Author Topic: G5 Chip Article (Read 3841 times)

kelly · « **on:** July 01, 2003, 08:23:23 AM »

Kind of Geeky.

Why the G5 Runs So Fast

http://new.lowendmac.com/myturn/03/0701.html

Mrious_be · « **Reply #1 on:** July 01, 2003, 10:14:20 AM »

Intersting, although it was abit TO for me

so i stopped reading after a little.
Let's just say that Apple have an awesome processor at it's hands now

krissel · « **Reply #2 on:** July 01, 2003, 11:46:14 PM »

Much of what that article tries to put forth is detailed in this one at Apple's site.

If you can slog through it you will understand what they are saying about the two FPUs that have the ability to do more math per unit. But I also ran across some caveats that make me think that to fully take advantage of the speed of the G5 we will have to upgrade many of our applications to ones rewritten for the G5's different architecture.

Here are a few things to note:

*the larger cache line size (128 vs 32) will create a bubble effect with programs written to take advantage of the smaller size
*longer latency to memory
*increased latency to G5 Velocity engine pipeline
*no L3 cache (programs that were optimized to it will have to be reworked)
*small chunks of data are now inefficient and large chunks will have to be written for the L2 cache

I'm wondering how well older programs will do on the new hardware. I'm sure Apple chose the most updated of programs to test the new machines.

One other thing I found interesting on that page:

QUOTE

A new OSX API has also been proposed that will provide a superset of the existing Gestalt, sysctl, and _cpu_capabilities functionality. This API will be named OSSystemInfo and is scheduled to appear in Smeagol and Panther.

Smeagol

sandbox · « **Reply #3 on:** July 02, 2003, 02:42:44 AM »

smeagol

Aliases: Gollum, Trahald, Slinker, Stinker
Date of Birth: 2430 T.A
Race: Hobbit of the Stoor strain
Height: between 3-4 feet, when not hunched over
Date of Death: 3019 T.A.
Alignment: Evil

Smeagol / Gollumn transform

Technically Speaking

QUOTE

This build, code-named Smeagol, will run on the new chip but won't take advantage of many of its key features, including 64-bit support. Sources said Apple's goal for Smeagol is to deliver Mac OS X performance at least "on par" with what Jaguar could achieve on Motorola G4 chips running at the same speed; the move will allow Apple to ship the new hardware before Mac OS X 10.3, a k a Panther, can take advantage of all the new processor's capabilities.

giantmike · « **Reply #4 on:** July 02, 2003, 08:42:25 AM »

You never write a program to use cache, the hardware does it by itself. Therefore, most of those points are null.

As for the increased latency, I could see that on the velocity engine, as that was added on at the end of the 970s production stage. However, increased latency to RAM makes no sense, as there is no L3 cache, which should shorten the time looking through cache, and the frontside bus and memory itself is so much faster.

As with any CPU, specific optimizations are needed if you want to get the most out of every clock cycle. However, with updated compilers, most programmers should just have to run their code through a new compiler, with the correct flags set, and they will get 90% of the optimizations for the G5. The remaining 10% would take hours of work, and in most cases, not be worth it.

Even without recompiles of the program, the G5s will still be wicked fast on all of the programs that run on today's G3 and G4 computers.

krissel · « **Reply #5 on:** July 02, 2003, 10:23:06 PM »

Some quotes from the article that led me to question the performance of present applications:

QUOTE

DCBZ in the G5 has three detrimental effects on the performance of existing programs...

QUOTE

As the CPU frequency has increased at a much faster rate than the memory frequency, the relative time to access memory has increased in the G5 vs. the G4.

QUOTE

Avoid small data accesses. Due to the increased latency to memory, the longer cache lines, and the nature of the CPU-to-memory bus, small data accesses should be avoided if possible. The entire system architecture has been designed to optimize the transfer of large amounts of data (i. e. maximize system memory throughput). As a side effect, the cost to handle small accesses can be very high and is quite inefficient. If possible, allocate data in large chunks to better amortize the overhead to access memory. Adjust to the smaller cache. High-performance programs that have tuned themselves for the presence of an L3 cache will need to be re-worked to fit in the (now larger) 512 KB L2 cache.

I realize this article is to developers to warn them about the changes they should consider in new code. I imagined a long pronounced rewrite process necessitating new upgrades and consequent costs to the consumer. But I wasn't aware that a simple recompiling of already existing apps would basically eliminate or minimize these problems.

That's good to know.

() · « **Reply #6 on:** July 07, 2003, 11:48:03 PM »

Thanks for the information, very nice to read about the processors on the G4 and G5 as well as the P4.

i am on a imac G4 and was wondering if I need to upgrade to a G5, but would rather see if there will be a G5 chip to have installed on my G4.

If you have any info about upgrading G4's to a G5 let me know.

Luv ya!

sandbox · « **Reply #7 on:** July 10, 2003, 01:34:40 AM »

a bit more on the topic.......

Mac OS X 10.3 Panther will not be a 64-bit OS
By Tony Smith
@ The Register

kelly · « **Reply #8 on:** July 10, 2003, 09:34:43 AM »

Very nice Dan Knight Article.

The Power of the G5 and the Future of the Mac

http://new.lowendmac.com/musings/03/0710.html

Bill · « **Reply #9 on:** July 10, 2003, 11:33:33 AM »

QUOTE(sandbox @ Jul 9 2003, 11:34 PM)

a bit more on the topic.......

Mac OS X 10.3 Panther will not be a 64-bit OS
By Tony Smith
@ The Register

Got fleas hatching on the right coast SB?

With the heat on the left, they're as big as dogs. <gr>

kelly · « **Reply #10 on:** July 14, 2003, 07:58:16 PM »

Another interesting Article.

Things I didn't know. Of course that covers a lot.

"... a dual, unidirectional bus.".

http://www.macopinion.com/columns/engine/0...7/10/index.html

They need to get these out to see what they're really like.

Bill · « **Reply #11 on:** July 14, 2003, 08:18:42 PM »

"They need to get these out to see what they're really like.

"
Agreed.

Here's one on opinionated testing tactics.

QUOTE

If there's ever been a personal computer that everyone's gotta get, it's the Apple Power Mac G5.

QUOTE

Unfortunately, it appears that Apple has stolen a particularly tainted page from traditional PC vendors. Once, Steve Jobs marketed the Mac as the tool for individualists (think 1984's Super Bowl commercial) and then, for people who had some intellectual capacity (the Think campaign) and eventually, as a computer that simply worked without headaches and aggravation (the switch campaign), Apple now is taking on the PC industry with "benchmarketing."

It's right out of 1992, when PC vendors pumped up their efforts to cheat on benchmarks to claim a meaningless performance edge over a competitor. Sad to say, those benchmark shenanigans worked to a degree. For example, all the popular video card companies that remain today at one time exploited loopholes in benchmark design.

eweek

Bill · « **Reply #12 on:** July 31, 2003, 11:48:47 PM »

The Smell Of Fear
More
the inquirer

kelly · « **Reply #13 on:** August 01, 2003, 10:31:08 AM »

That guy makes me seem like a pretty lukewarm mac fan.

Whatever.

They need to get them out so people can see how a Dual G4 1.4 GHz

compares to the G5 1.8 for example.

And which should you buy.

There's not much doubt the Dual G5 2 GHz will smoke.

Bill · « **Reply #14 on:** August 01, 2003, 05:35:02 PM »

"Every time a PhotoShop cookoff is run on a PC and a Mac, the false accusations fly like a manure in a Texas tornado."

Had to read that twice.

News:

Author Topic: G5 Chip Article (Read 3841 times)