October 3, 2003

by Don Engstrom

Probably the biggest difference between these three upgrade cards, the Sonnet Encore/ST, the PowerLogix PowerForce G4 Series 233 and the GigaDesign G-celerator is the way they look.

All three cards run at 800 MHz (there are also 1 GHz, 1.25 GHz (GigaDesign) and 1.20 GHz (Sonnet & PowerLogix), models. All three cards have 2 MB of L3 cache (in addition to 256K of L2 cache running at full processor speed). All are fairly easy to install and each card, in our performance testing, turned in about the same scores. Both the Sonnet & GigaDesign cards have a commendable 3 year warrantiy, and the PowerLogix cards have an adequate 2 year warranty... it is always nice to see a manufacturer stand behind their products!

Then the manufacturers part ways....

Sonnet has always tried to keep things simiple. They want you to be able to stick a processor upgrade into your machine, enjoy the extra performance and otherwise just forget about the new brain you have transplanted. This strategy has been a big winner for them, pushing them forward to be the dominant player in the United States upgrade market.

As you can see above, the Sonnet Encore/ST card (the one on the left in the image), is rather spare looking. Sonnet uses miniaturized components and utilizes the old processor's heatsink for cooling ... rather than providing you with a new one. You can make no performance adjustments to the card, which auto-configures itself to your machine once it has been installed.

Many of the components on the GigaDesign G-celerator are not miniaturized. GigaDesign says,engineering in larger capacitors (which you see on the card standing upright like skyscrapers), will allow them to ramp up to the next generation of processors more easily. The GigaDesign cards come with their own copper heatsink and fan. The heatsink adds a lot of heft to the G-celerator, which must weight in at, at least 5 times that of the Sonnet Encore/ST and PowerLogix cards.

On top of the GigaDesign card there is a set of jumpers you must set to get the card to perform at the correct speed. Using the jumpers It is possible to try to over-clock the processor to see how far you can push it before it becomes unstable, However this is not something that GigaDesign actively encourages. The jumpers' default setting is for a machine with a 100 MHz bus. If you have a machine with a 133 MHz bus you will have to modify the jumper settings using the detailed instructions that come with the card.

Those of you that see a processor upgrade as more of an unfinished adventure, rather than a settled story, will appreciate the ability to fiddle with the jumper settings.

PowerLogix stands somewhere between Sonnet & GigaDesign. Like Sonnet, the PowerLogix cards utilize the original heatsink, and has miniaturized components (picture above left). However the PoweForce, like GigaDesign card, has a series of jumper settings that must be set correctly depending on the bus speed of your machine. Similarly the default setting is for a machine with a 100 MHz bus. Again the inclusion of the jumpers allows propeller-heads to see if they can push the upgrade a little faster than its rated speed. But most of us will be content to set the jumpers to their default, and leave it there.

Compatibility

The Sonnet Encore/ST PowerLogix PowerForce Series 233 and GigaDesign G-celerator cards are compatible with the following machines:

Power Mac G4 (AGP Graphics)
Power Mac G4 (Gigabit Ethernet)
Power Mac G4 (Digital Audio)
Power Mac G4 (QuickSilver 2001)
Macintosh Server G4 (450 & 500 MHz)

In addition the Sonnet & PowerLogix cards are compatible with the G4/Cube. However installing the Encore/ST or PowerForce in a Cube requires some additional hardware. In the case of Sonnet you must buy an adapter which Sonnet will sell you for $30 or $50, (the price depends on the speed of the card you have purchased). With the PowerLogix upgrade you need to add a small fan, which they sell separately for $13.

The GigaDesign and PowerLogix cards can also be installed in the Quicksilver 2002 machines. Since these machines already run at 800 MHz, 933 MHz or dual 1GHz, it is unlikely that you would find it cost effective to add an upgrade card to one of these setups at the current speed/price ratio.

The GigaDesign cards require Mac OS 9.2.1, 9.2.2 or OSX, and PowerMac firmware version 4.2.8 for all non Quicksilver models. The Sonnet cards require Mac OS 9.2 through Mac OS X (OS 9.2.1 or 9.2.2 may be required to perform Apple firmware updates). PowerLogix cards require Mac OS 9.2.1, 9.2.2, or Mac OS X.

We tested the cards on a Power Mac G4/400 with AGP graphics. The machine was running Mac OS 10.2.4.

Cards from all manufacturers ran though both our performance tests and our stress tests without any problems.

Installation

Installation of any of these upgrades is very easy and straight forward. There is no software to install (unless you do not have one of the required Mac OS versions on your machine. In this case you will have to update the system software). You need to uninstall the orginal processor daughter card, which is attached to the motherboard by three phillips type screws. The trickiest part is reinstalling the original heatsink ... if you have opted for a Sonnet or PowerLogix card. The stock heatsink is held onto the the daughter card by two clips that must be reattached once you have the new upgrade card in place. A set of pliers is used for this operation, and you need to be gentle as you ease the spring loaded clips into place. As long as you are careful, it is no big deal.

All in all it shouldn't take you but a few minutes to get any of these upgrades installed. A testament to the quality of Apple's design, as much as that of the upgrade manufacturers. If you are going to install the Sonnet or PowerLogix upgrade into a Cube, according to the literature, the process is somewhat more complex. Indeed Sonnet recommends professional installation.

Performance

All of the upgrade cards performed well. They also all turned in about the same performance boost. This is not surprising given that the hardware configuration of each card is identical, and there is no software driver to get in the way of performance. In raw processing capability, all three of the 800 MHz cards turned in slightly better than twice the performance of our test machine, which in stock configuration runs at 400 MHz.

In some tests the results with the cards installed was significantly greater than what you would expect from an upgrade clocked at only twice the original processor speed. For example in our AppleWorks Search & Replace test the upgrade cards turned in about two and a half times the performance of the stock machine. Also in one of our QuickTime encode tests the upgrades were more than twice as fast. We think that the reason for the lopsided results, in these two tests, is down to the differences in the caching schemes of the upgrade cards, which are more robust than the stock machine configuration. The stock machine has no on chip L2 cache. The upgrade cards have 256K running at full processor speed. The stock machine has 1 MB of backside cache (off chip L2 cache), whereas the upgrade cards have 2 MB of L3 cache. Cache is high-speed memory which allows the processor to access pre-fetched data quickly, negating the need to trot all the way back to the (slower) main memory.

Though the performance numbers of the upgrade cards were good, in most of the test results below you'll see that they turned in less than twice the speed of the stock machine. Performance ranged anywhere from 65% to 90% better than the machine in its original state.

We also noted a nice performance boost in on-screen graphics and in certain games, much better frame rates.

Then there were the puzzling results where the upgraded machine actually performed worse than the stock setup ... though by a very small amount.. In our drive test and our MP3 encode test, all three processor upgrades turned in scores about 14% worse that the stock configuration. We have no hard information on why this would be. Both these tests should respond to the better performance, and enhanced configuration of the upgrades, but for some reason the performance dropped off. We can only think that it is related to the extra L3 cache somehow ... though we don't understand why. A mystery!

Final thoughts

Upgrades from all three companies worked like champs during our testing period. They all turned in very good performance numbers over our stock machine. The increased power will be welcome by anyone stuck with an older machine, and who frequently make use of processor humgry applications. We even saw a nice speed boost in some of our game tests. The fact that our stock machine did not have a L3 cache, and that all the upgrade cards came with 2 full MB of L3 cache plus a L2 running a full processor speed, dramatically helped performance of the upgraded machine in certain tests. Of course there is the law of diminishing returns. Installing a 800 MHz upgrade in a machine with a stock speed above 533 MHz does not make a lot of sense to us.

So whose card to choose?

All manufacturers have something special to recommend them - none of which is related to relative performance ... where they are about equal in our testing.

Sonnet is the big kahuna. They have succeeded where others have failed, by keeping things simple, having a diverse product line, and being persistent. Sonnet has had its share of hiccups along the way, but have learned from them, and moved on to become the dominant player in the American market. They also have operations worldwide. If you like the comfort of a known commodity, want an upgrade that you install and then forget about, and don't mind paying a price premium for this ease, a Sonnet upgrade may be the one for you. [see our note below on the price changes that occurred after this was written]

GigaDesign is the new kid on the block. A small company, they currently only manufacture upgrades for G4 machines. They saw a hole in the marketplace for low-cost processor upgrades and moved agressively to fill it. It is nice to see another company enter the marketplace with a solid upgrade card. It is also good that they believe enough in their product to offer a 3 year warranty. What GigaDesign lacks in a track record they make up in low-prices. The 800 MHz G-celerator card's manufacture's suggested retail price comes in 24% less than the cost of an equivalent Sonnet card. This amounts to a $70 difference ... which is not chicken feed. [Editor's note: since this was written, Sonnet & PowerLogix have dropped the price on their 800 MHz cards to almost match the cost of the GigaDesign card ... you will see the new prices listed below].

The GigaDesign cards may also appeal more to those that want to tinker with their upgrade .. just to see just how much performance they can squeeze out of it. Having jumpers available allows the adventurous the ability to try to over-clock the upgrade, to see if they can push it a little faster than its rated speed. So if you are a tinkerer, you might want to consider the GigaDesign card.

The PowerLogix card also has user adjustable jumper setting, that may allow you to over-clock your upgrade card slightly. This depends on the individual processor that has been installed on the upgrade card. All processors are tested to work reliably at the rated speeds, but some upgrades may work at higher clock-speeds. Then again it is also possible that they will become unstable at any speed higher than that for which they are rated. It is the luck of the draw.

PowerLogix has a 2 year warranty whereas Sonnet & GigaDesign both offer a 3 year manufacturers warranty. If a processor upgrade doesn't go bad within the first month's of its use, it is unlikely to fail for many years to come. However the comfort-zone for the consumer is always enhanced by a longer warranty period. If we were able to award ratings on a 1/4 point basis we would knock a 1/4 off the rating of the PowerLogix card because of its truncated warranty.

One of the original big three upgrade manufacturers, PowerLogix has been struggling somewhat in the recent years against the marketing muscle of Sonnet. However, they seem to be finding their footing again, coming out, often well before others, with a variety of upgrade solutions. Indeed they are currently the only ones offering a dual processor upgrade for the G4 Towers.

The bottom line is that any of these processor cards, for about $300, will nearly double the processing power of a stock 400 or 450 MHz machine ... depending on the applications you are working with. While the upgrades won't give you the 'over-all' performance of a new machine, compare the price tag to; $1,499 for a new G4/1 GHz Tower, or the uncertainty of a used G4/800 which you can pick up on eBay for around $1000 ... sans warranty.

These upgrades provide you with an affordable way to extend the life of your current machine, and will add value to it, when it comes time to take the big plunge ... upgrading to an all new Power Mac.

Note: We plan to add addendum to this review when we take a look at the performance of the 1 GHz and 1.2/25 GHz in this family of upgrades from Sonnet, PowerLogix & GigaDesign. Check our Processor Upgrade Hub for links.

Product: Encore/ST G4/800/800/256K/2 MB

Company: Sonnet Technologies

MSRP: $300

Hits: Very good at processor intensive tasks, speedy processor caches. Reasonable price/performance ratio. Works in Cube. Very good stability. Three year warranty

Misses: none significant

Requirements: Power Mac G4 (AGP Graphics), Power Mac G4 (Gigabit Ethernet), Power Mac G4 (Digital Audio), Power Mac G4 (QuickSilver 2001), Macintosh Server G4 (450 & 500 MHz), Cube. OS 9.2 through Mac OS X. Installation in Cube requires additional hardware

Rating: (5 possible)

Product: G-celerator G4/800/800/256K/2 MB

Company: GigaDesign

MSRP: $289

Hits: Very good at processor intensive tasks, speedy processor caches. Very good stability. Reasonable price/performance ratio. 3 year warranty

Misses: The only caution is that they have a much shorter track record than the others. Does not work in the Cube.

Requirements: Power Mac G4 (AGP Graphics), Power Mac G4 (Gigabit Ethernet), Power Mac G4 (Digital Audio), Power Mac G4 (QuickSilver 2001), Power Mac G4 (QuickSilver 2002), Macintosh Server G4 (450 & 500 MHz). OS 9.2.1, 9.2.2 or OSX, and PowerMac Firmware version 4.2.8 for all non Quicksilver models

Rating: (5 possible)

Product: PowerForce Series 233 G4/800/800/256K/2 MB

Company: PowerLogix

MSRP: $299

Hits: Very good at processor intensive tasks, speedy processor caches. Very good stability. Reasonable price/performance ratio. Works in Cube

Misses: Warranty not as robust as the other two manufacturers. They have been struggling somewhat in their competition with Sonnet

Requirements: Power Mac G4 (AGP Graphics), Power Mac G4 (Gigabit Ethernet), Power Mac G4 (Digital Audio), Power Mac G4 (QuickSilver 2001), Power Mac G4 (QuickSilver 2002), Macintosh Server G4 (450 & 500 MHz), Cube. OS 9.2.1, 9.2.2, or Mac OS X. Installation in Cube requires additional hardware

Rating: (5 possible)

"Real World" Tests

The tests below are from our suite of real world application tests. These tests feature a diverse selection of applications commonly used by the Mac community. The test suite was designed to render an accurate and well rounded picture of the upgrade's performance. All of the tests below, were timed with a stopwatch. The times were then converted to percentages, relative to out test machine, the Power Mac G4/400, which is set to 100%. For all scores, higher numbers are better. Also included, for comparison purposes, is a stock Power Mac single processor G4/1 GHz, which is the currently shipping low-end Tower (January, 03).

The cache setup for all the options below are as follows:

Power Mac G4/400 : 1MB of L2 backside cache @ 200 MHz (half processor speed)
Sonnet Encore/ST G4/800: 256K of on chip L2 cache @ full processor speed and 2MB L3 cache
GigaDesign G-celerator G4/800: 256K of on chip L2 cache @ full processor speed and 2MB L3 cache
PowerLogix PowerForce Series 233 G4/800: 256K of on chip L2 cache @ full processor speed and 2MB L3 cache
Power Mac G4/1.0 GHz: G4/800: 256K of on chip L2 cache @ full processor speed and 1MB L3 cache

Desktop Tests

Nice little boost on the boot process

For some reason the upgraded machine preformed a little worse than the stock machine in the drive speed test. Perhaps it has something to do with the caching schemes of the upgrade cards, which are quite different from the stock machine. As you can see though, a processor upgrade in an older machine won't be able to keep up to the hard drive speeds the latest model computer

The Folder Copy test duplicates a folder filled with thousands of items. The File Copy test duplicates a single file

This is one of the oddest tests results. It is a processor intensive task. In all configurations the processor is completely saturated, however there is very little variation in performance results. So there is some significant bottleneck in the process that is constricting data flow.

Let 1K Windows Bloom is a simple carbon application that opens and closes 1,000 windows. Nice significant boot to screen redraws

Two folders with many items are searched using OS X's new search function.

Large Document & Database Type Tests

A Macro (series of complex actions) was run in Microsoft's Excel program, which is part of Office X

Stresses the processing & memory systems of the machine. This test takes place in a large AppleWorks document. This is a raw processing power test in which the size of the L2/L3 Cache plays an important part. The scores between the processor upgrades are within the margin of error.

Number Crunching & Rendering Tests

The Fractal program is a good test for assessing the fundamental processing potential of each setup. All the upgrades are evenly matched

A filter Effect is applied to an iMovie

BareFeats Photoshop Test Suite - mimics a production type workflow in Photoshop. Again nearly identical scores for the upgraded machine

Encoding/Decoding Tests

The default encode compresses a QuickTime movie for streaming on the Web

Preps QuickTime Movie for import into iMovie

No clue as to why an MP3 encode would actually drop on the upgraded machine. As you can see this happened for all three upgrades and we ran the test several times to double check. Again we think that the difference in caching schemes might be the culprit, though we don't know why this would be the case.

Multitasking

MP3 Encode, AppleWorks search & replace and folder copy are all carried out at the same time.

This is straight processing work

Gaming

Giants is tuned to take advantage of the processor(s). It  also helps to have a better graphics card. The upgraded machine turned in between 15 and 16 fps, compared to roughly 11.5 fps for the stock 400 MHz machine

4x4 EVO2 is a demanding racing game. The upgraded machine turned in roughly 10 fps.