Accelerate Your Macintosh! - Mac L2 Cache DIMMs

Click for Universal Drive Adapters!

A Click on this Banner shows your site support to my Sponsors

The Gurus Certify
This Site is Cool.

The Cache Crop:
A detailed look at Macintosh Cache Dimms
Last Updated: July 16, 1997

Note: This article was written in 1997, before there were G3 and G4 CPU Upgrades, which don't need motherboard cache dimms. The comments below apply only to systems still running a 601/603/604 CPU. L2 cache dimms (esp. 1MB) are not generally being made now in 2000, since demand is literally nil after the G3 and later CPU Upgrades became popular. (Especially since 1MB L2 cache dimms were often $100 or so, and G3/350 CPU upgrades with faster backside cache on the card are available for just a bit more - a much better value. See the OWC Site Specials page for the last place I saw L2 cache dimms available and for more modern upgrades also (RAM, CPUs, Drives, etc.)

=Summary= Why upgrade your cache?: For systems without a G3/G4 CPU (still running a 604/603/601 CPU), a 1 meg cache is your best performance boost bargain (along with adding Ram). For maximum performance 604E accelerator cards with adjustable bus speeds (XLR8, Newer, PowerLogix) should be run at the fastest speed possible. Many fixed speed cards also have 48 or 50mhz bus speeds, and therefore may need a "fast cache" to run reliably. The G3/PowerPC 750 and later CPU cards have their own very fast "backside" L2 cache and don't depend on motherboard caches for performance. The comments here are directed only at systems still using a 601/603/604 CPU.

Size Counts: Forget what some magazine experts have said - we've proven that larger caches do produce better real world performance. We've done more accelerator & cache testing than anyone else on the net, and a quality 1 meg cache is one of the best performance bargains you can buy. With the faster CPU's, get the largest cache you can afford and that will run with your processor card.

Best Cache: Easy Pick - The PowerLogix Ultra 1 meg cache (P/N UC1MB). Up to 25% better performance than a 256k cache (varies by task) and the only 1 meg cache I've seen that runs with the 1997 XLR8 604e CPU Upgrade cards. In most machines, the less expensive standard Rapidcache version performs just as well. [Note, this cache is no longer made of of 2000.]

Best Card with any Cache: I know this is getting monotonous, but the PowerLogix PowerBoost Pro is far and away the best 604 CPU Upgrade card tested in 1997 to maximize any cache speed, due to its wide selection of bus speeds. Read about it in our review. [Note: see the CPU Upgrades page for a complete list of reviews]

L2 Cache Dimm Models tested:

Newer Tech. 256 cache dimm

Two Stock 256K Apple cache dimms

Two 512k Cache dimms - from MacCPU & XLR8

Four 1 meg Cache Dimms (IDT & Lifetime Memory)
and the current speed kings - PowerLogix's RapidCaches
Size Counts!- 1 meg cache 25% faster than 256k in real world test!

Spy Pix: - Take a peek at the Apple/Micron 256k, PowerLogix 1 meg, and IDT 1meg caches.
Update: 8500/120 with 1meg cache scores
(stock 8500/120 with 256k cache scores are Here)
10/13/97 - For a slide showing bus speed/cache speed - click Here
"slide show" info on hi-speed caches (PPC750) is Here.

Newer Technology 256k Cache:

Looking at this cache I didn't have a lot of hope it would run as fast as the rest of the lot, as it had the slowest rated chips of any of the cache dimms we've seen. However it had 8 Motorola sram chips (at first I thought it was a 1 meg cache), and Daystar claims Motorola static ram chips make the most reliable Mac cache. As it was the only cache to boot reliably with the Dual Processor cards we had, we believe them. The Sram chips were rated at 12ns, and the tag at 15ns. Fast enough for the Apple bus speeds on paper, but in the 8500 world where even the phase of the moon seems to affect cache operation - I had my doubts that it would work above 50mhz bus speeds - but it did.

It was the only cache to boot reliably with our troublesome MP (Dual CPU) card, but nPower still reported failed memory tests. Removing the MP card and installing any other cpu card reported all memory passing again. What does that say about our MP card?

With single cpu cards, the Newer cache was fine, running as high as 52.8mhz bus speeds reliably even with the cache sensitive XLR8 cards. However if you're after maximum performance I'd pass on anybody's 256k cache - get a 512k or 1 meg if possible. You'll buy yourself as much as 25% in performance with the larger cache. As far as 256k caches go however, the Newer is probably the best aftermarket 256k available.

Apple 256k Caches:

The two Apple caches were of different OEM designs. The first one contained 13ns Motorola Sram chips, and 11ns tag ram chip (This one is gone, so the speeds are to the best of my recollection). This cache maxed out at 45mhz with the XLR8 card, with all memory installed (see memory notes for more info). The Newer card was not tested with this cache, as I "traded" the cache with a local machine that was destined to never run faster than a 40mhz bus, and had what I guessed was a superior 256k cache installed (read on).

The second Apple 256k cache had Micron 11ns Srams (surface mount) and a 8ns tag ram. A picture is Here. This module had a lower parts count, and based on my experience with Micron ram and sram, I guessed that this was a superior module - it was. The Newer card ran at 52.5mhz bus with this cache and all memory installed, and at 56.25mhz bus when the two stock 8meg dimms were removed, leaving the four 16meg dimms installed (in a different 8500 motherboard). This was the fastest bus speed we obtained with any of the cache and card combinations so far.

512K Cache:

I tested two 512k caches, both were from the same OEM (IDT). MacCPU and XLR8 evidently have the same supply. These caches used IDT 9ns Sram and a 12ns (per marking) Tag. I am puzzled by the slower Tag ram chips on this and many other Mac caches (but not all of them), as this goes against standard "copy back" cache design theory that states the Tag ram must be faster than the Sram. Slower tag ram is used in the pipeline cache of PC models, but no Mac uses a pipeline cache to my knowledge. The IDT 512k cache allowed the XLR8 card to run at a 47.6 mhz bus speed reliably in one motherboard, but not another. The XLR8-180 actually ran fine for about 30 minutes at the 50mhz bus speed, completing the MacBench 4.0 cpu/fpu tests but locked the machine during the disk testing. The Newer card fared better with the 512k cache, running 52.5mhz bus speeds reliably.

1 meg Caches - the performance choice:

After a long absence, 1 meg caches that work are back. No longer do you have to buy a Power Computing machine to get the benefits of a large cache. Since we've probably done more Mac Accelerator testing than anyone else on the web, believe it when we say the faster processors really need the larger caches. Real world and benchmark performance increases of 25% are typical. Virtual PC also benefits greatly from a large cache. A quality 1 meg cache is the best "bang for the buck" out there.

IDT 1meg cache:
This was the first reliable 1 meg cache to hit the market this year, and is still sold by many vendors. The IDT cache design uses SEC 8ns static rams, along with 12ns tag ram chips. A picture of the IDT cache is here. Although the XLR8 card failed to run at 45mhz with the 1 meg cache, the Newer card has run rock solid at 52.5mhz bus for some time.
The IDT 1 meg cache has a bus speed limit of 53mhz in most machines.

Note: My industry sources say the SEC 8ns chips used on the original IDT 1 meg cache we have are not made anymore, and new modules may have 9ns chips (should not make a difference really).

LifeTime Memory 1meg cache:
I've just received and did limited testing on Lifetime Memory's $375 1 meg cache. It was loaned for testing on the Newer MP card, which has not run any other cache but Newer's in this particular 8500. The MP card would not run with this cache either, generating at best the typical "bus errors" and failures to start. It seems MP cards do not like the larger cache - at least our 2 samples. The cache did run fine with the Newer single cpu card at bus speeds of 52.5mhz. The next higher setting was 56.25mhz bus, which would not boot completely.

The Lifetime 1 meg cache consisted of four Micron 10ns srams, and two 8ns tag rams, and appears of very high quality. It did not however provide any improvement in bus speed for the Newer single cpu card in this 8500. Other users have reported better results with this cache in single cpu Mac's.

The KING of Caches - PowerLogix 1 meg RapidCache:

XLR8 Owner's - your cache has arrived! PowerLogix's "Ultra" 1 meg cache (PowerLogix part number UC1MB). For the first time, I was able to run a 1 meg cache with the XLR8-225mhz and 233mhz card, with bus speeds of 47mhz/235mhz cpu on the 225 card, and 49mhz bus/245mhz cpu on the 233 card. Based on successful testing of the Powerlogix Pro card at up to 55mhz bus speeds, I'd say this is the cache to have for all fast processor cards. On less cache sensitive cards like the Newer Technologies upgrades, the less expensive IDT cache may be fine, but you still may get better overall performance (via higher bus speed settings) with the PowerLogix Ultra cache. If I had to pick only one cache to keep - this would be my choice without a doubt.

Update: 7/13/97
Both the Ultra and standard PowerLogix RapidCache 1meg models easily run at 55mhz bus speeds in my 8500. They may do faster speeds, as 55mhz seems to be my motherboard limit. For a MacBench graph of what a stock 120mhz 8500 scores with the 1 meg cache - click Here.

A few words regarding RAM:

Problems that at first appear cache related may in fact be memory related. For best results remember the following guidelines:

Match dimms in the Ax/Bx banks to allow interleaving (75/85/9500 series)
(Make sure dimms are same type, style, speed - preferably the same vendor)
Arrainge dimms starting at the upper banks (A4/B4 in a 8500), working down.
Largest dimms in the top bank usually works best.
Sometimes moving or removing dimms has helped solve problems in Apple 75/85/9500s.
Always buy 60ns dimms, they cost no more and are recommended in fast machines.
Some users report that disabling interleaving (moving dimms) has helped problem Mac's.
If you suspect ram/cache problems - run the extensive Ram test contained in the nPower Control panel (p/o Daystar's MP software). It works fine with single cpu cards and is the best memory test out there bar none. It has detected memory problems that were not shown with other ram test utilities.

Our System Ram:
Up until July 97, our 8500 system Ram consisted of two 8meg 60ns marked single sided stock dimms, and four 16meg dimms (one matched pair, and one mismatched (not the same vendor/style) pair). All memory was interleaved. In an attempt to squeeze more bus speed out, we tried moving simms to defeat interleaving, with no improvement in results.

Update:7/97
Tests run from this date forward will use a matched 60ns pair of 64meg dimms from MacGurus. This is done to minimize the issue of ram mismatching while still giving me plenty of ram for serious graphics work. If you're looking for the best prices on ram and cache, always check MacGurus. (And in case you're wondering, I paid for the Ram, as usual).

UPDATE: MacBench doesn't lie - Cache size does matter!
Some "experts" have attempted to discount the approximately 25% performance increase I've seen in MacBench 4.0 with the 1 meg cache versus the 256k caches. Well MacBench has been proven to be more accurate than the "experts" realize, as tests in Photoshop have shown a "real world" increase in performance almost exactly the same as MacBench 4.0 reports!

A PowerMacintosh Resource Page article cited times to run a large file Gaussian blur with a 256k and 1 meg cache. As quoted below the difference is dramatic with the 1 meg cache. (Quoting from the page linked above)
"The best example I can cite is this: I did a full-blown 250-point radius Gaussian blur on a 600 DPI, 78 meg file.

With the 200MHz card and the 256K cache, the rendering took 2295 seconds, around 38 minutes.
With the 200MHz card and the 1 meg cache, it took 1652 seconds, almost 28 minutes.
With the 400MP card and the 1 meg cache, it took 315 seconds, just over 5 minutes. Absolutely freaking incredible! "

Contrary to the "experts" published opinions, it is obvious (and common sense) that cache size does make a difference in the real world. Of course Power Computing has known that all along.

Summary:

Surprisingly, MacBench seems to favor the large caches to a great degree. I say surprising as most benchmarks don't usually show a big increase with larger caches (at least on the other platforms). The Mac's combination of (relatively) slow bus speed and high processor/bus speed ratios are the reason for the increased need for cache in my opinion. In most cases the 1 meg cache delivered an additional 25% in performance compared to the 256k cache. The Photoshop example cited above shows it does benefit real world applications too - by almost the exact MacBench percentage increase. As these results show, its obvious that the cpu is running more efficiently with the larger caches. With the faster processors and bus speeds, get the most cache you can afford and run reliably in your system. Prices for quality 1 meg caches are not that much more expensive that the 512k variety, in some cases as little as $25 depending on vendor/brand.

Keep in mind that although all of a OEM's cache may use the same brand/speed of chips, each lot or module may vary. Some lots may be faster than their rated speed. The other more common variable is your system ( ram, cards, motherboard ). I'll state up front I'll never buy another 8500 system, or those based on that motherboard design due to the many cache issues (and memory issues as well) seen on this board design. Apple is supposedly releasing new 8600/9600 systems with cache on the processor card, running at up to 100mhz bus speeds at the August 97 MacWorld expo.

CHRP designs may finally bring the Mac world "up to speed" as far as motherboard and cache technology. Not only will it reduce costs and help standardize the use of lower cost components (Apple licensing policies willing that is), but it is rumored to use pipeline caches (popular with all PC's now) and run bus speeds of 66mhz to 83mhz, far exceeding the 40mhz to 50mhz bus speeds of current Apple designs. (Power Computing is an exception - their PowerCenter Pro series ships with a 60mhz bus).

The popular opinion is that a lack of termination, and/or unoptimized cache trace/socket locations are to blame for many of the Apple cache problems. Whatever the reason, hopefully others have learned from the nightmare all Mac tweakers have went through regarding cache. The CHRP motherboards should be much better in this respect, with bus speeds of 66 and possibly 83mhz rumored to be on the drawing board. 83mhz bus speeds are working in the PC world with common, cheap ( $20 for a 512k module!) pipeline burst cache designs. Hopefully CHRP will bring that home to the Mac as well.

= Where to Buy =
L2 Cache dimms are now getting rare since G3 and G4 CPUs don't need them.
Check Other World Computing for availability.

= Top of Page =
= Back To WWW.XLR8YOURMAC.COM =

Systems | CPU Cards | SCSI | IDE | Firewire | Cache | Video Cards | Tips/Misc. | Forums | FAQ

Last Revised: 7/16/97

Copyright © 1997.
No part of this sites content is to be reproduced in any form without written permission.
All brand or product names mentioned here are properties of their respective companies.
Users of the web site must read and are bound by the terms and conditions of use.

	The Gurus Certify This Site is Cool.
The Cache Crop: A detailed look at Macintosh Cache Dimms Last Updated: July 16, 1997
Note: This article was written in 1997, before there were G3 and G4 CPU Upgrades, which don't need motherboard cache dimms. The comments below apply only to systems still running a 601/603/604 CPU. L2 cache dimms (esp. 1MB) are not generally being made now in 2000, since demand is literally nil after the G3 and later CPU Upgrades became popular. (Especially since 1MB L2 cache dimms were often $100 or so, and G3/350 CPU upgrades with faster backside cache on the card are available for just a bit more - a much better value. See the OWC Site Specials page for the last place I saw L2 cache dimms available and for more modern upgrades also (RAM, CPUs, Drives, etc.)
=Summary= Why upgrade your cache?: For systems without a G3/G4 CPU (still running a 604/603/601 CPU), a 1 meg cache is your best performance boost bargain (along with adding Ram). For maximum performance 604E accelerator cards with adjustable bus speeds (XLR8, Newer, PowerLogix) should be run at the fastest speed possible. Many fixed speed cards also have 48 or 50mhz bus speeds, and therefore may need a "fast cache" to run reliably. The G3/PowerPC 750 and later CPU cards have their own very fast "backside" L2 cache and don't depend on motherboard caches for performance. The comments here are directed only at systems still using a 601/603/604 CPU. Size Counts: Forget what some magazine experts have said - we've proven that larger caches do produce better real world performance. We've done more accelerator & cache testing than anyone else on the net, and a quality 1 meg cache is one of the best performance bargains you can buy. With the faster CPU's, get the largest cache you can afford and that will run with your processor card. Best Cache: Easy Pick - The PowerLogix Ultra 1 meg cache (P/N UC1MB). Up to 25% better performance than a 256k cache (varies by task) and the only 1 meg cache I've seen that runs with the 1997 XLR8 604e CPU Upgrade cards. In most machines, the less expensive standard Rapidcache version performs just as well. [Note, this cache is no longer made of of 2000.] Best Card with any Cache: I know this is getting monotonous, but the PowerLogix PowerBoost Pro is far and away the best 604 CPU Upgrade card tested in 1997 to maximize any cache speed, due to its wide selection of bus speeds. Read about it in our review. [Note: see the CPU Upgrades page for a complete list of reviews]
L2 Cache Dimm Models tested: Newer Tech. 256 cache dimm Two Stock 256K Apple cache dimms Two 512k Cache dimms - from MacCPU & XLR8 Four 1 meg Cache Dimms (IDT & Lifetime Memory) and the current speed kings - PowerLogix's RapidCaches Size Counts!- 1 meg cache 25% faster than 256k in real world test! Spy Pix: - Take a peek at the Apple/Micron 256k, PowerLogix 1 meg, and IDT 1meg caches. Update: 8500/120 with 1meg cache scores (stock 8500/120 with 256k cache scores are Here) 10/13/97 - For a slide showing bus speed/cache speed - click Here "slide show" info on hi-speed caches (PPC750) is Here.

Newer Technology 256k Cache:
Looking at this cache I didn't have a lot of hope it would run as fast as the rest of the lot, as it had the slowest rated chips of any of the cache dimms we've seen. However it had 8 Motorola sram chips (at first I thought it was a 1 meg cache), and Daystar claims Motorola static ram chips make the most reliable Mac cache. As it was the only cache to boot reliably with the Dual Processor cards we had, we believe them. The Sram chips were rated at 12ns, and the tag at 15ns. Fast enough for the Apple bus speeds on paper, but in the 8500 world where even the phase of the moon seems to affect cache operation - I had my doubts that it would work above 50mhz bus speeds - but it did. It was the only cache to boot reliably with our troublesome MP (Dual CPU) card, but nPower still reported failed memory tests. Removing the MP card and installing any other cpu card reported all memory passing again. What does that say about our MP card? With single cpu cards, the Newer cache was fine, running as high as 52.8mhz bus speeds reliably even with the cache sensitive XLR8 cards. However if you're after maximum performance I'd pass on anybody's 256k cache - get a 512k or 1 meg if possible. You'll buy yourself as much as 25% in performance with the larger cache. As far as 256k caches go however, the Newer is probably the best aftermarket 256k available.

Apple 256k Caches:
The two Apple caches were of different OEM designs. The first one contained 13ns Motorola Sram chips, and 11ns tag ram chip (This one is gone, so the speeds are to the best of my recollection). This cache maxed out at 45mhz with the XLR8 card, with all memory installed (see memory notes for more info). The Newer card was not tested with this cache, as I "traded" the cache with a local machine that was destined to never run faster than a 40mhz bus, and had what I guessed was a superior 256k cache installed (read on). The second Apple 256k cache had Micron 11ns Srams (surface mount) and a 8ns tag ram. A picture is Here. This module had a lower parts count, and based on my experience with Micron ram and sram, I guessed that this was a superior module - it was. The Newer card ran at 52.5mhz bus with this cache and all memory installed, and at 56.25mhz bus when the two stock 8meg dimms were removed, leaving the four 16meg dimms installed (in a different 8500 motherboard). This was the fastest bus speed we obtained with any of the cache and card combinations so far.

512K Cache:
I tested two 512k caches, both were from the same OEM (IDT). MacCPU and XLR8 evidently have the same supply. These caches used IDT 9ns Sram and a 12ns (per marking) Tag. I am puzzled by the slower Tag ram chips on this and many other Mac caches (but not all of them), as this goes against standard "copy back" cache design theory that states the Tag ram must be faster than the Sram. Slower tag ram is used in the pipeline cache of PC models, but no Mac uses a pipeline cache to my knowledge. The IDT 512k cache allowed the XLR8 card to run at a 47.6 mhz bus speed reliably in one motherboard, but not another. The XLR8-180 actually ran fine for about 30 minutes at the 50mhz bus speed, completing the MacBench 4.0 cpu/fpu tests but locked the machine during the disk testing. The Newer card fared better with the 512k cache, running 52.5mhz bus speeds reliably.

1 meg Caches - the performance choice:
After a long absence, 1 meg caches that work are back. No longer do you have to buy a Power Computing machine to get the benefits of a large cache. Since we've probably done more Mac Accelerator testing than anyone else on the web, believe it when we say the faster processors really need the larger caches. Real world and benchmark performance increases of 25% are typical. Virtual PC also benefits greatly from a large cache. *A quality 1 meg cache is the best "bang for the buck" out there.* IDT 1meg cache: This was the first reliable 1 meg cache to hit the market this year, and is still sold by many vendors. The IDT cache design uses SEC 8ns static rams, along with 12ns tag ram chips. A picture of the IDT cache is here. Although the XLR8 card failed to run at 45mhz with the 1 meg cache, the Newer card has run rock solid at 52.5mhz bus for some time. The IDT 1 meg cache has a bus speed limit of 53mhz in most machines. Note: My industry sources say the SEC 8ns chips used on the original IDT 1 meg cache we have are not made anymore, and new modules may have 9ns chips (should not make a difference really). LifeTime Memory 1meg cache: I've just received and did limited testing on Lifetime Memory's $375 1 meg cache. It was loaned for testing on the Newer MP card, which has not run any other cache but Newer's in this particular 8500. The MP card would not run with this cache either, generating at best the typical "bus errors" and failures to start. It seems MP cards do not like the larger cache - at least our 2 samples. The cache did run fine with the Newer single cpu card at bus speeds of 52.5mhz. The next higher setting was 56.25mhz bus, which would not boot completely. The Lifetime 1 meg cache consisted of four Micron 10ns srams, and two 8ns tag rams, and appears of very high quality. It did not however provide any improvement in bus speed for the Newer single cpu card in this 8500. Other users have reported better results with this cache in single cpu Mac's.
The KING of Caches - PowerLogix 1 meg RapidCache:
XLR8 Owner's - your cache has arrived! PowerLogix's "Ultra" 1 meg cache (PowerLogix part number UC1MB). For the first time, I was able to run a 1 meg cache with the XLR8-225mhz and 233mhz card, with bus speeds of 47mhz/235mhz cpu on the 225 card, and 49mhz bus/245mhz cpu on the 233 card. Based on successful testing of the Powerlogix Pro card at up to 55mhz bus speeds, I'd say this is the cache to have for all fast processor cards. On less cache sensitive cards like the Newer Technologies upgrades, the less expensive IDT cache may be fine, but you still may get better overall performance (via higher bus speed settings) with the PowerLogix Ultra cache. If I had to pick only one cache to keep - this would be my choice without a doubt. Update: 7/13/97 Both the Ultra and standard PowerLogix RapidCache 1meg models easily run at 55mhz bus speeds in my 8500. They may do faster speeds, as 55mhz seems to be my motherboard limit. For a MacBench graph of what a stock 120mhz 8500 scores with the 1 meg cache - click Here.

A few words regarding RAM:
Problems that at first appear cache related may in fact be memory related. For best results remember the following guidelines: Match dimms in the Ax/Bx banks to allow interleaving (75/85/9500 series) (Make sure dimms are same type, style, speed - preferably the same vendor) Arrainge dimms starting at the upper banks (A4/B4 in a 8500), working down. Largest dimms in the top bank usually works best. Sometimes moving or removing dimms has helped solve problems in Apple 75/85/9500s. Always buy 60ns dimms, they cost no more and are recommended in fast machines. Some users report that disabling interleaving (moving dimms) has helped problem Mac's. If you suspect ram/cache problems - run the extensive Ram test contained in the nPower Control panel (p/o Daystar's MP software). It works fine with single cpu cards and is the best memory test out there bar none. It has detected memory problems that were not shown with other ram test utilities. Our System Ram: Up until July 97, our 8500 system Ram consisted of two 8meg 60ns marked single sided stock dimms, and four 16meg dimms (one matched pair, and one mismatched (not the same vendor/style) pair). All memory was interleaved. In an attempt to squeeze more bus speed out, we tried moving simms to defeat interleaving, with no improvement in results. Update:7/97 Tests run from this date forward will use a matched 60ns pair of 64meg dimms from MacGurus. This is done to minimize the issue of ram mismatching while still giving me plenty of ram for serious graphics work. If you're looking for the best prices on ram and cache, always check MacGurus. (And in case you're wondering, I paid for the Ram, as usual).

UPDATE: MacBench doesn't lie - Cache size does matter! Some "experts" have attempted to discount the approximately 25% performance increase I've seen in MacBench 4.0 with the 1 meg cache versus the 256k caches. Well MacBench has been proven to be more accurate than the "experts" realize, as tests in Photoshop have shown a "real world" increase in performance almost exactly the same as MacBench 4.0 reports! A PowerMacintosh Resource Page article cited times to run a large file Gaussian blur with a 256k and 1 meg cache. As quoted below the difference is dramatic with the 1 meg cache. (Quoting from the page linked above) "The best example I can cite is this: I did a full-blown 250-point radius Gaussian blur on a 600 DPI, 78 meg file. With the 200MHz card and the 256K cache, the rendering took 2295 seconds, around 38 minutes. With the 200MHz card and the 1 meg cache, it took 1652 seconds, almost 28 minutes. With the 400MP card and the 1 meg cache, it took 315 seconds, just over 5 minutes. Absolutely freaking incredible! " Contrary to the "experts" published opinions, it is obvious (and common sense) that cache size does make a difference in the real world. Of course Power Computing has known that all along.

Summary:
Surprisingly, MacBench seems to favor the large caches to a great degree. I say surprising as most benchmarks don't usually show a big increase with larger caches (at least on the other platforms). The Mac's combination of (relatively) slow bus speed and high processor/bus speed ratios are the reason for the increased need for cache in my opinion. In most cases the 1 meg cache delivered an additional 25% in performance compared to the 256k cache. The Photoshop example cited above shows it does benefit real world applications too - by almost the exact MacBench percentage increase. As these results show, its obvious that the cpu is running more efficiently with the larger caches. With the faster processors and bus speeds, get the most cache you can afford and run reliably in your system. Prices for quality 1 meg caches are not that much more expensive that the 512k variety, in some cases as little as $25 depending on vendor/brand. Keep in mind that although all of a OEM's cache may use the same brand/speed of chips, each lot or module may vary. Some lots may be faster than their rated speed. The other more common variable is your system ( ram, cards, motherboard ). I'll state up front I'll never buy another 8500 system, or those based on that motherboard design due to the many cache issues (and memory issues as well) seen on this board design. Apple is supposedly releasing new 8600/9600 systems with cache on the processor card, running at up to 100mhz bus speeds at the August 97 MacWorld expo. CHRP designs may finally bring the Mac world "up to speed" as far as motherboard and cache technology. Not only will it reduce costs and help standardize the use of lower cost components (Apple licensing policies willing that is), but it is rumored to use pipeline caches (popular with all PC's now) and run bus speeds of 66mhz to 83mhz, far exceeding the 40mhz to 50mhz bus speeds of current Apple designs. (Power Computing is an exception - their PowerCenter Pro series ships with a 60mhz bus). The popular opinion is that a lack of termination, and/or unoptimized cache trace/socket locations are to blame for many of the Apple cache problems. Whatever the reason, hopefully others have learned from the nightmare all Mac tweakers have went through regarding cache. The CHRP motherboards should be much better in this respect, with bus speeds of 66 and possibly 83mhz rumored to be on the drawing board. 83mhz bus speeds are working in the PC world with common, cheap ( $20 for a 512k module!) pipeline burst cache designs. Hopefully CHRP will bring that home to the Mac as well.

= Where to Buy = L2 Cache dimms are now getting rare since G3 and G4 CPUs don't need them. Check Other World Computing for availability.
= Top of Page = = Back To WWW.XLR8YOURMAC.COM =
Systems \| CPU Cards \| SCSI \| IDE \| Firewire \| Cache \| Video Cards \| Tips/Misc. \| Forums \| FAQ

Last Revised: 7/16/97 Copyright © 1997. No part of this sites content is to be reproduced in any form without written permission. All brand or product names mentioned here are properties of their respective companies. Users of the web site must read and are bound by the terms and conditions of use.