Originally Published: Monday, 13 November 2000 Author: Ross Sanders
Published to: enhance_articles_hardware/Hardware Articles Page: 1/1 - [Printable]

A Different Approach: Part II of V

Ross' last article for Linux.com chronicled the beginning of his adventure to find an appropriate RISC-based system to run Linux. Learn what happened after last week's adventure, and find out his plans to meet the goal of finding his ideal RISC-based machine!

   Page 1 of 1  

To all the people who posted and e-mailed advice about my first article in this series: a hearty thanks. It's nice knowing we have readers of diverse tastes about computer systems.

A re-cap of my requirements: at least 3 free PCI slots, RS-232 support, a good quality 3D video card, solid, proven performance from the CPU, full support from a Linux distribution, and a low-enough price that I don't have to form an IPO just to buy one. In Apple's case, this leaves out the PMac Cube and any laptops.

At first look, the hype about the PowerMAC G4 is extremely impressive. Apple says you can have a supercomputer sitting on your desk computing your data, be it rendering a movie or orbital calculations at a full gigaflop or more (that's 1 billion floating point calculations per second for those who don't know that term). Unfortunately, that's only equivalent to supercomputing technology from when I was still playing gold box Dungeons and Dragons games on my old Commodore 64. Supercomputers of today routinely do teraflops, but who can afford an IBM Deep Blue in their closet?

There is a PDF file on the Apple Web site that explains vector processing and supposedly how they reached the numbers for their performance calculations.

G4 Velocity Engine PDF

I personally found the numbers explanations woefully inadequate, since there's no direct mention of the configuration used or of the actual algorithms being processed. For a "distinguished scientist," the author did a bad job of presenting results and their origin. I tossed the printout of the PDF file into file 13, believing it more marketing hype than solid information. Somehow, I doubt they were using the $1,800US basic system for those benchmarks.

Searching out good benchmarks for the G4 systems has proved difficult not only on Apple's site, but elsewhere across the Internet. Apple's "real life" benching with Photoshop didn't carry much weight with me. I don't use Photoshop and most likely never will. What I needed was more information.

I found a set of SPEC marks at ugeek.com comparing a 450 Mhz G4 with several x86 variants. I was starting to become less impressed of the performance of the G4 systems from Apple. I don't know how benchmarks were compiled. Of course, configuration is everything (as is optimization) when comparing computer systems. SPEC benchmarks don't take advantage of on-board vector processors all the PowerMac G4s have. No software will take advantage of it unless it's specifically coded and optimized for the vector processor just like MMX or 3DNow. The CPU by itself without the Velocity engine performance gains really isn't all that impressive. How much more power can you get out of the processor with software that can take advantage of that part of the computer? Unfortunately, I've found no benchmarks that tested WITH the vector optimizations that were not done by Apple. I remember the differences with vector processors my college professors used in the bigger computer systems. They could speed computations sometimes by a factor of 3 or more if utilized properly. Naturally, I cannot compare an IBM mainframe's results (or a Cray for that matter) with those of a G4 simply because it is very bad form to compare a computer of today with one from 10 years ago (Apple might want to take note of this!). Today's vector processors are so different and the way each individual program utilizes them (if at all) varies so much that you can't compare them. On the same computer with 2 different applications you can get speeds wildly different. I will have to carefully consider this "apples and oranges" problem more if I want a PowerMAC G4 on my desk. Technology comparisons should be current with their peers or not at all. Current technology produces terraflop computers that deserve the name "super" computers, Apple claims to sell a supercomputer for a student's price. That's hogwash and marketing hype.

If I buy the PowerMac, the very first thing to go will be the ATI Rage card in favor of something faster and better supported in Linux, possibly a GeForce or a 3Dfx. There are many options. Either way, I want better than what ATI has to offer.

I also mentioned the problem of needing an RS-232 port. PowerMacs have none, which is a drawback. I'm aware of USB to RS-232 adapters, but why would you kill a fly with a sledge hammer? I consider it a minor strike against the Macs.

So let's re-cap. What do we have in the PowerMAC G4 systems? Gigaflop to 4 gigaflop performance claimed by Apple which is what a supercomputer was about 10 years ago (and these numbers are questionable without finding hard data). SPECfp95 at 20.4 and SPECint95 at 21.4 which are computed without the benefit of the vector processor and are unimpressive. A vector processor built in with really as yet unproven performance outside of the marketing hype. Yellow Dog Linux claims to have support for the Velocity engine in some of its software so support is there if limited. A video card that will be replaced as soon as the system arrives. And, no native legacy RS-232 support.

Add to this, three open PCI slots and one 2x AGP slot which is admittedly old news since newer AGP motherboards in the PC world use 4x. Expandability is there for sure. It's RISC but with a weak processor that can't compete without the Velocity engine.

This week's homework for the performance hogs out there: help me find decent CPU benchmarks/true accounting of the G4 capabilities with the vector processor under Linux (not using MacOS). What I'm specifically seeking is hard information that explains the configuration used (especially if the vector processor is used), the operating system used (Linux here naturally), and the results. If I can find more information I'll bring a follow up on what I've discovered to you and paint a more complete story minus the hype.

Next week's topic is the venerable but popular Alpha processor. Around much longer than the PPC CPUs , Alpha has data is more readily available with less hype to muddle through. Come back as I walk through these systems that thoroughly deserve the reputation they have gained as the workhorses of the scientific community. At the end of this series, I will tell you my choice for my next computer and why I picked it.


Ross is still out of his mind, leave messages at the (BEEP.)




   Page 1 of 1