Hello,

I'm looking for a PowerPC-based system as I really like some of its features like AltiVec, OpenFirmware and above all low power usage.

Looking for something in my price range I found only used Apple products and I don't believe that Mac Pros will have a low energy consumption.
I was unable to find a place to obtain a G4-Pegasos (as I read they are not produced anymore and there are problems due to lead and the RoHS certification) and these PWRficient processors still seem to be far away from affordable real world usage.

So I'm "stuck" with the Efika, but I'm not sure if it will be sufficient for my applications. Most of the time I don't need a lot of processing power, as I am only coding, surfing and listening to music and I'm pretty sure that the Efika will be powerful enough for these things. But compilation is an entirely different thing and I tend to compile a lot to test changes in my code. I'm rather impatient, but my current P-M 1.86GHz/1GB system already limits my work flow and my Athlon Classic 650MHz/256MB is way too slow for lots of testing.

I've read about Efika2, but it seems to be based on the same 400MHz CPU. Are there any plans for an Efika with a faster CPU and more RAM? I believe that 1.2GHz/512MB RAM should be OK.
Which other (affordable) PowerPC-based systems are planned in the next time? (I've read about an AMCC processor as well - any plans for consumer products?)
Although I was not able to find any used POWER5-system: How is their power consumption?

kind regards

Which compiler are you using? As simple as it sounds, you can speed up compiling by using a faster compiler.

If you're coding in C or C++ on X86-based PC's you might try using the free OpenWatcom compiler from http://www.openwatcom.org -- it's brutally fast on modern CPUs! :) (and it's able to compile for MS-DOS, Win 3.x, Win 9x, NT, XP, Vista, OS/2, Netware and Linux)

On PowerPC-based systems, there might be fast compilers for C/C++ also (on Apple, I think Metrowerks Codewarrior has a good rep).

There might be other compilers worth looking at. Like the Comeau C++ compiler, or parallelized commercial compilers (which are expensive).

I use the GNU compiler. I haven't tried any other compiler, but I'm not sure the code base (most of the code is not written by me) will be compatible with other compilers, nor have I tested any other compilers. I'm only hobbyist after all.

But in general the compilation speed is pretty fast, it's only that I use a clean copy of the source tree for compilation testing before commiting any changes and thus recompile everything. (Bad experience of the past.) I don't mind compile times of less than 5 minutes, but if compilation takes 30 min, it is just too long.

Compiling the Debian Linux kernel on my EFIKA box with GCC 4.1.2 took over 3 hours. Perhaps MorphOS 2.0 will run GCC faster, but generally, hard number crunching is where CPU speed matters (and with compilers, I/O speed). And then it becomes important what compiler and OS you're using.

You might try an older version of GCC, 2.95.x for instance, which is said to be much faster than the 3.x or 4.x versions, but it's not ISO compliant.

Yeah, a good, free, lightning-fast compiler for PPC would be great!

On the Amiga, I had the SAS/C compiler which was able to run from memory and cache huge amounts of include files. I wish something like that existed on modern hardware. (OpenWatcom does alleviate the problem somewhat; it's really fast, but runs only on x86 currently)

I'm a compiler writer, but I don't have much spare time currently; I will look into it as time permits.

You can use distcc or icecream to do distributed cross compiling on all the x86 machines you have available.

We are using icecream and a 15 node compiler farm (most p4 celerons@2.5ghz and a few amd64) to (cross)compile a lot of stuff including Openembedded for efika and it saves a lot of time.

Once we even added our efika to the build cluster, but it got very few build jobs mainly because icecream is calculating the "power" of each machine and tries to send the build jobs to the fast one first.

An efika based build cluster with would work fine assuming you have so many efika's around :)

G5 and POWER5 are insane power-hungry monsters.

There is a reason why the PowerMacs had water cooling..

If all you need are fast compile times you need a Core2Duo or
the new Q6600 Core2Quad. You will get very low relative power-
consumption with those processors. 65W maximum on four cores..

There are today no comparable (and available) alternatives in
the PowerPC-world.

If you want PowerPC, I would recommend a dual G4 PowerMac
or a Pegasos 2. Both second-hand.

Thanks a lot for your answers.

I'll probably go for a efika and a (cheap and pretty energy efficient) AMD dualcore x86_64 box for development.

Since I've given up on playing computer games the only reason for a fast CPU is compiling, but being a hobbyist I don't mind waiting five minutes instead of 2 when I can save a couple of hundred Euros. Or I'll wait for Barcelona.

What PowerPC based systems can I await in the future? The P.A. Semi system, some Italian 566MHz system called Samantha and Genesi's Efika2?
If so, then going for an Efika now seems like a good choice to me.

Not really.

It should be noted that the Debian kernel build does not take so long because of compiling kernels - certainly, if you configure your kernel right, cut down on the modules and enabled features, and don't wrap it in a Debian package script solution or sandbox (like Gentoo genkernel or doing it by hand), the most I have ever waited for a kernel is 25 minutes on the Efika. Usually it takes 14 minutes (and 32 seconds, so says 'time').

The Debian make-kpkg script, Perl, rules, packaging and archiving all takes a millennia to run and optimizing the compiler doesn't make a bit of difference to it.

_________________
Matt Sealey

@Neko: All I did was typing "make" after setting the kernel configuration. That took over 3 hours. No idea what was happening behind the 'make' command. The second time around, when I configured in the NVidia support, making the kernel took only a couple of minutes. I don't think I will have to compile the kernel again. My optimizing the compiler remark was meant generally, not particularly for Linux. I won't touch the GCC source! lol