make -j# Tests

[cheapie]

I recently got an 8-core CPU (AMD FX-8350 4GHz) and I did some testing to see how many parallel compilation jobs got the best speed. These tests were done compiling "The Powder Toy" (it uses SCons, but the -j option does the same thing as on make).

-j1 (one core only) takes 97.877s
-j4 (number of "modules") takes 27.201s
-j8 (number of cores) takes 19.586s
-j9 (cores+1) takes 19.463s
-j10 (cores+2) takes 19.583s
-j16 (cores*2) takes 19.861s

As you can see, cores+1 got the best speed on this computer, but not by much.

[Elysarian]

Re: make -j# Tests

cores+1 is the recommended setting in Gentoo, I've never tested other values myself though.

nice bit of info there that proves the guys who write the manual aren't just pulling numbers out of thin air

[eccerr0r]

Re: make -j# Tests

Very difficult to answer this question if you have a multithread core. I agree with the single thread CPU case: cores + 1 is good. But that number will change for a multithread box.

I think I set my multithread box to number of threads the processing units can handle, and no more...

[cheapie]

Re: make -j# Tests

Very difficult to answer this question if you have a multithread core. I agree with the single thread CPU case: cores + 1 is good. But that number will change for a multithread box.

I think I set my multithread box to number of threads the processing units can handle, and no more...

This particular CPU is sort of halfway there. It has 4 "modules", each comprised of 2 integer cores and a shared floating point unit. So, essentially, with integer operations, it's an 8-core 8-thread CPU, but with floating point operations, it's a 4-core 8-thread CPU.

[shovenose]

Re: make -j# Tests

This particular CPU is sort of halfway there. It has 4 "modules", each comprised of 2 integer cores and a shared floating point unit. So, essentially, with integer operations, it's an 8-core 8-thread CPU, but with floating point operations, it's a 4-core 8-thread CPU.

An Intel Quad Core w/HT is as much an "8 core" CPU as an FX-8350 is.
As in, it isn't.

That said, while something like an i7-4770K is faster than an FX-8350 both are excellent processors with plenty of CPU power to spare to run pretty much everything 99.95% of people will want now and for years to come.

Intel's main advantage is more efficiency and less power consumption. The CPU costs more initially but it will pay for itself over its useful life span because it uses less power.

[Elysarian]

Re: make -j# Tests

An Intel Quad Core w/HT is as much an "8 core" CPU as an FX-8350 is.
As in, it isn't.

That said, while something like an i7-4770K is faster than an FX-8350 both are excellent processors with plenty of CPU power to spare to run pretty much everything 99.95% of people will want now and for years to come.

Intel's main advantage is more efficiency and less power consumption. The CPU costs more initially but it will pay for itself over its useful life span because it uses less power.

agree with pretty much everything said in this post.

was having a discussion with my wife on the subject of pc upgrades recently and came to the conclusion that neither of us needs an upgrade in the foreseeable future ( I have an FX8120, she has a 6 core version, we both have 16GB of RAM, I have a GTX760, she has a GT640).

About the only upgrades we see as possible in the next 12 months would be SSDs and possibly a GPU upgrade for her machine.

In the past I've always tried to effectively double my CPU "performance" each time I've upgraded:
first pc was 486sx25
then a dx2 66
p166 mmx
AMD K62 350
Athlon 750
Athlon XP 4000 (bigger jump as longer time span)
Phenom quad core
then my current FX8120.

[mariushm]

Re: make -j# Tests

Here it was:

k6-2 333 Mhz (bought by parents)
486 66 Mhz (first year of university, gifted to me)
duron 1100 mhz (bought with my own work money in 2nd year of university, same with the rest)
barton 2500+ (1833 mhz)
intel d805
intel q6600
fx-8320