Performa 5320: what clock speed?

[Franklinstein]

I recently bought a Performa 5320 in part to do a side-by-side comparison with a 5420 because, according to specs available on multiple sites online and in the Apple service manual, the 5320 supposedly runs at 120MHz. Imagine my surprise when my 5320 arrived, in its original box, with both the label on the box and the markings on the processor specifying 100MHz.

The only resource that identifies the 5320 as a 100MHz system is this one. This wouldn't be the first time that mainstream and even Apple documentation was wrong; the endless versions of the same machine in the 90s was pretty confusing and this one, being a non-US system, probably wasn't scrutinized very often.

My question: does anybody else have a 5320? If so, what is its clock speed?

 

[Innes]

So I guess what might have happened here is that the 5230 model was a configuration only for Japan, hence why the Japanese resource you have linked is saying 100. Everything else refers to the computer as 5230CD which is what (according to Wikipedia) Europe and presumably the rest of Asia got.

Why did you picked a 53XX over a 54XX?

 

[Cory5412]

The weird part is that after so many years only one single reference lists the correct speed. Usually in the intervening years someone's emailed someone at everymac/LEM or another of the million different mac spec dexes and gotten them to update their listing for the machine's speed.

My initial thought was that it might have been a system whose motherboard was swapped, perhaps for a board such as the 5260. OS compatibility is usually "fine" when in close proximity or using a certain model's install disk on an older system, and you might never notice except if you look at System Profiler, which might list the exact configuration it thinks you have.

Just casually looking at the service manual at http://tim.id.au/laptops/apple/legacy/powermac.perf_5200.5300.pdf Apple lists 120MHz in there too.

Granted, given that these are almost certainly written in English in the USA and then (if I had to guess: not localized into other languages) distributed, the person who wrote this likely did not have a very good opportunity to review an actual Performa 5320.

The box is definitely the wildcard, and it is the main bus in my personal "what if someone swapped the board at some point?" theory.

 

[Franklinstein]

The vast array of confusing Performa models is one reason Apple had such a poor reputation in the mid-90s: nobody knew exactly what they were getting, higher numbers were often lower spec'd than similar models, and sometimes the only difference was bundled software. Ridiculous. 

So I guess what might have happened here is that the 5230 model was a configuration only for Japan, hence why the Japanese resource you have linked is saying 100. Everything else refers to the computer as 5230CD which is what (according to Wikipedia) Europe and presumably the rest of Asia got.

Why did you picked a 53XX over a 54XX?

Yeah the models all say "5320CD" or whatever on Everymac but I haven't seen too many late-model Performas after the 61xx series that kept the "CD" designation on any of the labels; CDROMs became essentially standard in 1995ish so it wasn't really useful for differentiation anymore.

I was mostly curious if someone in the EU or Asia-Pacific region (I know there are some Aussies who frequent the forum) had seen a similar example of the 100MHz variant of the 5320. Was it an undocumented configuration? Was this model was originally 100MHz but stepped to 120MHz immediately after introduction? Is it a documentation error? I'm trying to figure that out. These machines are fairly unloved so I'm not sure there are too many still kicking around.

Anyway I got a 5320 because it was supposedly the fastest (@120MHz) of the 52/53xx series, the machines LEM refers to as "Road Apples," and I was going to do a side-by-side performance comparison with a 5420, which is a 120MHz PCI-based model. Also the 5320 is a non-US machine so I figured I'd pick one up while I could get it cheap. I paid about $30 for mine (including shipping) in its original box; I'll add pictures of the box labels to the list of things that I'm supposed to get pictures to post here. The 5420, another non-US model (in black!), was about $50 including shipping, though it didn't include its box or accessories and the previous owner was obviously a smoker (yuck). As an aside, the black plastics of the 5420 don't seem quite as fragile as the beige plastics of the other models, though it's still pretty easy to snap off the bezel tabs if you're not careful.

 

[Franklinstein]

Here's the box label. Up top it says PowerPC 603e processor, clock 100MHz, 256k 2nd level cache memory. 

 

[Cory5412]

My money is on either documentation error, or last minute production change, similar to the 366MHz model of the Power Macintosh G3 that was announced, listed for sale in almost every catalog reseller in very early 1999, and then never materialized.

In terms of 5320 vs 5320CD, my guess is they are the same machine. Usually instances of "same number, different configuration" were Performa vs. Power Macintosh vs LC situations, which added to the confusion for the 630 and 6100/6200+ families.

I believe a (small) handfull of 6100 configurations had "6100" vs "6100/xxCD" configurations but as far as I know no 5200+ or 6200+ systems were shipped without optical drives.

As also mentioned, it wouldn't be the first incorrect bit of documentation, and since as far as I can tell every single mac-dex is based on Apple's USA spec database, it wouldn't surprise me if someone at Apple in the USA flubbed typing down that machine's information or was given incorrect information at some point and hadn't seen anything else about it to be able to correct it.

In regards to the performance: Even if it had been 120MHz, I think you'll find the 5420 is faster, and I don't think anyone will be very surprised by that result. I suspect you'll also find that the 5320 is "fine" -- being one of the fixed/revised machines with the 603e CPU which had the better designed L1 cache. 256k of L2 will help the 5320 along as well.

Another interesting comparison would be a 6200/75 and a 7200/75. I have a 6200 variant coming at some point and I intend to compare it with what I've got hanging around, although I don't happen to have a 7200/75 at the moment.

[general '90s mac industry commentary below this point]

In general, Performa was the brand sold to homes, Power Macintosh was the brand sold "professionally" and LC was the brand sold into education, until the 5400/6400 when that was narrowed to Power Macintosh for both education and "professional" and Performa for home, and then later simplified to all models presenting as "Power Macintosh" -- although the 6400 and 6500 both had "for $MARKET" as a configuation title in some media. (for example, there's a "6500/250 for Small Business" configuration listed in the service manual. These aren't often reflected on mac-dexes such as everymac or LEM, however.)

Anyway, the glut of performas was part of Apple's image problem at the time, but when you get right down to it, most of the mid '90s Apple spent floundering -- money they might have used into building up better machines was wasted on all manner of side-projects.

Part of Jobs' simplifying the product line and introducing the iMac was to address a problem Apple knew it had in the mid '90s, which was that nobody was excited by the next incremental performa upgrade. Even dumping the Performa name in favor of "Power Macintosh 6500/xxx for Multimedia" didn't excite anyone. I'm not even sure a 6500/400 would have. In the clone era, people chose clone vendors because they were less expensive and in many cases because buying one was easier and simpler than getting your hands on an Apple Mac, even one that was "easy" to choose, like a member of the 7/8/9 series.

Of course, those had almost exactly the same problem the performas did: Apple frequently added models to the lineup, with spec difference that were often invisible or buried a couple numbers deep. For example, there were cacheless 7200 and 6400 configurations, and additional CPU speeds got piled onto the stack often without Apple doing a good (or any) job re-aligning (say, 6400/180 is the release model, 6400/200 and 6400/225 are released, 6400/180 never gets discontinued or discounted, for example.)

At the high end, Apple had trouble with, of all things, pro Mac configurations that took forever to actually ship because Iomega had frequent production delays and backlogs with Zip drives, and for some reason ("dumping") Apple felt compelled to include Zip drives with the 8600 and 9600, instead of just letting users of those big professionally-oriented machines choose on their own what to add. It wasn't until 1998 that Apple would partner with some more reasonable retail resellers (CompUSA) and start offering direct sales online, so until then it made more sense to go buy a UMAX or a Power Computing from one of their web sites.

And ultimately that speaks to something I think is as true today as it was in 1998: In general, people don't want to spend a lot of time and energy balancing a bunch of different factors of computer performance against their budget. If you go spend $3000 on a machine, you want to trust that you got around $3000 worth of machine. There were a few gems in the Performa lineup, but (especially internationally where low-cache and no-cache models were most popular) a few machines were duds and often there would be no good reason given as to why this 6400/180 is way slower than another 6400/180. (There's also the somewhat related issue of, essen

Apple did a slightly better job of listing all the specs in 1997, but organizing and comparing them wasn't quite as easy as it was later on when the web was that much more ubiquitous. Add to that, different catalog resellers would have different deals and often major-looking catalog and back-of-magazine resellers would be selling old-stock and used machines way past when those machines stopped being "available" (a stuffed channel that was rejecting some new shipments was another problem Apple acknowledged in its annual reporting in the mid '90s leading up to the iMac.)

(I realize nobody said this, here, but these issues are why people were saying Apple was on its deathbed, which was, just to be clear, never really completely true.)

I don't think Apple was on a course to die in 1997 if they hadn't changed anything bigger, but I will argue that Jobs saved them from a death in the early 2000s if they didn't make a few major changes by then.

 

[Franklinstein]

In regards to the performance: Even if it had been 120MHz, I think you'll find the 5420 is faster, and I don't think anyone will be very surprised by that result. I suspect you'll also find that the 5320 is "fine" -- being one of the fixed/revised machines with the 603e CPU which had the better designed L1 cache. 256k of L2 will help the 5320 along as well.

Another interesting comparison would be a 6200/75 and a 7200/75.

Right, so I doubt there would ever be a question as to whether the 54xx is faster than a 53xx at the same clock speed. Rather, I'm interested in a definitive conclusion to the whole argument: how bad were they, really? The main drivers are both the original LEM "Road Apple" article, where these machines were basically described as the WORST COMPUTERS EVAR!!11!, and the rebuttal from Taylor Design, where he does correctly dispute several of LEM's crazier claims but then basically hand-waves away the rest of it, with things like 'every computer uses bridge chips', 'PowerPC native code ran as fast as on every other Power Mac', etc.

Basically my review would focus on: Are they really "road apples" to be avoided? Or are they simply legacy-focused machines that do they best they can? If they are poor performers, what makes them so bad? Obviously the small L1 cache of the 603 used in the first 52/62xx machines made 68k emulation a bit pokey but later models, with their 603e chips, negated that a bit. Also Connectix's Speed Doubler emulator may have been small and/or efficient enough to improve the basic 603's performance; I'll try to test that too (I also have a Performa 5210). My comparison would be to see exactly what the performance cost of all the legacy hardware is, compared to a machine with all-new, modern parts, where they are otherwise the same (processor, processor and bus speeds, L2 cache, RAM +/-8MB, hard drive, OS load, 10bT network).

Another interesting point: nobody has an proper overclocking solution for the 52/53/62/63xx except for a really old Japanese site (now visible only with the Wayback Machine) that does some complicated stuff with replacement clock crystals. However, a comparison of empty pads and resistors on the underside of the processors on the 75 and 100MHz boards suggests that an easy overclock may be possible; I'll try moving some resistors around to see if there are any positive results.

 

[Cory5412]

Just casually, given that you already have the machines and it sounds like you may have set them up and started to use them -- what are your initial impressions?

The heart of what LEM wrote, in 1997 when they first started publishing, basically boils down to that if it's 1997 and these machines still cost several hundred dollars and you have a choice between, say, a 6200 and a 7200, is that you should get the 7200.

I'd be interested in a more modern take on it, and my perception of that (and what I'll likely write on my own blog when I've had a chance to poke at the 6220 a bit) is whether or not they "can be used" -- the descriptions you see of the 6200 describe a computer that, more or less, should literally fail to boot up.

My take on the entire thing has long been that the 5200 and 6200 and their immediate family (up through the PCI change) were the cheapest PowerPC-based Macs you could get and were often among the cheapest you could get, period (though: some 5xx and 6xx machines stayed on sale for far too long.)

Considering their performance has to be done thinking about the context of a world where you can get an entire computer and some start-up software and usually a printer and a monitor for about $1900, or you can pay 3x that and get a computer (the 9500) that should be just around 2x as fast on paper and includes the following amenities:

• power cord
  
• mouse
  
• unformatted text editor
  

Of course, those are the extremes, but even stepping up to the next machine after the 6200 close to doubled your cost once you actually assembled a working computer and put some software on it.

That comparison of course stops making sense when the iMac and the Power Mac G3 crossed over in price and your $1299 could buy you either an expandable system with fewer included amenities or a most-in bundle system, trading some legacy compatibility for most of what you'd need to get started up front, without trading off any performance.

One more thought: From a modern perspective, I don't think that "reviews" of vintage hardware make sense, especially in any supply-limited ocmmunity. We shouldn't, at this point, be telling anybody to shy away from any Mac they can get their hands on.

I think it's important to be aware of what you're getting (something LEM does insanely badly in the modern context, for continuing to host extremely factually incorrect articles with little or no revisions showing up-to-date research or reflecting the needs of people "shopping" for these machines in modern times) but I don't think there's a good reason to classify any given vintage Mac as an "avoid this one".

Again, it's not like we're shopping for three-year-old PowerBook G3s to run OS X on and cache-having /233s cost the same as cacheless /233s on the used market.

but then basically hand-waves away the rest of it, with things like 'every computer uses bridge chips', 'PowerPC native code ran as fast as on every other Power Mac', etc.

To address this, specifically: An important thing to note here is that to my knowledge, Mac OS did not completely shed 68k code until literally Mac OS X. Every single release up to that point was frequently lauded as "even more PowerPC-native!" but as far as I know, apple either never really finished the job, or they only finished it mostly in the very newest versions of OS 9.2, which won't run on anything so old. (9.2 requires either PCI, by which point "the pain points" were gone or a G3, I forget which.)

So, the fairest way to evaluate this specifically is to run 9.1 with as much RAM as you can fit into either of these machines, and the newest software you reasonably can, such as the PPC versions of IE/OE4 or IE5/OE5 and Office 98 or 2001. Those applications have stiff-ish system requirements though, and so you might run into the other problem: A machine from 1995 with limited upgrade potential just isn't well equipped for things that were new several years later, at a time when everything in computing was moving very fast.

(That said: Anecdotally, IE4/OE4 run "fine" on my 840av under 8.0 or 8.1 with 24MB of RAM, but really only one at a time, I haven't had a lot of reason to try IE/OE4 on my next closest system, the 6100/66, but it'll be something I make time to do under 7.6.1 and 8.1 on that system as well.)

Given that the author has close to the exact machine you do, and says that in their experience it's "fine" I think it's okay to accept the hand-wavey explanation as being sourced both in the technical fact of reducing the amount of 68k emulation gets rid of one of the biggest pain points of the machine, and their experience using it.

That said, if your point is to prove something about the original group of 603 machines, the 5200/6200, doing it on a 5320/100 with the 603e (which addressed one of the bigger pain points in 68k emulation specifically) and had a 256k cache as a cherry on top is... not very valid.

One more thing to note: Initial shipments of 5200s had 8MB of memory installed from the factory, and there's a reasonably good chance, especially for those of us around the right age to have used these things when they were new in our schools, 7.5 on 8MB of RAM is an extremely bad look and probably exacerbated everything else about the machine. (Incidentally there were a handful of 7200 configurations with 8MB of RAM as well, I don't know why Apple thought that was a good idea except that Apple has pretty much for its entire existence included too little RAM in its computers.)

 

[Franklinstein]

Yes I can get that when comparing a 62X0 or whatever to a similarly priced 7200 in 1997ish, when these things were relevant, the 7200 would rightly be described as the one you should buy. Part of the problem with the LEM article was that they demonized the lesser machines when they didn't have to go that far. I mean, they could have said, "at the same price, the 7200 is 30% faster and has PCI for future expansion where the 62X0 is a little pokey but can use your old RAM and expansion devices." Boom. Done. I know which one I'd choose (not the 62X0).

I know that the original 5200 is a lesser machine (slower processor, half L1 cache, slightly slower bus (37.5MHz vs. 40MHz in later models)) that has 68k emulation issues due to its smaller L1 cache, but the basic architecture is exactly the same as the faster 53xx. Thus, a comparison of two machines that are identical except for the logic board should illustrate the performance impact of the use of legacy chips, as mentioned previously. Also every 52/53/62/63xx came with a 256k L2 cache; it's soldered onto the same DIMM as the ROM, so you really can't not have it. I have the expansion 256k L2 cache for the 5420. Any of the LEM-cited problems with serial ports or slow disk transfers or stuttering audio or whatever will be exactly the same on a 120MHz 603e-based 53xx as they are on a 75MHz 603-based 52xx (unless the problems are exclusively caused by slow 68k emulation performance, which is doubtful). Plus, if my experiments with overclocking yield fruit, I can potentially upclock or downclock a 52xx or the 53xx to within a few ticks of the other for a closer comparison. 

With the 5320 and 5420 I'm trying to compare apples-to-apples: everything the same except the logic board's support chips and expansion devices so that I can illustrate the impact (or lack thereof) of the extensive use of legacy devices and the requisite bridging between the various buses used in the 52/53xx. Now I just have to find either a 120MHz version of the 53/63xx board.

I'll be comparing both System 7.5.5 and OS 8.6 to see if there's a notable improvement in performance from the steady increase in PPC native code (I imagine there would be). Both boards will have 64MB of RAM installed (72MB for the 5420, since it has 8MB soldered) so low RAM and virtual memory won't be factors, though I may run a low memory comparison to see how each machine performs with extensive virtual memory usage; as noted, 8MB of RAM in 7,5,5+ is not fun for anybody regardless of which machine you're using.

 

[Cory5412]

.

Also every 52/53/62/63xx came with a 256k L2 cache; it's soldered onto the same DIMM as the ROM, so you really can't not have it.

Aha, forgot that detail, thank you.

Any of the LEM-cited problems with serial ports or slow disk transfers or stuttering audio or whatever will be exactly the same on a 120MHz 603e-based 53xx as they are on a 75MHz 603-based 52xx (unless the problems are exclusively caused by slow 68k emulation performance, which is doubtful).

Notably, these issues would also be present on the Performa 630 series, as well, since the 6200 was built pretty much by adding a 603 to the existing architecture. It's easy to find the developer notes Apple published for each machine and verify this in the block diagram.

With the 5320 and 5420 I'm trying to compare apples-to-apples: everything the same except the logic board's support chips and expansion devices so that I can illustrate the impact (or lack thereof) of the extensive use of legacy devices and the requisite bridging between the various buses used in the 52/53xx.

A good bit of information to accompany this part of your article-or-articles (and, a good exercise for you if you haven't done it) would be to look at the block diagram for the machines.

As Taylor Design noted, the setup in the 6200 is pretty much what a modern northbridge/southbridge looks like, so it wouldn't surprise me if a newer machine mirrored that, despite switching out almost all of the relevant components. It's been a while since I looked, however, I'll see if I can make itme to do that later today.

 

[Franklinstein]

As Taylor Design noted, the setup in the 6200 is pretty much what a modern northbridge/southbridge looks like, so it wouldn't surprise me if a newer machine mirrored that, despite switching out almost all of the relevant components. 

So the point about newer systems using bridging is relevant, but these machines used far too much bridging in their double quest for cheapness and legacy support. Most of the contemporary PCs only had the one or two bridges: processor to the north bridge (memory/primary expansion bus controller), and then the south bridge/legacy device controller (for an older expansion bus and slower I/O devices). The 52xx design would be the equivalent of (at the time) a Pentium 100 with a 66MHz/64-bit bus talking directly to a bridge chip (the L2 cache and BIOS being the only other things directly on the processor bus), and that bridge chip then talking on a 32-bit 33MHz 486 bus to the memory, video, SCSI, ATA controllers, and the third bridge chip, which in turn ran the serial ports, floppy drive, sound, and any expansion devices on an 8MHz 8-bit ISA bus. Now, Apple's engineers did a pretty commendable job making it work, but it's incredibly inefficient and entirely unnecessary unless you have a mandate from management to reuse as much old crap as possible.

Any decent design would have put the processor and memory controller on the same high-speed bus and then bridged the processor/memory bus to the slower I/O and expansion bus. This is what the 54xx did: the processor talked directly to the memory controller at full speed and full 64 bits (and also directly to the new Valkyrie-AR video controller), and the only bridge was to the PCI bus; there were no unnecessary bridges partly because there was no legacy stuff to contend with. The thing that really killed the 52xx's performance was that first bridge chip (Capella) and subsequent halving of the memory bus width so that Apple could re-use the 630's old F108 memory controller.

I'd like to know exactly how much of a demand existed for the continued use of legacy expansion devices, or if it was a perceived customer requirement that wasn't really there. I mean, the only things that work in the CS slot were the internal modems and Ethernet cards; the only things that still worked in this model's LC PDS were Ethernet cards and maybe really slow video cards. Did someone really need two Ethernet cards in one of these? Is an internal modem a big enough deal to design your entire computer around? Or did Apple have 50,000 chips left over from building the Quadra 630 and decide they just had to use them instead of write them off?

 

[Cory5412]

Did you have a chance to take a look at the develoepr notes?

The notes for the 6200 are linked from and hosted on the Taylor Design web site: http://taylordesign.net/downloads/references/PowerMac5200-6200.pdf

A mirror of the 630 developer notes are here: http://mirror.informatimago.com/next/developer.apple.com/documentation/Hardware/Developer_Notes/Macintosh_CPUs-68K_Desktop/Mac_LC_630_Quadra_630.pdf

The main difference is that there is a "Bus Translation Unit" slotted between the CPU and the rest of the system bus for what appears to be addressing purposes. Another bridge connects the data bus. Given that the performance improves dramatically with the improved cache structure of the 5300/6300 and friends with their 603e CPU and its bigger, unified L1 cache, I think it's at best misguided to simply blame the Capella (aforementioned Bus Translation Unit") for all of the 6200's problems.

I can't speak to what demand there was to actually use the same expansion as the previous LC/Performa series systems. Perhaps the initial thought was that these systems would work with the IIe card, or that some of the (admittedly, relatively few) other non-Ethernet upgrades were worth preserving. The TV/Video system was probably worth implementing on the new series, but there's also the argument of leaving that kind of things for higher end machines with more room for expansion. Though, especially given that Apple's gimme to the education market as late as 1999 was to have ATi build a Rage GPU with analog video input, I'm gonna guess that education was doing something with those video and TV/Video systems. Whether that's laserdisc playback in HyperCard stacks, k-12 level video production as part of lessons, or something else, I couldn't tell you for sure. So, the more important part of the 630 to that market was probably the a/v system, and Apple probably legitimately saved money by putting a 603 and a Capella on the 630 board instead of building an entirely new architecture.

Apple could probably have saved itself some space on these systems by going the route of the x100 Power Macs and building Ethernet onboard and then using either a dedicated PDS slot or omitting card-based internal expansion (and also implicitly: a/v functionality) entirely. It would likely have resulted in a "better" system that simultaneously failed to meet core needs of one of its largest audiences.

Of course, this gets into a different kind of issue, which is that we can probably have years worth of discussions about different possibilities for re-engineering these systems. For example, why wouldn't Apple have just re-shaped the 6100 platform for these systems? (That would be the more natural progression, I think, than introducing PCI on the low end education/consumer system first.)

 

[Cory5412]

More generally, and I don't know if there is a good answer to this at all, it sounds like the question you're trying to ask is whether or not it would have been possible to build a better system.

The philosophical answer here is that the 603 in the 6200 needed to have the issues it did in order for the better cache design for the 603e to be evident.

 

[Cory5412]

one more thought: back in the '90s, schools loved to do upgrades, and they loved to "employ" students to get "technology class" credit to do, among other things, relatively easy machine upgrade and maintenance tasks, and opening up an entire fleet of machines to swap out cards or upgrade RAM wasn't necessarily out of the question.

It wouldn't at all surprise me if school districts ordering a fleets of 6200s and swapping existing LCPDS Ethernet cards from old I/II/III systems into the new 6200s was a thing that actually happened at some point. It wouldn't even surprise me if some of those places just slotted the new 6200 in place and kept using the 12" monitors and older keyboards and mice.

(Though in the '90s a popular move was to order a new computer lab and then relocate your old computer lab into a cheap manufactured building out back or distribute the machines directly into classroom to increase overall computer access.)

 

[Gorgonops]

that bridge chip then talking on a 32-bit 33MHz 486 bus to the memory, video, SCSI, ATA controllers, and the third bridge chip, which in turn ran the serial ports, floppy drive, sound, and any expansion devices on an 8MHz 8-bit ISA bus.

Technically speaking your standard Pentium/Pentium-II era Southbridge like the PIIX4 does have a "bridge after a bridge" because the way they implemented ISA is as a bridge device (PCI ID: 8086:7110) on the PCI arbiter. IE, the ISA bus is technically not directly hung off the CPU's native bus but is "double-translated", even though it happens inside the same IC. (And then note that legacy I/O like the serial/parallel/floppy controllers were then typically handled by a FDC37B787 or equivalent "Super I/O" chip that sat on said double-bridged ISA bus.) So far as I can tell every Intel chipset from the 430FX onward put the ISA bus logically behind the PCI bus.  (Which raises some interesting questions that I believe I've seen people argue about on places like the VCF with regard to what the newest machines with "real" ISA bus slots are; that matters to some people for obscure reasons. I'm having a really hard time finding a datasheet for the 430LX and NX to figure out how the heck they implemented ISA.) So from that point of view sticking a "bridge after a bridge" is completely standard practice and there's nothing wrong with the 5200. (Or, at least, that doesn't really belong the on the list of things intrinsically wrong with it.) The major hit against the architecture is mostly that the RAM is on a 32 bit bus instead of 64 bit.

Looking at the block diagram my biggest question really would be is what if any additional latency the "543D Latches" that sit between the CPU and everything that's south of Capella introduces to RAM and I/O transfers. RAM addressing and refresh happen on the "Quadra Side" of the Capella BTU but the data bus technically doesn't go through there. As long as the "Quadra side" is clocked at some even fraction of the 603's native bus speed (so there's no delay in address propagation) I'm going to ignorantly suggest there isn't a huge issue with this, but if those latches have to deal with running asynchronously that could suck.

 

[Cory5412]

My guess is, given that the architecture scaled up to 120MHz and most people with those machines don't talk about having a lot of trouble, those 543D latches are probably "not that bad" regardless of whether or not they introduce latency or anything of that nature.

(Especially since: you're buying a performa on which to run math blaster or clarisworks, not a 9500 on which to run photoshop or avid or protools.)

 

[Gorgonops]

those 543D latches are probably "not that bad" regardless of whether or not they introduce latency or anything of that nature

Yeah, I was mostly just thinking of the sort of "problems" that a specifically RAM-pounding benchmark might point out. These machines were built a few years after the start of the era that DRAM speed started substantially falling behind CPU cycle speed (thus the widescale deployment of increasingly sophisticated caching strategies) anyway, so even if the RAM really does run at a significant speed penalty compared to the 603e's Platonic Ideal it might not show that much except under relatively rare circumstances.

 

[Franklinstein]

My guess is, given that the architecture scaled up to 120MHz and most people with those machines don't talk about having a lot of trouble, those 543D latches are probably "not that bad" regardless of whether or not they introduce latency or anything of that nature.

(Especially since: you're buying a performa on which to run math blaster or clarisworks, not a 9500 on which to run photoshop or avid or protools.)

I would hope that complaints would fall off after the myriad service recalls were performed (bad ROM, bad logic board, and/or bad internal VGA cable).  The primary concern with the memory bus on these is that the controller is behind a bridge and the bus is only 32 bits wide as opposed to a 603(e)'s native 64 bits. 

As stated, one of the primary design goals, in addition to being inexpensive, was to enable the use of legacy hardware (LC PDS cards, CS cards, RAM) to further entice institutions into upgrading. A good number of performance compromises had to be made to allow this but all in all, once the original 603 was replaced with the 603e and System software improved, they weren't too bad; even the original 603-based ones got better with OS 8 and a bit more RAM. 

I remember running The Playroom on something similar to these, if not these exact machines (most probably in Power Macintosh 5200/75LC guise) back in elementary school (~1994) and it seemed to go ok to my inexperienced eyes. You're right in that they're more of an appliance than something you'd actually lust after like the 9500, and I don't know too many people who wax lyrical about their vacuum cleaners or microwaves, either; they're just there, doing their jobs in an unremarkable fashion.

 

[Franklinstein]

So I did a little experimenting: attempts to clock up the 52xx board ended with only non-bootable configurations. Further reading of the 603 user manual reveals that the 603 supports only 1, 2, 3, or 4:1 bus multipliers; no half-multipliers allowed. So it doesn't look like a 52xx is overclockable without changing the bus speed; clocking the existing chip to 3x would yield a ~112MHz configuration, which is well beyond the rated speed of the chip (and above the apparent 100MHz maximum for the fastest 603). Perhaps if the bus clock was reduced to 33MHz it would be possible to overclock the processor to 100MHz but that may not be stable, depending on the way clock generation works for the various frequencies used in the 52xx.

The 603e does support half multipliers, but so far I have only managed to downclock a 53xx board to 80MHz from 100MHz; there may be other resistors that need to be relocated to yield an overclock instead. If I had a 120MHz board to compare it would be much easier.