FreeBSD: A Faster Platform For Linux Gaming Than Linux?

[b15hop]

Not having the auto-selection of arch/opts/flags is "the point". The ability to customize multiple facets of the OS is Gentoo's strength, and should be leveraged as such. I thank those individuals running ~x86, and those that test all of the combinations and permutations of cflags and library versions. It is their efforts that ultimately make pre-baked distributions more reliable. Gentoo users suffer so the rest of us don't have to. In regards to benchmarking, any Gentoo result is only representative of the user's Gentoo installation, not all Gentoo installations. I'd be willing to concede a bit if the benchmark author was the one constructing a targeted Gentoo installation for his/her benchmark. Alternatively, if Gentoo-BSD were to a point where meaningful comparison data could be obtained, I'd probably be interested in the comparison.

Very true. I don't envy those testers one bit. That's a LOT of testing to do. The combinations are mountainous in quantity.

[scjet]

Great article and review about FreeBSD, err I mean PCBSD

Being that PCBSD (which is really just FreeBSD + KDE) is probably the most widely used "Desktop" of all the BSD/Unix's,
and, Ubuntu is clearly the most widely used desktop of all the Linux's, then this is a very good look into a freindly comparison.
Obviously this is a "FAIR" comparison.
and so what if PCBSD is actually faster than Ubuntu in Ubuntu's own backyard, cause that is exactly what this review proved at this point in time.
btw, unless you've been under a rock, it's definitely not the first time where Linux apps run as good or in some cases faster on FreeBSD, than on some other Linux distro's. That's been goin' on for years now.
And regarding "...Clearly, BSD devs did something right. Again. ..." Yes they certainly did, just ask MacOSX, 'cause it sure ain't Linux under that hood.

I don't use Ubuntu much anymore, I've been using Arch for a couple years now.

Anyway, Kudos to PCBSD for comin' through as a very nice BSD/Unix Desktop alternative.


Thanks again Phronix, for this good Review

Cheers.

[XorEaxEax]

When there were broken benchmarks here, of Linux vs OpenSolaris, I did not argue too much. OpenSolaris used gcc 3.4, 32 bit, vs Linux gcc 4.3, 64bit. I did not whine about OpenSolaris, and I do not whine about Linux now.

I don't know what ' I did not argue too much' means, but argue you should since a there's certainly a big difference between gcc 3.4 and gcc 4.3 in the code they output, which obviously affects the performance of benchmarked binaries.

Anyway, it would be interesting to see BSD running Linux software. I have read people claiming that BSD is faster than Linux, when running Linux software. That would be interesting to see if it is true.

Yes that WOULD be interesting, so would this particular test, IF phoronix could conduct it in a way that would make the results meaningful. But sadly these tests are so very flawed. I mean seriously, 3d performance comparisons running with different compositors enabled, of which one atleast is well known to have severe impact on 3d applications/games? They couldn't make the test more pointless if they tried (unless of course if this was specifically to test nvidia proprietary on Ubuntu 11.04 vs FreeBSD 8.2, but that is NOT how the test was presented)

Now, of course you as a well-known BSD/Solaris fanboy will look at the results and say GREAT! Just like linux fanboys would have done the same had the results been reversed. But neither of you care one lick for the actual facts anyway, so what else is new. I on the other hand (and I suspect alot of others with me who can look at an operating system objectively) would like to know the actual performance, no matter what it would be, because we find these things interesting and therefore it's a darn shame that Phoronix does all these very interesting tests and yet so poorly constructed that the testresults end up being close to worthless.

Clearly, BSD devs did something right. Again.

I'm sure the BSD devs have done tons of things right, but what exactly were you referring to?

[kebabbert]

So in short, your main claim is that:
You don't care if the benchmark is broken because some unnamed idiot (which undoubtedly represents all the Linux users world-wide) started it.
Following this logic, all non-Linux users world-wide are complete idiots given the 3 y/o like comment made by tbyte. *

When there were broken benchmarks here, of Linux vs OpenSolaris, I did not argue too much. OpenSolaris used gcc 3.4, 32 bit, vs Linux gcc 4.3, 64bit. I did not whine about OpenSolaris, and I do not whine about Linux now.

I am saying that I dont see the same fanatic Linux fanboy behaviour from BSD or Solaris users, maybe because if you are the best, then you dont have to fight all the time. Only people feeling inferior or using inferior productst, must fight all the time. Bad self esteem, maybe.

However, you should not call Kraftman an "unnamed idiot", he might be angry with you. Then you will surely have problems as he fights with everyone.

Anyway, it would be interesting to see BSD running Linux software. I have read people claiming that BSD is faster than Linux, when running Linux software. That would be interesting to see if it is true. Clearly, BSD devs did something right. Again.

[Dorsai!]

I am only using general optimization (-o2 -march=native) and I am sure that this is (or better should be) the best configuration for my compiler/machine configuration. You are of course right that there are always applications and programs that break the rule but in general and as a rule of thumb you can say that most (maybe ~90%) of the source code out there runs better optimized for your machine than optimized for some machine. Everything else should be considered either a bug in the application or a bug in the specific arch backend of the used compiler.

And mind that such bugs can strike the "other" arch as well. In average -march=native should beat -march=pentium or somesuch in every machine configuration, well except for a real pentium that is where it should be equal ;-).

Another thing to consider: -o3 could be called faster in general, but the code produced is way larger. In some applications that leads to slower execution. While that may be an argument for your side, I doubt that many binary distros will compile -O3 on a per application level as the IO throughput and memory speed a user has available can not be forseen. I guess -o2 can be seen as consens for most binary distros. (except maybe -Os for the "tiny" distros)

Then there is the memory aspect of "speed" in Gentoo. You have your own personal, slimmed and trimmed down binaries which will load considerably faster in most occasions then their binary distro counterparts. The optimization does not only increase the starting time of applications but also increases the subjective "snappyness" of an application. When you press a button it might make a difference if for example 20 MB of libraries have to be loaded into the RAM or 100MB before you see any action. Mind that most desktop distros build "everything with everything" as long as a user COULD need it.

I do not know how well spread the usage of --as-needed in the linker flags is among the binary distros, but seeing that most patches for even the most mainstream apps came as gentoo introduced them I am not so sure it is used. If applied globaly it removes several MB of unneeded libraries which will not have to be loaded on the application start. When it was introduced as recommended it reduced the loading time of KDE4 and its applications considerably.

Sure, Gentoo will not magicly transform a Pentium 1 into a modern gaming rig, but the ability to tailor it to your needs will increase its subjective speed visibly. Also, even 2% gain are per definition faster, although not by much.

[russofris]

Yes, and of course its faster. But we are compiling not only for that but for flexibility. The speed is just a cherry-on-top bonus.

You were dead on up until this generalization. Targeted compilation (+march +mcpu and friends) does not guarantee an increase in performance for any specific app. Running the latest and greatest compiler does not guarantee an increase in performance for any specific app. Choosing one compile over another(ICC versus GCC) does not guarantee an increase in performance for any specific app. It is only through iterative testing with all combinations and permutations that the best combination of compiler, version, and cflags that the best performing solution for a given archtype is discovered. Gentoo-as-an-OS does little to promote/achieve this unless you were to automate the recompilation of the app and test each combination. Gentoo-as-a-community 'does' indirectly achieve this due to to the variety of platforms/architectures on which Gentoo runs and the diversity of their users.

I assure you that the fact that your hypothetical id3-tech-based game runs 5% faster on your Gentoo installation than your Fedora installation is almost entirely accidental, and not a result of -o3 versus -o2 (or any other magic flags in your make.conf). That is.... Unless you've iterated through different flags/compilers to find the best working combination on the target app and its dependencies via a profiler, all of your "optimizations" may be having an equally adverse effect.

[Dorsai!]

The reason I use gentoo is not because it is 1337 ?berfast, but the power over my system I get.
I can enable/disable for example mp3/dvd/midi/gtk/... support in ALL application with only one small global flag.
I can completely remove parts from for example KDE which are impossible even to deactivate on most distros.
I can upgrade any software without worrying about config files being overwritten or outdated.
I can go back to any previous version of a package if I dont like an update (or it brings bugs).
I can install multiple versions of packages in parallel if it makes sence.
I can configure a kernel, compile it, deploy it fully reboot ready with initramfs (raid and crypto enabled) in just one command.
I can create my own packages in just a few lines of text (especially easy for most kde software).

Yes, and of course its faster. But we are compiling not only for that but for flexibility. The speed is just a cherry-on-top bonus.

OK, it's a bitch to setup, but you learn so much on the way and once it is set up you have a very maintainable system with all the freedom and comfort you can get.

[gilboa]

I am quoting someone else from slashdot, who observed this behavior. I am not the only one to found this out: When Linux wins, everything is good and fair. When Linux looses, something is wrong. Isn't this a bit odd?

Regarding if this benchmark is bad or good:
I know Linux people here that compared a single core 1GHz old SPARC vs 2.4GHz dual core x86 Linux, and drew the conclusion that "This benchmark is fair and correct: Linux is clearly faster than Solaris". When I objected, he called me names such as "FUD, Troll, Idiot, etc" (etc, etc, etc, etc)

So in short, your main claim is that:
You don't care if the benchmark is broken because some unnamed idiot (which undoubtedly represents all the Linux users world-wide) started it.
Following this logic, all non-Linux users world-wide are complete idiots given the 3 y/o like comment made by tbyte. *
I bow before your superior logic.

- Gilboa
(* Rather ironic given the fact that my software ran [at different times] on Linux, Windows, FreeBSD ... and Solaris)

[gilboa]

Do You know what the meaning of "linux shill" is, Bilbo ?
Measuring by the level of your logic I'm picturing you as a low level windows support.

Um.... Sure, whatever.

[russofris]

Maybe something Gentoo needs is some kind of auto configuration tool. Something that helps with flag selection and compiler optimisation... I don't see the point otherwise.

Not having the auto-selection of arch/opts/flags is "the point". The ability to customize multiple facets of the OS is Gentoo's strength, and should be leveraged as such. I thank those individuals running ~x86, and those that test all of the combinations and permutations of cflags and library versions. It is their efforts that ultimately make pre-baked distributions more reliable. Gentoo users suffer so the rest of us don't have to. In regards to benchmarking, any Gentoo result is only representative of the user's Gentoo installation, not all Gentoo installations. I'd be willing to concede a bit if the benchmark author was the one constructing a targeted Gentoo installation for his/her benchmark. Alternatively, if Gentoo-BSD were to a point where meaningful comparison data could be obtained, I'd probably be interested in the comparison.