Gentoo is Fast on New G5s

Durin_Deathless writes "According to a thread on the Gentoo/PPC forums, some Gentoo users have installed Gentoo on their new G5s without any problems whatsoever. Benchmarks are extraordinary: compiling kde on a G5 running at half speed takes 15 minutes, while it takes one hour on the fastest P4 available. Gentoo/PowerPC lead, Pieter Van den Abeele, reported that the machine currently runs at half speed due to fan controlling hardware not yet supported. The Gentoo team will post benchmarks, and will update installation instructions as soon as possible. There is some question as to what exactly was compiled, as the times seem impossibly fast even on the P4."

Comments

compiling KDE

[platipusrc]

There was a followup post where the originator of the thread stated that he actually only compiled KDEBASE and KDELIBS...not the entire KDE setup. So no, unfortunately the G5s don't bring us 10x the performance just yet :)

I can't wait to see

[iendedi]

How this thing runs Linux once it has been optimized for it (correct drivers, kernel patches, compiler switches, compiler version, etc..)

How teh fuck?

[bic2k]

Do the editors of slashdot even read what they link to? The forum thread was the most useless thread I've ever read. Nothing is proven. Nothing is confirmed. And some guy is claiming he installed suse 64bit for AMD on a G5... tell me there is something more interesting out there!!!

Re:How teh fuck?

[Fizzlewhiff]

I installed 64 bit HP/UX on my Pentium 4. It only took me one HP/UX CD and a dab of epoxy but it installed nicely on the computer. I am sure someone could do the same with a G5 and a 64bit AMD distro, although a Mac person would probably do a much nicer job since they are more artistic. They'd probably frame and mat it in a nice shadow box with some pretty flowers or colored macaronis.

Re:How teh fuck?

[bic2k]

nope

Re:How teh fuck?

[ernstp]

On thought that I had was that doesn't SuSE have a dist for IBM 64 bit zSeries too? They should run Power4, and Power4 binaries should be compatible with the PowerPC 970 if I understand correctly. Maybe he confused AMD and IBM... both ends with an M right?

Anyway, my first thought when I read the most recent messages in the thread was, vapor...

Re:How teh fuck?

[pyros]

Yeah, AMD ends in M.

Re:How teh fuck?

[ernstp]

Hey, I'm trying to come up with an explanatin here... it's close, ok?

Linking to a Forum?

[E1ven]

Oh my, the poor site.. If this ever goes to the FP, it's doomed--

Testing the G5, part 1
Today at work we were surprised to find a brand new black box sitting at our feet, it was the new G5! First thing I did was rip open the tape, and open the beautiful card board box. Once I got it open I noticed there wasnt much special about the manuals, cables and cds. So I pulled out the huge machine (its alot bigger than I thought AND ALOT Heavier!!) And stuck it on my desk. I quickly hooked it all together, and plugged in the monitor. Before I opened it I took the side of the case off and noticed the clear window inside. Its very attractive but I feel its weird because it doesnt go all the way to the top of the case so the DVD-R and harddrives are not behind the window. So I booted it up with the side on and I couldnt believe how SILENT it was. I had to open it again to make sure the fans were moving. So I was playing with it even more and I noticed that if you take the window off the fans go to full RPM and are LOUD. When the window is on they look like they are going in slow motion, no wonder its quiet. So after I got though the registration wizard apple makes you do, I got to my Macosx Desktop. The first place I headed was system profiler. I dont remember what was there exactly but it showed a Sony DVD-R, 1.6 G5, two 128mb DDR 400 DIMMS in slot 1 and 2 and it was running Macosx 10.2.7. Nothing else was really that surprising. Since I had to do real work I couldnt really play with my new machine, so I have to wait until tomorrow to play with it. The first thing I am going to do is Burn a PPC Live CD and see if it runs, hopefully it does, To bad it has a Nvidia graphics card. So Ill report more to you tomorrow.

Talk to you later

Back to top

Compiling KDE in 15 minutes?

Did you emerge 'kde' or 'kdebase' or 'kdelibs' ? Does that 15 minute figure also include dependencies?
_________________

I'm away for the weekend on a business trip.. but on Monday, I will post the benchmark tests and everything else for it.

KDE took about 3 hours to download and compiled in under 15 minutes... my Pentium 4 took roughly an hour to compile KDE...

I will run some tests on monday and run some kernel tests and such then post my results.

I used kernel 2.4.22 and kernel 2.6test4
both of which booted fine.. X only worked so far on 2.4.22

I've talked to benh (who gets his G5 from YDL in about a week) he told me that there is a remote possibility for the current kernel to work, but it wouldn't have SATA working and would run at half speed (due to the fans - 1ghz powerbooks had the same problem)

If compiling kde took only 15 minutes, how long is it going to take at full speed? 7 minutes?

Did you have to make a custom livecd to boot or did the current livecds work perfect?

----
I will report back to you on monday with all the information with I get back home

BTW... I did not use the livecd to bootup
I will post an install guide monday

[edit]Forgot to add... for KDE.. I was just talking about the KDEBASE files and KDELIB files.. not all the other stuff like games, and media, etc[/edit]
_________________

Re:Linking to a Forum?

[b1t r0t]

and I couldnt believe how SILENT it was

I gotta agree with the original poster about the silent part. I was in the Dallas area Apple Store a couple of weeks ago and noticed how quite the place was... except whenever I walked past an MDD G4! Every other Mac model (not just the G5) was wonderfully silent. Steve Jobs may be crazy, but he's a genius.

compile time?

[sdibb]

There is some question as to what exactly was compiled, as the times seem impossibly fast even on the P4.

No kidding... my Athlon XP 2500 took about 15 hours to compile KDE. You can't even download all the KDE packages in 15 minutes.

Besides, the actual "kde" ebuild is nothing more than a little flag that says yes indeed, I installed all the other KDE packages: kdebase, kdenetwork, kdemultimedia, kdeaddons, kdeedu, kdegames, kdegraphics, kdeadmin, kdeutils, kdeartwork and kdepim.

Fortunately, you don't need to install each one if you want to use KDE's basic functions.

Re:compile time?

[BrookHarty]

You can't even download all the KDE packages in 15 minutes.

You can if you download all the languages.

Re:compile time?

[Anonymous Coward]

I dont get the joke...someone please explain!

Re:compile time?

[Arielholic]

You can't even download all the KDE packages in 15 minutes

If you would have read the thread, you would have known that it took him 3 hours to download kde.

Re:compile time?

[damiam]

Actually, you can, if you're lucky enough to have a local mirror on a gigabit ethernet network. :-)

Hopefully, Apple will help

[ZackSchil]

I think Apple likes it very much when someone buys their hardware and runs Linux on it. The large margins on their boxen help cover OS X R&D (which is more expensive than you could possibly fathom, a full modern OS in 4 years, wait, what?!). They even have a reseller that is allowed to sell Macs pre-installed with Linux, unlike MS, who threaten any Wintel PC makers who try to offer Linux on their boxes with expensive licensing.

Plus, Mac OS X plays very nicely with Linux boxes and they know it. I just hope Apple will help the small Linux on Mac community integrate their software and proprietary hardware for at least full functionality. I have a feeling they will.

Re:Hopefully, Apple will help

[ComputerSlicer23]

Uhhh, there's a lot more then 4 years into Mac OS X. I was reading Mac OS X runtime docs 4 years ago (they we're the best docs for OpenStep runtime I was using at a job, I quit 3.5 years ago, and I worked there 1.5 years, so I am pretty sure it's got at least 5 years in it at Apple that they've had public documentation). They have a lot of pre-existing software from FreeBSD, and Mach. They also have all of the code and coders that cames from NeXT. It's still no small feat by Apple, but how long was BeOS around? They had pretty good stuff from what I hear, and I'm not sure how long they have been around. QNX 4 was put together pretty quickly. It was a complete re-write of QNX. So was Neutrino.

Second, Linux could have been written in a lot less time if it had been designed from the beginning to end up what it is (it might not be as good, but it could have been pretty good a lot sooner then it was). The concept that SMP existed. Having somebody who knew what the hell they we're doing at the beginning of it. Linus is a damned genious now, but when he started it, he wasn't a C programmer at all. Which leads me to guess, he wasn't much of a UNIX programmer at the time (let alone an experienced kernel programmer). It wasn't like he designed around the concept of having SMP, or even optimized disk accesses. Scalability wasn't a big deal. Running with more then 8MB of RAM was impossible (he only had 8, so if you had more and wanted to use it, you had to fix it yourself). Second, it's a whole heck of a lot easier to write an OS when the platform is relatively fixed (yeah it needs to work under x86, but if it doesn't, that's not Apple's problem).

Kirby

Re:Hopefully, Apple will help

[nacturation]

Linus is a damned genious now, but when he started it, he wasn't a C programmer at all. Which leads me to guess, he wasn't much of a UNIX programmer at the time (let alone an experienced kernel programmer).

When I first started using Linux, it was at version 0.2 back in 1991/92 or thereabouts. It was blazingly fast on my 486 -- naturally, text mode only but it made MSDOS shit its pants. What I saw back then was certainly not the result of someone who's "not a C programmer at all".

Re:Hopefully, Apple will help

[ZackSchil]

I was more contrasting Apple's development of Mac OS X to that of Windows XP. When Apple took NeXt Step, it was about as far as Windows NT 3 in terms of usability on a modern consumer desktop. Then they banged out an NT 4 equivalent in the public beta back in 1999. They pulled off 10.0 to 10.1 around 2001 to meet Windows 2000. Then came Jaguar in 2002 to meet Windows XP. They kept up with Microsoft and were, in fact, blasting by them. Longhorn has been delayed to 2005. Mac OS X 10.3 comes out late this year! Apple is blasting through with Mac OS X in hyper-development. It's costing a lot put the point is, they have nowhere near the type of resources Microsoft has! Not even close! But they're dong it and Apple fans like myself are footing the bill (quite literally, I'm happy to do it, $120 a year is cool for this speed of OS progression, though others think not)

Re:Hopefully, Apple will help

[ComputerSlicer23]

I'll point out that's not possible.... :-)

That problem at that point was that the people who where writting the code you ran under DOS we're stupid. The OS got out of your way. Other then the possibility of the filesystem was slower, DOS should have ran faster. Possibly only mildly faster, but faster.

Linux could multi-task, which means there is a scheduler. The scheduler is pure overhead in terms of speed. So now you are down to comparing apps to apps. Maybe a little bit of filesystem thrown in for giggles. I supposed DOS could have had a crappy timer interrupt or something, or crappy serial I/O interrupts. However, as I recall, most of the actual O/S of DOS was really in the BIOS, so blame the MoBo maker. The BIOS implemented all of the functionality to talk with hardware so DOS didn't have to write any real drivers.

As far as Linus not being a C programmer, go read his biographies and the early days of Linux. Go read interviews with his old roommate. He wasn't a C programmer. He originally started doing it in assembly. In fact the whole project was to learn more about the x86 assembly, it started as a terminal emulator.

He didn't know how sprintf worked. His roommate Lars implemented the original sprintf for the kernel for him.

Kirby

Re:Hopefully, Apple will help

[WatertonMan]

One should point out though that Longhorn will contain features that are not in 10.3. The much anticipated metadata savvy Finder and file system aren't part of Panther, for instance. (Despite expectations which only increased when the BeFS guy joined Apple)

Re:Hopefully, Apple will help

[2nd Post!]

Compare vapor to vapor, right?

What will Apple release in 2005 to match Longhorn?

All I know is that as an enduser, Apple is very much meeting my expectations. As an enduser, are users of Windows XP and eventually Longhorn having their expectations met?

How many RPC vulnerabilities a year is that again?

Re:Hopefully, Apple will help

[Anonymous Coward]

That problem at that point was that the people who where writting the code you ran under DOS we're stupid

The people who were writing the code you ran under DOS could probably spell the words "writing" and "were". That means they weren't as stupid as you.

Re:Hopefully, Apple will help

[ComputerSlicer23]

Possibly they could spell better then I can. That has little to do with the discussion at hand. I also misused the first "That" at the beginning too. Hell I wouldn't fit in on Slashdot if I could spell and use proper grammer. As soon as they make a language that compiles as a form of communication, I promise, I'll run all my slashdot posts thru it?

Sorry, I generally don't proofread replies while at work.

Kirby

Re:Hopefully, Apple will help

[ComputerSlicer23]

Uhhh, you better hope an NeXT head never, ever reads that... They make Apple geeks seem like fair weather faithful. :-)

I was not much of a NeXT user (only used it a handful of times). However, their development kit, especially their AppBuilder stuff was incredible (I used that on other platforms). I've heard about the feature set the OS had in 1991 was pretty impressive. I used to work in a room full of people who talked about the wonders that NeXT was at the time. NeXT was an incredible OS from everyone I've heard from (I know at least 3 independent sources of people I trust who say that). It's largest two failings, were interrelated. It was too expensive, and nobody made third party applications for it. It was right there with Amiga and OS/2 in terms of wonderful OS that nobody used.

Re:Hopefully, Apple will help

[Anonymous Coward]

Other then the possibility of the filesystem was slower, DOS should have ran faster.

And other than the possibility that DOS was inefficiently coded ... and other than the possibility that the display driver was less efficient ... and other than...

Still not possible, Kirby?

Re:Hopefully, Apple will help

[ComputerSlicer23]

Uhhh, do you know anything about the guts of DOS? DOS really doesn't do anything. DOS isn't an OS in the technical sense. It's a boot loader, and let's you provide access to BIOS routines. It wasn't DOS that was slow. It was the BIOS routines.

That's why Windows 3.11 could run on it. It's why DOS Extenders ran on it. DOS didn't *DO* anything other then command.com really, and a little bit of filesystem stuff. Once you started running DOS you pretty much had complete control of the hardware. The BIOS did most of the heavy lifting on serial I/O, writing to the screen, reading or writing from the floppy or harddrive. That's why the BIOS had to be reversed engineered before you could make another PC run DOS. The BIOS did all the work. It's also why DOS ran a load more hardware then Linux did (at the time).

You go get a copy of DOS, go get a copy of an old Linux 0.2 kernel. Fire up program that calculates primes. Neither one of them is particularly faster or slower then the other, assuming you used similar compilers. Do that again with Windows, and you'll probably find that Windows runs about 2-5% slower (last time I checked), due to context switching speed and generic overhead of the (GUI updates, and other subsystems you don't have to have on a Linux machine).

Kirby

Re:Hopefully, Apple will help

[tf23]

NeXTStep itself wasn't too expensive, it was the hardware that was overpriced.

Granted, even the hardware was before it's time. But schools had a hard time purchasing it. When they did, they loved the systems (atleast at Ohio State we did).

But it was so expensive, especially when someone would question the purchase against a PC or Mac at that time.

Once NeXTStep for Intel came out, that made things a little easier. But, by then, it was almost too late.

The real test.

[BrookHarty]

KDE might be nice, but try compiling Mozilla with all options, email/irc/etc... Thats what I'd like to see as a benchmark test.

Re:The real test.

[Anonvmous Coward]

"Thats what I'd like to see as a benchmark test."

Not to sound cynical, but is that really all that useful of benchmark? You only compile Mozilla once.

Re:The real test.

[Anonymous Coward]

"You only compile Mozilla once."

You're not a Gentoo user, are you?

Re:The real test.

[Anonvmous Coward]

"You're not a Gentoo user, are you?"

No, I use Windows. I don't have to deal with having to compile stuff.

I realize it's useful in the *nix world (even done a bit of that when I tried Linux out) but we're not talking about time critical stuff here unless you're a developer. (Granted, that could be the killer app for this machine, if the number of developers out there was a lot higher.)

I prefer rendering benchmarks where time really does equate to money, but unfortunately my previous argument about mass consumption pretty much sinks that idea too.

Re:The real test.

[Anonymous Coward]

Don't worry. Your benchmarks are as useless as ours.

Re:The real test.

[2nd Post!]

I dunno, I compile Mozilla about once a week.

There's Camino, there's Mozilla, and there's Firebird, so there's three different builds to play with on the same source tree. Then there's the fact that new bugs get fixed all the time.

I used to do it nightly, maybe a year ago, before Camino was developed.

Re:The real test.

[commodoresloat]

Actually, the real test will come when somebody measures how fast this thing can copy a 17 megabyte file from one folder to another.

Re:The real test.

[ernst_mulder]

I thought you needed a 9600 to copy 17MB files? Now I'm confused. By the way, does the G5 finally do multitasking? :-)

Re:The real test.

[b1t r0t]

How does that test anything other than the hard drive? Let's try it on a 4-drive striped RAID, okay? And are you measuring the time to just "copy" the file, or are you including the time to write the cache back to disk?

17 megabytes? That's so small, I don't think I would even see the copy progress dialog half the time.

Re:The real test.

[Hangman Jim 99]

True. How about a 600 meg file from one drive to another?

Re:The real test.

[Hangman Jim 99]

But, this doesn't test CPU speed of course - but I have to assume the origionator of this thread knew that ;)

Re:The real test.

[cbreaker]

Compiling software involves a lot of CPU heavy computation and drive access. Although you might not compile things on a daily basis, it *is* a decent measure of performance.

If you can't realize that, well, I can't help you...

Re:The real test.

[cant_get_a_good_nick]

He's referring to some dork troll that keeps posting about how (some OS/Hardware config, usually MAcOS) is pitifully slow copying one 17Mb file from one place to another and how his P.O.S. 486 can do it so much faster.

Well duh, 5>4

[ArmorFiend]

G5 is lots faster than P4? This is so obvious to even the most clueless home computer user! I mean, come on, how can a P4 possibly compete with a G5? The G5 is clearly 1 ahead!

Can slashdot please tell us something that's not common knowledge? I mean geez, next the'll be like: Saddam Masterminded 9/11. Well no duh!

place your bets

[BortQ]

So how long before IBM starts selling G5 boxes running linux and openoffice?

Re:place your bets

[CoolMoDee]

I remeber hearing talk about an IBM PPC 970 (g5) workstaion. IBM want's more of the community to help with Linux on PowerPC, so making a semi-afordable linux power box would defiently be a way to get that, espically if it runs linux out of the box. I know I would buy one..as would a lot of other linux geeks.

Re:place your bets

[confused one]

If they do, they'll call it a workstation and charge IBM Unix Workstation (RS6000 or Intellistation Power) like pricing... Can we say base pricing "from $8130"

Buy the Apple.

Re:place your bets

[IM6100]

G5 is an Apple whizzy-word. IBM doesn't use that name to describe their Power PC parts. IBM could come out with an equivalent part and put it in a box, but 'G5' is an Apple marketing thing.

Re:place your bets

[IM6100]

The quality of RS6000 hardware tends to justify the price paid. It's not cheap 'consumer grade' hardware, the market segment that Apple plays in.

RS6000 boxes are made to last. Apple hardware is made to sell the customer another box in two to three years.

Re:place your bets

[BortQ]

I've had my current blue and white macintosh for 4+ years now and it's still going strong. I use it for software development, web, email, and some movie editing.

Of course it's becoming obsolete and I'm drooling over the G5s, but Apple hardware usually lasts a loooong time. I still have a mac SE in the basement that worked fine the last time I plugged it in.

Re:place your bets

[phoenix_rizzen]

IBM will never sell a "G5" box, as G5 is an Apple moniker for any Apple box that uses a PPC970 CPU.

IBM has plans to sell Linux workstations using the PPC970 chip. However, they will never be called "G5".

Re:place your bets

[commodoresloat]

As soon as they pay their linux license fees to SCO.

Re:place your bets

[Anonymous Coward]

My Commodore 64 also worked fine the last time I plugged it in. What's your point?

Re:place your bets

[Anonymous Coward]

Actually, IBM has/does sell G5 and even G6 systems! They just aren't what we think of when
we say G5.

http://www-1.ibm.com/servers/s390/pes/

Re:place your bets

[Anonymous Coward]

My Commodore 64 also worked fine the last time I plugged it in. What's your point?

Hard of reading are you son?

Apple hardware usually lasts a loooong time

Re:place your bets

[fr0dicus]

-5 pedantic guff.

The above post, condensed to 5 words

[AvantLegion]

"This one goes to 5!"

Re:The above post, condensed to 5 words

[Compulawyer]

It's 1 faster! (for when you're in the middle of compiling and need that extra oomph! to get you over the top).

Re:The above post, condensed to 5 words

[b1t r0t]

But does it have Dubly?

Re:The above post, condensed to 5 words

[Mitchell Mebane]

Man, I can't wait for the G11!

in other news:

[croddy]

a gentoo user is claiming incomprehensible performance boosts due to foo! film at eleven!

thanks, /. !!

[Anonymous Coward]

Now I have a new set of unsubstantiated and probably completely made up statistics to further my Apple *and* Gentoo advocacy!

Wait'll my Windows-using loser friend hear about this! As soon as they finish re-installing after SoBig and MSBlaster that is!

What he compiled...

[Lally Singh]

kdebase & kdelibs.

Re:What he compiled...

[Lars T.]

Four times faster than on his P4.

Re:What he compiled...

[You're All Wrong]

Do you trust that guy to be able to accurately report on what he's done? I certainly don't.

Anyway, if you want benchmarks, why aren't you looking at www.spec.org?
Oh yeah, because Apple have had a hissy fit and don't submit any results any more.

YAW.

Re:What he compiled...

[drsmithy]

Four times faster than on his P4.

"P4" covers a *lot* of possible configurations. It's about as helpful as saying it was four times faster than his "G4".

Re:What he compiled...

[Anonymous Coward]

Oh yeah, because Apple have had a hissy fit and don't submit any results any more.

Let me translate.

Oh yeah, because Apple had a hissy fit and doesn't submit results any more.

Re:What he compiled...

[aristotle-dude]

Because Intel used a non-free compiler specifically optimized for the X86 chips with cheating fpu flags to get those scores. Because spec tests and most artificial benchmarks are meaningless and vary each time you run them.

Re:What he compiled...

[be-fan]

SPEC tests aren't articifical benchmarks. This is an official list of the SPEC benchmark programs:

164.gzip 1400 Data compression utility
175.vpr 1400 FPGA circuit placement and routing
176.gcc 1100 C compiler
181.mcf 1800 Minimum cost network flow solver
186.crafty 1000 Chess program
197.parser 1800 Natural language processing
252.eon 1300 Ray tracing
253.perlbmk 1800 Perl
254.gap 1100 Computational group theory
255.vortex 1900 Object Oriented Database
256.bzip2 1500 Data compression utility
300.twolf 3000 Place and route simulator
CFP2000 contains 14 applications (6 Fortran-77, 4 Fortran-90 and 4 C) that are used as benchmarks:
Name Ref Time Remarks
168.wupwise 1600 Quantum chromodynamics
171.swim 3100 Shallow water modeling
172.mgrid 1800 Multi-grid solver in 3D potential field
173.applu 2100 Parabolic/elliptic partial differential
equations
177.mesa 1400 3D Graphics library
178.galgel 2900 Fluid dynamics: analysis of oscillatory instability
179.art 2600 Neural network simulation; adaptive resonance theory
183.equake 1300 Finite element simulation; earthquake modeling
187.facerec 1900 Computer vision: recognizes faces
188.ammp 2200 Computational chemistry
189.lucas 2000 Number theory: primality testing
191.fma3d 2100 Finite element crash simulation
200.sixtrack 1100 Particle accelerator model
301.apsi 2600 Solves problems regarding temperature, wind, velocity and distribution of pollutants

Many of these are real world programs (even onse like AMMP and SixTrack, which you may not be familiar with) and they are all real-world tasks.

Re:What he compiled...

[You're All Wrong]

As a large company with many departments it's perfectly fine to refer to a company in the plural. It's a shame you don't understand the use of an emphatic modal too.

And you did that as AC - sheesh, what a tosser.

YAW.

Processor performance has never been my limit

[MerlynEmrys67]

Long time ago... I learned that the hard disk speed is the most important factor in compile speeds. Once you have LOTS of memory, then get the fastest SCSI hard drive you can get your hands on... forget 7200 RPM IDE drives, think 15K RPM SCSI disks. Every time you have to open a new file for compiling... you have to spin (on average) 1/2 of the disk to get your head positioned... it makes a huge difference

Re:Processor performance has never been my limit

[Anonymous Coward]

not only does the disk have to spin on average 1/2 way around to get the head positioned over the file proper, but it needs to do this potentially multiple times beforehand while searching through inodes just to locate the file!
*gag*

Re:Processor performance has never been my limit

[TheRaven64]

Another major factor is the IDE Vs SCSI question. SCSI supports command reordering, so it will resequence a series of requests so that they are executed in the order which requires the least movement of the head. IDE does not do this. Since compiling usually involves accessing a lot of small files, this can make a huge difference, as can having enough cache on the controller to keep the inode tables in cache.

Re:Processor performance has never been my limit

[Anonymous Coward]

Actually, IBM IDE drives support this ... although not always reliably. :) And there are a couple Maxtor (I think) IDE drives that support this as well, but you never hear about it being advertised.

That is, if you are talking about Tagged Queueing.

Re:Processor performance has never been my limit

[Dominic_Mazzoni]

Long time ago... I learned that the hard disk speed is the most important factor in compile speeds. Once you have LOTS of memory, then get the fastest SCSI hard drive you can get your hands on... forget 7200 RPM IDE drives, think 15K RPM SCSI disks. Every time you have to open a new file for compiling... you have to spin (on average) 1/2 of the disk to get your head positioned... it makes a huge difference

This may have been true once, but not anymore. Processor speed is by far the most important thing affecting compilation speed. This is ESPECIALLY true with C++ (e.g., KDE), which tends to take significantly longer to compile than C.

Disk speed is much more significant for linking. But for a large project, linking only represents a small fraction of the total time to build, so speeding up linking by 2x doesn't speed up the overall build by much.

I'm not saying you shouldn't get fast disks. I'm just saying that in a choice between 2x the processor speed, or 2x the disk speed, you should take the processor speed every time if compiling is your main task you want to optimize.

Re:Processor performance has never been my limit

[b1t r0t]

so it will resequence a series of requests so that they are executed in the order which requires the least movement of the head.

I think Unix-like operating systems have been doing that in the OS itself for, what, two decades now? Oooooh, now the hard drive can do it too, for when you're running crappy Redmondware.

Re:Processor performance has never been my limit

[Anonymous Coward]

Slightly off topic... but I am building a new machine to work with a very large Java project.

What type of machine would offer the best java compilation performace? P4? Athlon XP? Opteron?

Re:Processor performance has never been my limit

[TheRaven64]

You are assuming that the OS actually knows the physical geometry of the drive. Since most IDE drives are run in LBA mode this is not the case, since they are addressed by the controller as a large linear address space.

Re:Well duh, 54

[Anonymous Coward]

Your comment reminded me of this bash.cx quote.

----

fastb - lol apple have the g5 out. its must be slower than the p4 because they have so many more revisions to get through! next up is h1, h2, h3, h4, h5 then i1, i2... up to p1, p2, p3, p4! LOL!

kev[0] -I just read that three times and I think I'm dumber for it.

----

Wow... informative headline

[cubal]

I mean, is that or is that not the vaguest bloody headline you've ever read?

"Linux is fast on a new fast computer"

I mean, come on...

Re:Wow... informative headline

[nacturation]

I mean, is that or is that not the vaguest bloody headline you've ever read?

"Linux is fast on a new fast computer"

I mean, come on...

I guess my submission was rejected... "Gentoo is slow on an old G3"

Re:Wow... informative headline

[IM6100]

If you REALLY insist on outdoing folks, the phrase is: "NetBSD is S.L.O.W. on an SE/30"

But it's damned cool running X on a dinkyscreen Mac.

15 minutes?!

[vasqzr]

Anyone remember how fast the G3 and G4 chips were in SETI compared to Intel chips?

Re:15 minutes?!

[commodoresloat]

yeah -- they didn't find aliens either.

Bullshit

[Anonymous Coward]

This guy is a liar. Look through some of his posting history. He claimed to have downloaded America's Army source, compiled and run it on Linux/PPC.

if it was real

[Raleel]

even if it was real (people at the top said it was only a small subset), I would be skeptical of anything actually running fast on said system. generally, if it compiles fast, it's not very optimized.

compile times don't impress me any more, although they sometimes do reflect overall (disk i/o included) performance.

Re:if it was real

[Tom7]

True. To do a good test, one of the computers would have to be cross-compiling to the other architecture, so that they'd both be running the same code!

About twice as fast.

[jhesse]

...at least when running the spiffo graphics. (for some reason, the mac used *much* less cpu for that)

They were about neck-and-neck without the spiffo graphics, although the mac seemed slightly faster.
(Hard to tell, since they were different clockspeeds *and*
datasets.) Averages here...

This is on P2 and P3 chips. The Celerons were 3-5 times slower because it couldn't keep the data in cache.

Keep in mind that Seti@Home doesn't use Altivec or MMX.

Re:About twice as fast.

[Anonymous Coward]

Spiffo?

Re:About twice as fast.

[Anonymous Coward]

It is an industry term.

Re:About twice as fast.

[Anonymous Coward]

yeah...they dont care about properly optimising
their clients unlike other projects OR about supporting other platforms

Re:About twice as fast.

[FredFnord]

Thank Charlie Fenton for that. He's the programmer that did the Mac port of Seti@Home. And he really, really knows what he's doing, and has been programming on the Mac since before a fair number of Slashdot readers were out of diapers.

-Fred

Re:Getoo won't run on my G5

[Anonymous Coward]

Umm, I don't even think the 2Ghz G5s are shipped in volume yet. Plus linux probably doesn't have drivers for some of the hardware yet (SATA, the coprocessor, etc.). I've done 100 mb file transfers on a G4, and it didn't hurt performance at all, and I've played 4 quicktime movies while running a make -j 16 on a G5 simultaneously without noticing dropped frames or audio stutters - and the compile was going fine. So i'll have to call liar+troll on this one. Go back to your bridge :-)

Re:Getoo won't run on my G5

[siddhartha03]

They're only shipping single processor units I thought...

make vs. ant

[pmz]

This reminds me of someone who said that Ant was the bestest greatest build tool, because it built our whole app in like 15 seconds. Of course, make sucks.

He failed to recognize that it is trival to get make to behave like ant in compiling Java applications by restructuring the rules (pass all the .java files in one command line).

Re:make vs. ant

[FredFnord]

> He failed to recognize that it is trival to get make to behave like ant...

Trivial. What a great word. In this case it means 'if you know exactly how everything should happen, and your make allows arbitrarily long command line expansion (which not all of them do, certainly) and you don't mind something that is, frankly, a hideous hack, then yes, it's... 'trivial.'

-Fred

Predictive compiling in Xcode

[goombah99]

Dont forget the Apple GCC compilier in X-code can do predictive compiling and just in time linking of module not needed to launch. making development builds insanley fast.

and if you are so lucky to work in a room with a lot of these there's the distributed compile via rendevouz.

Re:Getoo won't run on my G5

[Anonymous Coward]

Watch out for that flamebait, buster, because there's a well-known troll on the other end of the line. Besides, anyone who can install Gentoo in 2 minutes (or 20) is full of shit, no matter what system they're on.

Re:First Post, Suckas

[Anonymous Coward]

wtf? this is obviously on topic. it has the word "G5" in it and I used Gentoo, too.

Re:Well duh, 54

[shfted!]

You clueless dolt! The P4 is faster than the G5. P comes after G, and is clearly 9 letters ahead!

ram disks?

[Maelikai]

on smallish projects I've seen huge increases in compile speed when using a ram disk, even if only for the compiled output.

with up to 8 gb of ram I wonder if this will be a more common technique on the G5s...

Re:ram disks?

[MerlynEmrys67]

This used to be a huge win for me 10 years ago... put all of the temporary files onto a RAM disk - order of magnitude speed up.

However I have found what makes a much better case today is to just let the ram sit as a disk cache and let the system determine what the best use is.

It is really fun what you can do with 4 Xeon's and 16 GB memory as well - I loved the compile speeds you got in that system

Re:Getoo won't run on my G5

[Anonymous Coward]

Well duh the iPod and Safari won't work, they aren't compiled for use during a linux install. Stupid troll

Re:Well duh, 54

[worm eater]

Acutally, when you multiply the the P by the 4 you get a much bigger number than muliplying the G by the 5. See, you have to look at all the variables.

Maybe this explains the speeds seen

[Anonymous Coward]

I wonder if its possible, that the apparently fast compiles seen were simply because the G5 had simply compiled the code in the background as a predictive compile while they were editing the ported code??

Re:Maybe this explains the speeds seen

[gaelicwizard]

well as far as i know the predictive compiles in project builder don't run under linux... maybe i'm wrong... ;-)

JP

Re:Maybe this explains the speeds seen

[NitroPye]

Predictive compile in Xcode won't speed up a build of anything. It only speeds up the development turnaround process. Yes it is a xcode feature and has nothing to do with G5s specificaly. A G3 with xcode will do the same thing.

Re:Getoo won't run on my G5

[Anonymous Coward]

I have recently upgraded from a Mac 8600/300 w/64 Megs of RAM to a new G5 dual 2GHz with AGP 8X and PCI-X

Sure you did. So, you have a brand new G5, an 8600, an iPod, and you use Safari. Uh, who's a Mac fanatic? Please stop posting FUD to /. and get back to work. Microsoft isn't paying you for nothing. (Or ARE they?)

Worst Slashdot Article Ever

[Anonymous Coward]

Let's analyze this article:

some Gentoo users have installed Gentoo on their new G5s without any problems whatsoever.

No. Only one entirely unreliable user made outrageous claims including running Gentoo. Not users but user.

while it takes one hour on the fastest P4 available.

The user said "my Pentium 4" without saying anything about what that Pentium 4 was. For all we know it could be an old 1.6Mhz with 128MB.

Pieter Van den Abeele, reported that the machine currently runs at half speed due to fan controlling hardware not yet supported.

He says that it may be possible to get the kernel working and if it did then it would run at half-speed. There is no "machine currently" running it to confirm this and it also proves that the other guy is lying.

The Gentoo team will post benchmarks, and will update installation instructions as soon as possible.

According to Abeele - "As far as I know, *none* of the Gentoo developers that are working on support for the G5." Now I guess that eventually they will benchmark and update installation instructions but it is obviously not on the radar screen right now.

The author of this article and pudge should be whipped for putting this awful article up.

Even faster if...

[anarkhos]

Gee, just think what if gcc was compiled with xlc! /giddy

Not true

[Anonymous Coward]

I call your bluff.

If you've got enough RAM to cover regular operating overhead (40MB for me), the compile (biggest I've seen is QT, taking over 80MB), and the size of the expanded tarball you're compiling (150MB is usually enough) then most of the stuff you have to compile can sit comfortably in the disk cache and you won't have to wait for the disk at all. I see this all the time in my builds because I have 512MB RAM on the main box, when I pull a chip out and the compile size exceeds the disk cache the compile gets slow (OpenOffice is a great example). The factors I've seen as most important in compiling are (in decending order):

CPU Speed / CPU Arch - That's where the WORK goes on. Some archs are faster to compile for too, PPC comes to mind as a real 'easy' target for GCC.

RAM / Disk Cache - So you don't have to wait for the disk

Drive Speed - Really only matters on compiles larger than 1/2 available RAM when compiling from a freshly-unzipped tarball.

Now if you have a severe deficiency in any department you'll feel it, but a fast CPU and gobs of RAM will turn any machine into a decent compile-monster.

If you MAKE="-j2" then disk latency won't really matter nearly as much, the CPU can crank on a second thread while the data is coming from disk, that's why using a parallel nuild on a single-CPU system shows a performance increase.

Knoppix

[Anonymous Coward]

Hey, I can install Knoppix on my 2000+ AMD in two minutes.. wait, that's not installing

Re:Getoo won't run on my G5

[Anonymous Coward]

You god damn lame piece of shit, you totally butchered that troll. You're a fucking failure, go kill yourself now, for you will never amount to anything and are just a waste of good oxygen that other successful people desperately need.