Maybe the author of that article should have done some research first?
http://arstechnica.com/cpu/index.htmlSome nice stuff there, especially the PPC/x86 comparisons. Hell, 4 years ago they were already talking about the whole CISC/RISC thing being a moot point in modern CPUs:
http://www.arstechnica.com/cpu/4q99/risc-cisc/rvc-1.htmlJust about everything that people have brought up in this thread (pipeline comparisons, CPU design methodologies, CISC vs. RISC, PPC vs. x86, in-depth analysis of the PPC970, etc.) has been gone over, in detail, at that site I linked to. And heck, they even have links to their sources that they pulled the information from.
As for the author's assertion that an Alpha beats the P4 in floating point operations, did the author mean SpecFP2000 Peak:
http://www.aceshardware.com/SPECmine/index.jsp?b=2&s=0&v=1&if=0&cf=0&r1f=0&r2f=0&m1f=0&m2f=3&o=0&o=1...or SpecFP2000 Base:
http://www.aceshardware.com/SPECmine/index.jsp?b=2&s=1&v=1&if=0&cf=0&r1f=0&r2f=0&m1f=0&m2f=3&o=0&o=1It's interesting that the author originally wrote the article to justify to the management and their customers the choice of PPC over x86. I'm going to guess that PegasOS is trying to position itself to kiosks, embedded systems, the hobbyist market, and maybe business terminals. I make this guess based on:
http://www.pegasosppc.com/tech_specs.phpNow for his PPC vs. x86 article he mentioned Alpha. Alpha is fast, Alpha is neat, Alpha is expensive and not the product that he's trying to compare to x86. It's a totally worthless mention. I'd hate to be a customer who asks "why do you use PPC instead of x86?" and have someone come back with, "did you know that the Alpha CPU is faster than x86?" What's up with that?
Yappin' about power usage, etc. Talking about how the P4 is power-inefficient and then using the PM as a comparison--despite the fact that the PM uses a different architecture and pipeline than the P4. And power consumption, 'ell, it's not only the PM's that is low, there are other low-cost competitors who are also low-power:
http://www.via.com.tw/en/Digital%20Library/PR030708Antaur.jspHeck, if the author wanted to actually research power consumption:
http://www.sandpile.org/impl/pm.htmhttp://www.sandpile.org/impl/p4.htmhttp://www.sandpile.org/impl/k8.htmhttp://www.sandpile.org/impl/k7.htmhttp://www.sandpile.org/impl/crusoe.htmI'm too lazy to look up any PPC power consumption figures.
The sad part is that the author DID list Ars-Technica as a reference. I guess he skimmed the articles or something.
The even sadder part is that his business justification of why they chose PPC over x86 lists memory bandwidth as a "problem" for fast x86 CPUs. That's funny. I'm sure the P4 with it's 800 MHz (effective) FSB and dual-channel DDR400 memory subsystem is at a great disadvantage to the G4's 166 MHz CPU interface. Yup, you see, it works out to be more efficient because the G4 only scales to around half the speed of the P4. Heh, yeah, sure.
I also like how the author keeps referring to the Alpha, a dead-end product. Yeah, the 21364 is out now but HP's already announced that that platform has no future. Nor does PA-RISC. Looks like some of the old-school CPU manufacturers are ditching their "RISC" designs, eh? But that's ok, I mean, more and more computing institutions use commodity-level "CISC" CPUs like x86:
http://www.top500.org/list/2003/06/http://www.top500.org/lists/2003/06/trends.php"A total of 119 system, up from 56 six month ago, are now using Intel processors."
"149 systems are now labeld as cluster, up from 93."
Now I know many Itanium systems are there, but look at the increase in clusters. It doesn't take a rocket scientist to figure out that people are clustering cheap x86 CPUs more and more to get the best computing bang for the buck, supercomputer-wise. Someone else can pull up stats from all 500 installations to either support or disprove my assertion, I'm not quite curious enough to do it myself.
The author does finally admit that the PPC is crippled in terms of memory bandwidth and just plain slower than x86. But then he tries to minimize it by saying that people won't generally notice.
This piece has some facts in it, some bias, and is utter crap if it were originally written as a way to woo customers or provide business justification. The article touches on performance computing of which PegasOS's products are not. It talks about low power consumption, ok, there's a good fit. It tries to make the argument that Linux will equalize all issues and yet there are many many programs that will (and do) have endian issues since they were first developed on x86 and without regard for PPC/68k/Alpha/PA-RISC/MIPS. It's not like there is some magic wand that makes everything in Linux just magically work on all architectures.
Again, as a business justification, this piece tends to spend a large amount of time talking about Alpha and future, unreleased products like the POWER5 and POWER6. If I were a customer, would I not want to know why the PegasOS was the right choice for me NOW? With the CURRENT PRODUCTS? Or are people supposed to buy the current under-performing, bandwidth-crippled systems based on promises that a CPU due in 1 or 2 more years will be better?
The PPC has many strengths. I think many of them play into the PegasOS. However, this article doesn't expose them and spends too much time trying to minimalize the benefits of the main competitor, x86. The comparisons wildly fly to any CPU the author can imagine and somehow comparing a P4 to the Alpha is a reason why the PPC is better. And comparing the P4 to an unreleased CPU (G5) is a reason why PPC consumes less power/generates less heat. Whatever.
If I have enough money when the time comes, I'm gonna get me a PowerMac G5. And if AmigaOS 4.0 ever ships, I'll eventually get an AmigaOne to play around with. I'm no slavish Intel fanboy, honest. However, I do think this article was full of bias and poorly written. The guy might be a good EE but I don't agree with his opinions and I don't think he produced a good quality comparison or business justification, whichever this article was supposed to be.
