize                        ;   e_ehsize
              dw      phdrsize                        ;   e_phentsize
              dw      1                               ;   e_phnum
              dw      0                               ;   e_shentsize
              dw      0                               ;   e_shnum
              dw      0                               ;   e_shstrndx

ehdrsize      equ     $ - ehdr

phdr:                                                 ; Elf32_Phdr
              dd      1                               ;   p_type
              dd      0                               ;   p_offset
              dd      $$                              ;   p_vaddr
              dd      $$                              ;   p_paddr
              dd      filesize                        ;   p_filesz
              dd      filesize                        ;   p_memsz
              dd      5                               ;   p_flags
              dd      0x1000                          ;   p_align

phdrsize      equ     $ - phdr

section .text
global  _start
_start:       jnz     _start

filesize      equ     $ - $$

Built with nasm -f bin tiny_cpu_stresser_elf32.asm -o tiny_cpu_stresser_elf32

share|improve this answer

C, 52


Press Ctrl+C to exit.

enter image description here

share|improve this answer
If you do that in php, you will win! Just remove the $i=<> part and add <? to the beginning and you are good to go! – Ismael Miguel Feb 17 '14 at 21:43

Perl, 32


Now the embarrassing part: I foolishly put $t=time in front of $i=<> and was furiously trying to figure out why it exits a few seconds early.

Again, Ctrl+C to exit.

enter image description here

share|improve this answer

Unix C, 47

main(int a,char**b){alarm(atoi(b[1]));for(;;);}

Pass the time on the command line. Interrupt key (Ctrl-C) aborts.

share|improve this answer
if you use @ace's scanf trick, you can get this down to 39: main(a){for(scanf("%d",&a),alarm(a);;);} – Dave Feb 18 '14 at 8:32

Smalltalk (Smalltalk/X), 34

input: n; interrupt with CTRL-c or CMD-.

[[]loop]valueWithTimeout:n seconds

can golf better, if measured in days ;-) (just kidding):

[[]loop]valueWithTimeout:n days

enter image description here

or from a command line: enter image description here

share|improve this answer

This is not a serious attempt at it, but...

Bash, 12 bytes

:(){ :|:&};:

As found on Wikipedia.

WARNING: harmful code, don't run it on your computer!

- It produces 100% CPU load until system crashes;
- Allows user interaction to stop it (if you manage to kill all the forks, you can actually stop it...);
- You could give it a numeric input that represents the number of seconds it should run, but it won't use it.

share|improve this answer

PHP 43 40 bytes:

I hope this is an acceptable answer:



I could do like this: <?for(set_time_limit($_POST['t']);;); but it would lose flexibility and 3 bytes.

And i could cheat and do like this: <?for(set_time_limit($_REQUEST[t]);;);. It shaves off 2 bytes, but it's not a "standard" solution. Lets keep the game fair.

As @fireeyedboy and @primo suggested, you can also use this solution (34 bytes):


This allows it's use from the console, calling it like this:

php <filename> <time in seconds>

As i told, I'm not targeting the console solution, but they have to get the credit for this one.

Another answer could be this "monster", which is just both answers combined:


It's impossible to get key presses in php, without being on console, which I'm not targeting!

To stop it, you MUST abort the process (stopping the page from loading might stop the code)!

As a plus, it works in Android too! If you install a php server (free on Google Play).

To make it work, simply do like this:

You create a .php webpage and append ?t=<time in seconds> to the end of the url or submit a post (using a form or even ajax).

share|improve this answer
@e-sushi I fixed the answer and reduced 3 bytes. Not as small as the Bash solution, but close. And with flexibility! – Ismael Miguel Feb 17 '14 at 21:35
Run from the command line: replace $_REQUEST['t'] with $argv[1] then call it with: php -f cpustresstest.php <timelimit> and abort with ^C. – Decent Dabbler Feb 19 '14 at 7:09
Why not use $_GET instead of $_REQUEST? 4 bytes and you are using GET anyway – Kristoffer Sall-Storgaard Feb 19 '14 at 7:37
@KristofferSHansen Because the flexibility of my code will be severely hurt. – Ismael Miguel Feb 19 '14 at 9:26
@fireeyedboy That sounds like a good idea, but as I said, I'm not targeting to the console. – Ismael Miguel Feb 19 '14 at 9:27

Perl - 14 bytes


Sets a SIGALRM to be sent in input seconds, which terminates the script. In the meantime, it spins in a busy-wait.

Sample usage:

$ echo 4 | perl
Terminating on signal SIGALRM(14)

Perl - 12 (+1) bytes

If command line options are counted as one byte each, this could be reduced to 13 bytes using a -n:


Sample usage:

$ echo 4 | perl -n
Terminating on signal SIGALRM(14)
share|improve this answer
+1 for clever use of signals to concisely satisfy the exit requirement. – Mechanical snail Feb 23 '14 at 8:35

x86_64 assembly on Linux - 146 (source), 42 (assembled code)

The NASM minified source (146 bytes):

xor rdi,rdi
mov rcx,[rsp+16]
mov rcx,[rcx]
sub cl,'0'
jl k
imul rdi,10
movsx rdx,cl
add rdi,rdx
ror rcx,8
jmp l
mov rax,37
jmp s

Accepts a parameter on the command line specifying the number of seconds to run in the range (0, 9999999]; can be interrupted with the usual Ctrl-C.

You can assemble it with

nasm -f elf64 -o stress.o stress.asm && ld -o stress stress.o

In theory it would be necessary to add a global _start followed by a _start: label at the beginning, but ld manages to fix it by itself with little fuss.

The corresponding machine code (42 bytes):

00000000  48 31 ff 48 8b 4c 24 10  48 8b 09 80 e9 30 7c 11  |H1.H.L$.H....0|.|
00000010  48 6b ff 0a 48 0f be d1  48 01 d7 48 c1 c9 08 eb  |Hk..H...H..H....|
00000020  ea b8 25 00 00 00 0f 05  eb fe                    |..%.......|

(generated with nasm adding the BITS 64 directive)

A somewhat more readable version:

global _start

    xor rdi,rdi
    mov rcx,[rsp+16]
    mov rcx,[rcx]
    sub cl,'0'
    jl alarm
    imul rdi,10
    movsx rdx,cl
    add rdi,rdx
    ror rcx,8
    jmp argparse
    mov rax,37
    jmp loop
share|improve this answer

BrainFuck/Extended BrainFuck: 3


It will use 100% cpu on one core until aborted. All Brainfuck programs are valid EBF programs.

Zozotez LISP: 7 15 19

When using the little driver.

(:'r s) ; redfine read in the read-eval-print-loop

As a standalone expression without a driver: 15

((:'L(\()(L))))     ; setq a loop function and execute it    

Usage: echo '((\(L)(L))(\()(L)))' | jitbf

share|improve this answer

Python, 58 55 51

Wow... longer than the C one. There's got to be a better way. Still a tad long, but at least it beats the C solution!

import time;t=time.time;u=t()+input()
while t()<u:1
share|improve this answer
Haha, I love your first sentence. Personally I consider a C answer as the par. – ace Feb 17 '14 at 23:32
@ace (Barely) Fixed! – Bob Feb 17 '14 at 23:39
Now the other C solution beats yours! – ace Feb 18 '14 at 11:13
@ace Ah, I give up, At least this one is portable! :P (I actually looked at sigalrm earlier, but it's just too expensive to set up and use signals... Python can be rather wordy when its libs are required :[) – Bob Feb 18 '14 at 12:34

Java - 154 148 186

Weird error ate my Thread.sleep() part

public class Z{public static void main(String[]a) throws Exception{new Thread(){public void run(){for(;;);}.start();Thread.sleep(Byte.valueOf(a[0])*1000);System.exit(0);}}

and a more readable version:

public class Z {
    public static void main(String[] a) throws Exception {
        new Thread() {
            public void run() {
                for (;;)
        Thread.sleep(Byte.valueOf(a[0]) * 1000);

Spawns a new Thread with a nice endless loop (for(;;);) then on main thread a thread.sleep() and a System.exit(0) after timeout to exit; ctrl-c exits, too on cmdline wasnt able to shorthand that exit(). crashing wont work;

share|improve this answer

Linux sh and standard utilities, 14

Recent gnu coreutils includes a timeout utility which is helpful:

 timeout $1 yes
share|improve this answer
Nowhere near 100% CPU for me; it's throttled way too much by having to print I there another command? – Nick T Feb 19 '14 at 0:45
timeout $1 yes :|sh - 19 is probably the best you can do and get 100% utilisation. Tempted to steal this for my answer, but I'll be sportsman-like :) – Digital Trauma Feb 19 '14 at 3:41

Matlab - 19

tic;while toc<5;end

Replace 5 with desired execution time.

share|improve this answer

Go, 215 212 193 bytes (full)

package main
func main(){f.Parse()

Bonus, stresses all CPU's.

The Now() in the loop is there to kick in the scheduler, Now was the shortest function name I could find in my namespace

If I run go fmt the size increases to 286 277 254 bytes

share|improve this answer

Bash: 19 chars

function f(){ f;};f
share|improve this answer

Assembly: 16 bytes

_start:jg _start

Edit: Having not noticed the requirement to take a numeric input, i'm going to claim it does take one on the commandline, but ignores it =)

share|improve this answer
This takes a numeric input for the number of seconds the test runs? It looks to me like it simply loops infinitely. Whatever it is, remember to add an explanation in your answer. – Justin Feb 20 '14 at 19:54
Oh... totally read past that part of the question :/ – Riot Feb 20 '14 at 20:03

DOS Batch - 5 bytes


DOS Batch - 8 bytes


Second is a translation of the infamous sh forkbomb.

Ctrl+C breaks the program (unless you've tweaked the settings a little).

share|improve this answer

C#, 118

using a=System.DateTime;class b{static void Main(string[]c){var d=a.Now.AddSeconds(int.Parse(c[0]));while(d>a.Now){}}}


using a = System.DateTime;
class b 
    static void Main(string[] c) 
        var d = a.Now.AddSeconds(int.Parse(c[0]));
        while (d > a.Now) { } 

This requires a number as an argument which is the number of seconds to run. It will use 100% of one core for that much time or until crtl+c. I'm pretty sure this is as small as C# will go with its verbosity.

share|improve this answer

Batch, 2 characters


In essence, the program constantly starts itself over and over. Your results may vary, due to processor task allocation priority, but it works for me.

share|improve this answer

C# - 178 characters

using A=System.DateTime;class P{static void Main(string[]a){var b=A.Now.AddSeconds(int.Parse(a[0]));System.Threading.Tasks.Parallel.For(0,1<<30,(i,l)=>{if(A.Now>b)l.Stop();});}}

And more readable:

using A = System.DateTime;
    class P 
        static void Main(string[] a)
            var b = A.Now.AddSeconds(int.Parse(a[0]));
            System.Threading.Tasks.Parallel.For(0, 1 << 30, (i, l) => 
                if (A.Now > b)l.Stop(); 

Thats 178 chars in C# and uses all cores.

The only weakness that it is always ending because of the 1<<30 integer limit.

share|improve this answer

Java - 88 characters

class S{public static void main(String[]a){for(long i=0;i<Long.valueOf(a[0]);){i=i+1;}}}

This allows for 2⁶³-1 loops.

More Readable Version

class S {
    public static void main(String[] a) {
      for (long i = 0; i < Long.valueOf(a[0]);) { i = i + 1; }

C# - 87 characters

class S{public static void Main(string[]a){for(long i=0;i<long.Parse(a[0]);){i=i+1;}}}

More Readable Version

class S {
public static void Main(string[] a) {
    for(long i = 0;i < long.Parse(a[0]);i++) { i = i + 1; }

The program pegging the core

(This is on a 4 core system)

share|improve this answer
OP asked for 100% – Milo Mar 5 '14 at 4:44
The OP also specified that you only need to peg one core. It can go to 25% (which is 100% of 1 core). – Justin Krejcha Mar 5 '14 at 4:48
Not to be picky but your image shows 24.89% not 25% – Milo Mar 5 '14 at 4:48
True. It depends on what's happening on that core. If nothing is happening on the core, it will use the full 25%. – Justin Krejcha Mar 5 '14 at 4:52

EcmaScript 6:

z=z=>{while(1)z()};_=i=>(i+=1,i-=1,i++,i--,--i,++i,i<<=2,i>>=2,i+=0|Math.round(1+Math.random())&1|0,z(x=>setInterval(x=>z(x=>new Worker('data:text/javascript,'+_.toSource()),5))));setInterval(x=>z(x=>_(...Array(i=9e3).map((x,z)=>z*3/2*2/4*4e2>>2<<2))),5)

This will use 100% of the CPU on a single-core machine, and with Firefox, it has the added bonus that Firefox keeps using up more and more memory; the whole interface locks up and the only way to stop it is to kill Firefox in the task manager.

share|improve this answer

perl, 23 bytes

I can't figure out how to paste a literal control-T here, so I've typed $^T instead, but either works (the literal is 1 char shorter at 23 bytes):

$e=$^T+<>;1 until$e<time

$^T is just the time the interpreter started, so you can basically read that as time() since it is the first thing we calculate.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.