PowerPC Manufacturer ?

[xboxrulz1405241485 0]

I was wondering, are there any computer builders who build computers w/ IBM PowerPC 970FX processors (G5)?I really desperate for a computer that runs on PowerPC, I dun like the x86 architecture... I want to run OpenSUSE 10 beta 3 on it.Please don't tell me Apple is the only ones (for now)xboxrulz

[cyph3r 0]

I guess u mite wana have a look at this site. I was recently searching for stuff regarding PowerPC vs Linux cause i have a PowerPC Mac and it runs on a G3, But i wanted to turn it into a complete linux dependent system.

>> LINK <<

>> LINK 2 << I found this link to be helpful.

Cheers

[Jguy101 0]

I don't think you can buy any PowerPCs to build your own PC. However, some people have figured out how to install Linux on Xboxes (clicketh), and they'll probably figure out how to do it on Xbox 360s, which will have PowerPCs. Otherwise iff, like me, you REALLY don't like x86, go with an AMD Athlon, which uses AMD64 architecture. Very nice.

[xboxrulz1405241485 0]

umm.. AMD Athlons are x86 processors.xboxrulz

[the empty calorie 0]

Hahaha....

Anyways, Yes, there actually is a way to build your own G3/G4 based PowerPC machine. There's a company called Genesi which makes an ATX motherboard for use with either a G3 or G4 processor, and It also comes with installation disks for MorphOS and Debian GNU/Linux PPC. It seems to be more intended for people wanting to build their own Amiga-like computer, but does welll with GNU/Linux as well. I'm looking into getting one of these for myself, although the price is a little steeper than going with an x86 setup.

Oh, and here is the website:
https://genesi.company/

[Jguy101 0]

No, not the Athlons they're currently making, the Athlon64s. They all use the AMD64 architecture, although x86 stuff will still run on them. So, you can run an x86 OS on them, but it'd be better to a version optimized to take advantage of its 64-bit architecture.

[EdgeXC 0]

I dont think the IBM PowerPC 970FX processors is the wave of the future. IBM is behind in Intel and AMD processors. Intel and AMD are both coming out with new dual core processors. IBM is basically the average processor for basic business usage. Leave the gamming and other things to AMD. Intel would be more of Microsoft Office and things.

[the empty calorie 0]

Keep in mind, IBM and Intel/AMD, make completely different types of processors. I personally prefer IBM's offerings. IBM is behind? IBM's POWER architecture is much further ahead of Intel's...Keep in mind that IBM is known for making thew world's most powerful computers. In fact, the current recordholder IS an IBM machine, using POWER architecture. Intel offers nothing that even comes close. One thing you may be right about, the 970 series may not be their best product to go after (although I definetely wouldn't mind getting my hands on a 970 or two myself...), but don't forget what IBM is developing right now...Cell Architecture, which will blow the doors off of anything Intel has to offer. It's not like Intel is making major breakthroughs with putting multiple cores on a chip, it's been done for quite a while now. Any chip manufacturer can do that, and have been able to for the past ten years. And as far as IBM is behind Intel and AMD? Let's not mention that x86 architecture has been around since the seventies...Or that Microsuck has phased out Intel chips for the new upcoming Xbox 360, in favour of PowerPC chips... Intel has been too comfortable with x86 for so long to abandon it, and every other chip manufacturer took a big step ahead of Intel back in the early to mid nineties...Cell is coming....you better watch your step, kid.(I <3 IBM)

[kam1405241509 0]

Keep in mind, IBM and Intel/AMD, make completely different types of processors. I personally prefer IBM's offerings. IBM is behind? IBM's POWER architecture is much further ahead of Intel's...Keep in mind that IBM is known for making thew world's most powerful computers. In fact, the current recordholder IS an IBM machine, using POWER architecture. Intel offers nothing that even comes close. One thing you may be right about, the 970 series may not be their best product to go after (although I definetely wouldn't mind getting my hands on a 970 or two myself...), but don't forget what IBM is developing right now...Cell Architecture, which will blow the doors off of anything Intel has to offer. It's not like Intel is making major breakthroughs with putting multiple cores on a chip, it's been done for quite a while now. Any chip manufacturer can do that, and have been able to for the past ten years. And as far as IBM is behind Intel and AMD? Let's not mention that x86 architecture has been around since the seventies...

 

Or that Microsuck has phased out Intel chips for the new upcoming Xbox 360, in favour of PowerPC chips... Intel has been too comfortable with x86 for so long to abandon it, and every other chip manufacturer took a big step ahead of Intel back in the early to mid nineties...

 

Cell is coming....you better watch your step, kid.

 

(I <3 IBM)

1064321522[/snapback]

I'm really interested in this too ... but I came to a different conclusion ... I'm happy to be proven wrong though .

 

1. Supercomputers are about interconnecting 1000s of CPUs

Liking a chip architecture just because you like a supercomputer or mainframe doesn't correlate well. The latter are all about fast, custom interconnects between high-FP performing CPUs ... it's not just the CPUs themselves. Getting a bunch of Cell chips connected on a gigE LAN isn't going to match a supercomp .. no matter what Sony's marketing dept says! I'll show you what I mean: netlib has some benches on specific CPU architectures for linear algebra computations (LINPACK) ... the ones at the top were NEC's awesome vector processor scored 2177 Mflops (n=100), 3.6GHz Xeon 1MB = 1821, 1.9MHz POWER5 scores 1776, 1.6GHz Itanium2 = 1765, 2.2GHz PowerPC 970 = 1681, 2.6GHz opteron 852 = 1593, 3.2GHz Nocona Xeon EM64T = 1593, IBM eServer pSeries 690 1.7GHz = 1462, IBM IntelliStation POWER 275 1450MHz (POWER4+) = 1245, IBM eServer pSeries 630 6E4 1.45GHz = 1229, Cray T94 (4CPU) = 1129. All these are for the SINGLE CPU scores! Of course individual CPU scores are meaningless in supercomputers ... what matters is how several 1000 of them compute the linear algebra problem! Also, pricing is really important to me. It doesn't matter that Itanic benches 250MFLOPS faster than the faster Opteron if it ends up costing me more than twice as much .. plus the extra air conditioning etc required! I have seen a guy with an Itanium desktop, BTW, but that was before AMD starting to win 64-bit marketshare!!

 

I'm not sure where you get your stats about IBM from but, according to top500's latest report, 66.6% of the top500 supercomputers in the world are Intel based & 5% are AMD! 10.4% are Power & 5% PPC. Specifically, 35% are P4-Xeon/15.2% EM64T, 15.8% Itanium2, 5% Opteron, 4.4% Power4+, 3.2% PPC440, 3% Power4, 1.8% PPC, 1.6% Power5, 1.4% Power ... However the top 2 machines are IBM Blue Gene machines (one of them is at IBM T.J.Watson .. not really a "sale" ). #3 is a Itanic2 machine. Earth Simulator is now #4, and the Virginia Tech MacOSX cluster is now #14. There was an article at the end of last year stating about Opteron "winning three of the top 20 spots: the Shanghai Supercomputer Center's Dawning 4000A at #10, the Los Alamos National Laboratory's Lightning at #11 and the Grid Technology Research Center's Super Cluster P-32 at #19" .. things have changed now but will obviously keep changing again.

 

If you stick 1000s of POWER chips together, I'm sure you'll get a great supercomp, but I can't see how FP performance improves that much when we're talking about the same number .. I'm guessing you're talking about may be a couple of CPUs in a cc-NUMA or SMP workstation .. or maybe a cluster (please correct me if I've assumed wrong, BTW).

 

2. X86 is less of an issue these days

The x86 architecture is ugly compared to RISC designs, no doubt about it, but these days RISC chips have CISC add-ons, and vice versa (SSE3 etc)! Also, age is sometimes a good thing .. they've had lots of time to try to fix some things ... at last HW partitioning is doable . Intel did try to abandon x86, with Itanic/IA64 ... unfortunately v1 cost a bomb without enough of a performance increase to x86 apps!! It was aimed at FP performance (it did well in this, see above), and for being a proprietary UNIX system killer, but the developers ignored the masses and the masses of software written to that (and the fact that closed source software won't simply be recompiled overnight, as it were). Mass production reduces costs. X86 is open to all willing to compete, and hence we have the nice current situation of AMD & Intel pushing each other to improve. AMD came in understanding this and took over this slot, giving 64-bit memory addressing & integrated-memory controller scalability benefits without throwing the baby out with the bathwater.

 

Personally, I couldn't care less what architecture my main PC is, since the majority of my important code (i.e. for work) is my own and can be compiled to whatever happens to be i. fast & ii. cheap . The next most important other stuff are commercial x86 binaries for NT/Linux (some maths packages). The rest is opensource apps or win32/win64 on x86/x86-64 games (not as important since I'm happy to have a separate machine for those). It would be nice to have the good old days where there was tons of competition, I guess , but not just for the sake of it. I want there to be bang for buck. If I can get a massively popular (and therefore potentially cheap) x86 CPU that has good FP performance (like the Opteron in the above benches), then I'd go for that rather than a POWER/Alpha/etc that's similar in performance but with a slightly higher price tag .. unless there's a massive improvement (I'll get on to Cell in a second!!). Also the fact that I can run my CAD apps helps. BTW, did you know that IBM's official CAD app actually only runs natively on WIN32 ... for POWER it actually runs a WINE-like library layer ... that was quite a depressing shocker!! They advise you to run NT . But that's another matter!

 

Mind you, soon even this will be irrelevant! The instruction set is just one part of the equation. Look at what the now nearly dead Transmeta (amongst others .. there was this Russian design, plus several software companies are doing research similar to this) managed to do ... x86 code could be compiled into a very parallel VLIW underneath without losing that much in efficiency (if they had the manufacturing links they could've done better ... intel looks like it is going to pick up the gauntlet and go down this direction, taking aim at their own StrongARM CPUs according to the latest IDF because they don't want to pay ARM royalties anymore ... I actually love ARM chips, but that's another matter!). Intel claim they will soon bring out an ARM-like efficient chip that's x86-compatible .. or at least that's their aim .. I'm not entirely sure that's really possible, but anyway!!

 

3. IBM/AMD & Sun/AMD, Apple/intel, Cell

IBM itself isn't a good reason to vote for PPC, because IBM and AMD partner around chip fabrication, although IBM are now aligning themselves to Intel instead. Also, Sun & the startup they took over are helping AMD design very large-scale 64-bit x86 systems. Apple have given up on PPC and gone to x86 because the G5 hasn't had the expected speed bump for ages. IBM are focused on Cell, and Cell for PPC apps is damn slow.

 

I was really looking forward to Cell. It promised a focus back on pure uncompromised performance ala Seymour Cray & the DEC Alpha guys. But I've recently been reading a lot of negative press from developers using both PS3 & XB360 dev kits, and these are the 1st guys programming the machines. The former is basically a Cell workstation, and the latter is a modified Mac running at x% final performance (x being a fraction, but they can figure out final performance from that so it's not an issue). In both cases the developers are very disappointed .. so much so that many were prepared to publically complain. The main downer claimed is the limited amount of cache available to each of the Cell's VPUs, and the poor real-world performance of the multitude of VPUs given that they can't get enough data into them with such a puny amount of cache, and ditto for the low-clocked PPCs given that they weren't designed to do much! They were hoping to have a leap in CPU performance for AI/physics ... but programming multiple CPUs effectively is extremely difficult. Carmack's SMP quake could sometime actually run slower! Sure, you can say the developers have to figure it out .. I think it'll be a very long time in the making! Imagine using a PC with quad PII's rather than one fast P4 for gaming!! A whole load of algorithms are going to have to be figured out and rewritten in parallel with no guarantee of being any faster once the inter-process comms overhead is accounted for.

 

The reason these consoles went for PPC is because IBM enticed them with THEORETICAL performance-per-buck. This is great for marketing etc. Yes, there were demos of rubber duckies in bathtubs, and multi-stream HDTV decoding ... but in terms of actual game footable, Sony didn't show much .. and had even lied about some of the footage (later found to be prerenders)! A PC with PPU will beat probably them. The same thing happened in the previous generation, with Sony claiming supercomputer performance ... I still don't see games that look as good as those demos! If developers can't keep those parallel cores filled with the approriate data, performance will suffer. A good analogy is the P4's long pipeline that has a whole load of smart compiler tech to decide which data should be loaded next before it's needed ... predicting the future is never easy!

 

So right now I'm not too sure about Cell. I do hope these complaints are just initial problems .. since I'd absolutely love a cheap supercomputer on a card for maths .. I could then avoid going to my lab so often . But I have a feeling these guys are right .. they know what they're doing & wouldn't simply complain against MS/Sony just for the hell of it! They are putting their neck out doing so!

 

BTW, have you guys seen the stuff on Sun's Niagra? I'm looking forward to it .. there was a cool article in IEEE Spectrum magzine a few months back. If it's really as good as they are saying (one of their research prototype chips can supposedly match a 16-way Opteron on FP tasks, and they are talking about placing 128 of the cores on a single chip ... but maybe that'd be 2010!!). As with all things these days, I'll wait to see the real-world benches before making any final judgements! If there are any other cool CPUs anyone knows of, please put up a list on discussion. Others I know of are things like ClearSpeed etc ... but they are so damn bleeding expensive, I can't take them seriously. If you have a limited space requirement for supercomputing performance, then it's one of only a few similar companies doing this!

 

I'm not a pro-anything person. I have intel & AMD boxen, and nearly bought a G4 on a PCI card (didn't due to the poor performance of the cheap ones I could afford/justify!). I also (have to!!) use several OS's incl Windows, Linux & Solaris .. and even OSX has a few apps I really like (compiled to PPC .. I use my labs G4 notebook every so often!). It would be nice to get a definitive answer from someone who's used the above processors in serious workstation/cluster setups.

Kam.