Apple PHX 100

[Franklinstein]

Try running a 68k version of MacBench on the 6100 and original 62xx for comparison. The only valid complaint against these machines is that 68k emulation suffered on the 6200 compared to its contemporaries or successor models, and if MB4 is a fat binary or PPC native you won't not notice the effect with a native PPC comparison with the 6100. 

Nothing outside of supercomputers ran external cache at CPU speed, ever, at least not once CPU speed started to decouple from bus speeds in 1990-something. No way Apple was going to pay for 100MHz SRAMs for a bargain basement computer. If they could even get 100+MHz SRAMs in 1995, they would have put them in the 9500. Guess what? They didn't: the 9500 ran its L2 between 40 and 50MHz depending on processor bus speed. All Apple L2 caches ran at logic board bus speed until the 9600/300 and 350, which had specially designed processor cards that ran the onboard L2 cache at 100MHz (while ignoring the slower logic board cache), which is still far less than the 300/350MHz of the processor clock. Then of course there were the G3s and G4s with backside caches, and those never ran faster than 50% of clock speed, usually less. The whole point and largest benefit of an L2 cache is to have low-overhead SRAM memory (read: no refresh cycles required) available directly to the processor without having to go through/wait for other chips or narrower buses to get there. The 603 processor can only talk to external devices including L2 cache at logic board bus speeds, especially since not only is L2 directly on the 60x bus in these machines, but so is the ROM and Capella, and I guarantee none of those are capable of 100MHz operation.

As for a bottleneck between CS and LC PDS? I never said this was an exclusive problem for the 6200; it affects anything with Primetime including the Q630 and LC 575. Two networking cards could be problematic, if it was even possible to use legacy Mac OS simultaneously on two different domains or network types. Or CS Ethernet and a PDS video card. Or on the 575, CS Ethernet and a IIe card. It's not likely a problem anyone would encounter often, but there's the potential for degradation if both slots were active simultaneously, especially since they're both only 16MHz slots that may share the same '030 bus (the dev note isn't terribly clear if it's two '030 buses or one shared) and the Primetime has other things to do in addition to managing both expansion slots. 

My point was that there were a few errors or omissions in the guy's page, not that every word he typed was wrong; its still better than the LEM BS. 

 

[Franklinstein]

Small correction regarding full-speed external caches:

It could technically be argued that the Pentium Pro uses an external cache, because the cache and CPU are built individually on two separate dies, but I contend that it doesn't: this cache is built in the same process as the CPU and the two are ultimately bonded together in the same processor module. They are intertwined to such a degree that any error in bonding or in either die rendered the entire module as scrap; it couldn't be reworked, unlike a faulty memory module mounted externally. Thus, I define the PPro's cache as internal.

The Slot 2 Pentium II and III Xeons used a similar approach: the CPU and cache modules are both built on the same process as the CPU but this time all of the chips are attached to a carrier board on a special full-speed cache bus. These are proper external caches running at full processor speed, but nothing else in the system is on this bus or runs at this speed except the L2 cache; everything else is accessed by the 100MHz FSB (the later 133MHz FSB PIII Xeons use small on-die L2 caches, negating the argument and ultimately the need for a slot-based processor).

Regardless, both processors were targeted at the maximum performance money-is-no-object end of the IT field and both ended up in a variety of Top 500 Supercomputers (such as Sandia's ASCI Red) while the 603 was used in zero, so I wasn't totally inaccurate.

 

[Franklinstein]

To the cache issue specifically, where does it specify what bus speed the cache runs at? In the dev note, the block diagram shows it being essentially directly connected very closely to the CPU, but doesn't appear to list a speed.

In addition, from page 14 of the dev note:

So, I think that Daniel Taylor is probably correct in saying the 256k cache runs at the full speed of the CPU. Apple certainly thought so, and wrote as much as their dev note.

Actually, nowhere in that entire Dev Note does it mention system bus speed, so I can understand where there may be some confusion because of this omission; totally Apple's fault here. This is amended in the later 6300's Dev Note (again, the guy's using the 6200's Dev Note to talk about the 6300) but for reference, the 52/62xx uses a 37.5MHz system bus. This bus clock, provided by a single oscillator and central clock generator chip, is used by the processor for its external transactions (internally it is clocked at 2x bus), and it is also the base clock for Capella and F108 and by extension the L2 cache and system RAM. The clock generator chip provides Primetime with a separate 16MHz clock to run the CS/LC PDS slots, in addition to various other clock frequencies used throughout the system.

According to Motorola's user manual for the 603, this is a non-standard and unsupported configuration: the 603 is designed to only run at 1, 2, 3, or 4x bus speed, with a variety of supported bus speeds. However, 37.5 is not one of them; the 75MHz part is designed to be run at 3x a 25MHz bus. while the fastest 1:1 speed available for the 603 is 66MHz. In addition, no Apple support chips in existence at that time could operate beyond 50MHz, let alone up to 66MHz. If they did, don't you think the 9500 would operate at the same speed? Why would their flagship have slower parts than their cheapest machine?

Anyway the later 53/63xx is the same except it has a standard 40MHz system bus and now the 603e supports half multipliers, which is how we get 100MHz clocks from a 40MHz base (40x2.5). The rest of the clocks and Primetime's operation are still the same. 

 

[Compgeke]

Regardless, both processors were targeted at the maximum performance money-is-no-object end of the IT field and both ended up in a variety of Top 500 Supercomputers (such as Sandia's ASCI Red) while the 603 was used in zero, so I wasn't totally inaccurate.

But isn't the 603 more akin to a Celeron than a Pentium II/Pentium III? The higher end x86 boxes were the Pentium II and Pentium III, while your home consumer model was typically a Celeron or K6. The higher end Macs used a 604/604e while there lower end were a 603/603e. 

You could get something more consumer like a 7300 with a 604e just like you could get higher end consumer boxes with a Pentium II/III. Both cheaper than the full blown HP Kayak or Powermac 9500.

 

[Brett B.]

Just my own experience but I saw just as many PII/III boxes back in the day in use at home as I did Celerons and AMD K5/K6/etc chips.  Actually the AMDs were really not all that popular until the Athlon series came out.  From then on 2-3 years, it was probably 80% AMD chips in consumer PCs that I saw but the vast majority were HP machines.  

Most of the Celeron based machines I worked on were used in K-12 education.

 

[Cory5412]

Probably the best Mac equivalent to the Celeron is a couple 6400/4400 and 7200 models that shipped without L2 cache.

The 6200/6300 all did. I still haven't had a chance to look at the 6300 developer note to see what Apple said there. As far as I know, Apple never published a revision to the 6200 note, so it's possible that the speeds of the 603 bus are different from the two models.

Notably, the only things on the 603 bus are the CPU itself, the Calpella, and the ROM/Cache, so none of the existing chips from the 030/040 platforms involved in the 630/6200/6300 platform need to run at "very" high speed. Those buses (the '040 bus connecting Calpella to PrimeTime I/II/III, F108, and Valkryie, and the '030 PDS and i/o stuff coming out of Primetime II run at) run at their own speeds. The developer note for the 6200 does also state the speed of the 603 bus (it says it's the same speed as the CPU frequency) and the i/o bus coming out of PrimeTime II, but it doesn't specify the frequencies of the 040 data/address buses.

It's possible that the 040 bus in this machine runs at 37.5MHz, but the 603 bus is disconnected, logically and physically, from the 040 bus and doesn't need to run at the same speed. As the Taylor Design article says, basically  Calpella/F108 is the northbridge and PrimeTime is the southbridge.

Everything I've seen of newer platforms suggests that when things like "using '030 PDS cards" isn't a concern then 

Two networking cards could be problematic,

Let's be ENTIRELY clear here: In a different subforum on this very site, somebody put a 10/100 NuBus ethernet card into a Quadra 950 with a powerPC upgrade and it wasn't meaningfully faster than the onboard  network interface.

I put a file on my big fast file server on my fully switched gigabit LAN and my blue-and-white G3 can't routinely break a couple megabits of download speed, in IE. (Netscape does a little better.)

Classic Mac OS is HORRIFYINGLY bad at networking.

That said - yes, in some cases (with router software, ASIP supports multihoming as well) two network interfaces are supported. That said, even a 6200 cost a lot more than a real switch or router even in 1995 so nobody would've done it.

 Or CS Ethernet and a PDS video card. Or on the 575, CS Ethernet and a IIe card. It's not likely a problem anyone would encounter often, but there's the potential for degradation if both slots were active simultaneously, especially since they're both only 16MHz slots that may share the same '030 bus (the dev note isn't terribly clear if it's two '030 buses or one shared) and the Primetime has other things to do in addition to managing both expansion slots. 

Ethernet/video could be an issue, but, what's NuBus like? there's machines with video, ethernet and their disk all hanging off an 020 or 030-based nubus interfaces.

Ethernet+IIe card is a desirable configuration, but I have it on reasonably good authority (I asked a person who is deep in 8-bit Apple II) about it and he says this configuration is officially unsupported. "Apple says it won't work."

I have never seen any english-speaking Mac user use pretty much anything other than network or IIe card in an LCPDS slot. I'm well aware that other cards existed -- even outside of Japan -- but to my knowledge the installed base is pretty much zero, relative to the number of these machines that both were in use when they were new/current/relevant and now, in a "vintage" contxt.

Given the local love for the Color Classic and the existence of the CCII, I'm not at all going to be surprised to hear that this specific thing is different in Japan.

 

[Cory5412]

also, re LC-PDS devices: it wouldn't at all surprise me if the graphics and video capture devices are the kinds of things that don't work on the 6200 (or perhaps the 575/580/630) because of the lack of actual "processor direct" access.

 

[Franklinstein]

The 6200/6300 all did. I still haven't had a chance to look at the 6300 developer note to see what Apple said there. As far as I know, Apple never published a revision to the 6200 note, so it's possible that the speeds of the 603 bus are different from the two models.

Notably, the only things on the 603 bus are the CPU itself, the Calpella, and the ROM/Cache, so none of the existing chips from the 030/040 platforms involved in the 630/6200/6300 platform need to run at "very" high speed. Those buses (the '040 bus connecting Calpella to PrimeTime I/II/III, F108, and Valkryie, and the '030 PDS and i/o stuff coming out of Primetime II run at) run at their own speeds. The developer note for the 6200 does also state the speed of the 603 bus (it says it's the same speed as the CPU frequency) and the i/o bus coming out of PrimeTime II, but it doesn't specify the frequencies of the 040 data/address buses.

It's possible that the 040 bus in this machine runs at 37.5MHz, but the 603 bus is disconnected, logically and physically, from the 040 bus and doesn't need to run at the same speed. As the Taylor Design article says, basically  Calpella/F108 is the northbridge and PrimeTime is the southbridge.

You mean this, copied straight from the 5200/6200 Dev Note:

The internal bus structure consists of three internal buses; the 64-bit wide 603 data bus, the 32-bit wde 68040 bus, and the 32-bit wide I/O bus. The 603 bus is connected directly to the main processor and runs at the same clock rate. An external 256 KB second-level cache and 4 MB of ROM attach directly to the 603 data bus and help to optomize system performance. 

Yes it says the 68040 bus is 32 bits "wde", and that L2 cache is attached to the 603 bus and helps to "optomize" system performance. Typos and omissions aren't exactly foreign to these Dev Notes. Note my previous observation that there are exactly zero references to the system bus speed, only to processor internal clock and the 16MHz of the CS/LC PDS slots. It doesn't mention system RAM speed either.

Did you read previously where I noted that there existed ZERO 603 processors that could run at 1:1 processor:board speed exceeding 66MHz? I linked the 603 UM. It's not that many pages. That alone should be enough of a clue that anything physically external to the 75MHz processor on a 6200 is running at a lower rate, specifically no greater than half of that (which is 37.5, if you're curious).

Or maybe the fact that absolutely no Apple support chips produced in 1995 ran faster than about 50MHz, including Capella? Even if it did, there's the fact that neither the L2 chips or Capella changed with the increase to 100 or 120MHz 603e models. If the faster 603e chips ran their external L2 caches at 100 or 120MHz, why did they use exactly the same L2 cache modules from the 75MHz models? Were they somehow upward compatible with a >25% increase? And why did the faster 603e require a heatsink while Capella, now supposedly also running at 100 or 120MHz, didn't? It's because everything outside of the processor on the new models ran at 40MHz while the 603e ran at a multiple of that (2.5 or 3x) internally. Again, according to the documents from Motorola, there were exactly zero 603 or 603e chips that could exceed a 66MHz bus.

Go boot a 52/62/53/63xx, run TattleTech/Newer Gauge or Clockometer/Speedometer/Metronome/whatever and tell me what speed it has the system bus and L2 caches. I guarantee it's 37.5 on the 75MHz models and 40 on the 100/120MHz models.

Anyway going through my cache of Dev Notes, I don't have one for the 6300, only the 5260 which is basically the same as far as the board is concerned: it runs the 100MHz 603e instead of the 75MHz 603. I don't have any Dev Notes for any machines with the 120MHz 603e or the latest variants with soldered ROM and vacant L2 cache slots. These things don't have the greatest documentation.

I'll concede the CS/LC PDS thing as being a rare perfect storm, if it happens at all; I've never tried it because I have few non-Ethernet LC PDS cards. However, as you noted with the CS Ethernet and IIe card where "Apple says it won't work", I'll assume this is because the CS and LC PDS slot share the same 030 bus and only one can be active at a time. So basically choose one and forget the other (excepting CS modems, which are basically serial pass-through devices not on the 030 bus). Not that there were a ton of options outside of networking anyway.

 

[Cory5412]

I'll assume this is because the CS and LC PDS slot share the same 030 bus and only one can be active at a time

Your guess here is wrong. The stated/accepted reason is because CS devices want/use 32-bit addressing mode.

In reality, it might actually work, who knows. There's been some work, if I remember correctly, on running the IIe card in the 580/630, and that same work might allow it to work in a 575 with an ethernet Comm Slot device. (if it's been successful, I dont' know the status on that.)

However, CS ethernet should work with other PDS devices. that was never a documented as a limitation of any of these systems.

Or maybe the fact

Okay, conceded. We've gotten a little beyond the point of my statement which is that the Taylor Design article, while having a technical error or two (based on Apple's own writing) is the most correct available after-the-fact assessment of those machines.

I have a 6220/75 at home, as I've said, it's "fine" with PPC native code -- I've run 7.6.1 and 9.1 with a couple different apps, which TBH almost everything I happen to use is, I'll pull out some older 68k versions of things and poke around with it at that point.

The other thing to remember here is that, again, a 6200 with a monitor, keyboard, mouse, a raftload of software, and usually a modem or a printer cost 1/5 of what a 9500 with a mouse and a text editor cost. It cost around half of what a thusly equipped 7500 cost, and if you added a keyboard and monitor to a 7200, you were also at about two 6200s.

It was an insanely good deal, for which there was a compromise if you bought, borrowed, or already had certain kinds of older software.

 

[Gorgonops]

Wow, this sure turned a little heated. My one comment:

I have a 6220/75 at home, as I've said, it's "fine" with PPC native code

If PPC being able to run mostly native code gives these machines a new lease on life, that's totally great, but... in context, if these things really were dogs with 68k code because of the emulation simply being flatly inappropriate for the CPU that genuinely would have been a big deal when they were introduced, and you can't blame people for bearing a scar from it. Remember, these things were introduced running System 7.5.1, and with that version of MacOS almost the entire OS runs in emulation on PPC. It probably did genuinely suck.

But, yes, it's been completely run into the ground that the LEM article that was passed around for years about these machines is positively, completely, and utterly bunk, to the point of "not even wrong" in a lot of places. These boxes basically are just what the technote pretty much comes out and says, IE, they're low-end Quadras with CPU upgrades, and when Apple shoved them out the door Quadras with CPU upgrades were still considered viable platforms. They should mostly be judged by that standard.

(Although I guess you could be catty and point out that at least most Quadra upgrades let you switch back to the 68040 if the emulation speed got you down...)

 

[Cory5412]

The OS itself being bad until a couple years later would definitely be a bummer. A thing I should absolutely do, and probably will, is pop an original 5200/6200 install CD in my system and run some applications on it to see what that's like, relative to how 7.6.1 and 9.1 perform on it. I might do that in the mid-near future, because my network card isn't running so the OT/AS updates needed to hop on VTools aren't exactly relevant at the moment.

One thing that I'm interested in is if we've got an exhaustive list of what was in the bundles on all the variants of these things when they shipped, and whether those things were native/fat or 68k. From a "being scarred" perspective, I feel like (and, granted, I mean, I'm 31 today so if my family had one of these things at launch I would have been 7 and the understanding of computing a 7-year-old can have vs. a 30-something can have is very different) "something the machine came with" is unusable is pretty different than "something we borrowed or bought" is unusable.

That doesn't excuse it and to my knowledge Apple added no disclaimers or other notes about the potential for 68k-native software to have poor performance, but even if something is "clear" it doesn't mean it will be understood, see Microsoft Surface RT being marketed as "A PC!" for a more recent example.

As to why I bring it up so often: 1) it continues to get asked. 2) I want it to be abundantly clear to anybody reading this from afar who might have one of these machines or be able to find one, but, not, say, a Quadra 950 or a Power Mac 9500 that these machines are worthwhile both in 1995 and today.

From a 1995 perspective, yeah, if you were upgrading from something and the new machine you bought ran your particular apps (in this case, it appears to be that "games" is the only class of application impacted in any noteworthy way) much worse than your old one, you'd be unhappy. The 6200 wasn't really marketed to upgrading Mac buyers, but that wouldn't have stopped them from getting one.

From a 2019 perspective, it's not at all difficult to put a newer OS, speed-doubler, and maximize the use of PPC native apps and do whatever puttering you wanted to on this machine, about as well as any other vintage Mac would've done it. A corner of an edge of us are actually using these vintage Macs for something a 6200 would have a real hard time doing, and fewer of us than that group are *actually* doing those things, rather than just kind of touring them.

 

[CC_333]

I hate to add fuel to this fire, but I noticed last night that the 603 section in Wikipedia's PowerPC 601 article references the beyond-wrong LEM article as pretty much the sole reason why the 603 was so unpopular (in fact, according to the Wikipedia article, the 603 was the first to fully implement the PowerPC architecture). In reality, the 603 is actually pretty good, and the only reason it wasn't so great in the 6200 was the tiny cache, which the subsequent 603e and 603ev fixed, along with a die shrink so the whole works could be clocked higher. This has been discussed ad nauseum here, of course, so I won't blather about it anymore; I guess i just wanted to point out that the LEM article apparently has been around long enough that it has become quite pervasive, and thus there are many references to it (some of which aren't where we'd typically expect them) which assume it is fact when it isn't.

c

 

[Cory5412]

What's worse is if someone tries to go fix it, some rando wiki editor who knows literal zero things about Macs is going to revert it and point directly at LEM as the authoritative source about the 6200 as a platform, even though we have Apple's own documentation about the platform.

In reality, I hadn't thought of going to make those changes myself, but, I mean, I could certainly give it a go.

 

[Dog Cow]

What's worse is if someone tries to go fix it, some rando wiki editor who knows literal zero things about Macs is going to revert it and point directly at LEM as the authoritative source about the 6200 as a platform, even though we have Apple's own documentation about the platform.

Why not write the correct article in the Wiki here at 68kmla, then cite it in that Wikipedia article in place of the faulty article?

 

[Cory5412]

That's absolutely a good beginning point. I figured it would be easier to cite the actual develoepr notes, which we do have, they exist, etc etc.

The other thing to remember, I'm sure everybody knows this, is that it's not just one article on LEM, it's like five articles on LEM and almost every subsequent fluff piece on bad apple products includes this misinformation.

 

[Dog Cow]

The other thing to remember, I'm sure everybody knows this, is that it's not just one article on LEM, it's like five articles on LEM and almost every subsequent fluff piece on bad apple products includes this misinformation.

Yeah, but ya gotta start somewhere. 

 

[CC_333]

Yeah, but ya gotta start somewhere. 

Truer words have never been spoken!!

c

 

[Dog Cow]

 I figured it would be easier to cite the actual develoepr notes, which we do have, they exist, etc etc.

Oh you mean cite the Apple dev notes in the Wikipedia article? I think that's even better.

 

[dtaylor]

Nothing outside of supercomputers ran external cache at CPU speed, ever, at least not once CPU speed started to decouple from bus speeds in 1990-something. No way Apple was going to pay for 100MHz SRAMs for a bargain basement computer.

The access timing of RAM is a distinct issue from the speed of the bus that it's on. You'll find many examples...especially among early Macs...of the same speed RAM (cache or main memory) being used across computers with different bus speeds. It's possible that the developer note is incorrect and that the 603 bus ran at 37.5 MHz (2:1). But L2 cache access timing doesn't prove this so.

Case in point: many Macs at the time used 80ns RAM SIMMs which were interchangeable across machines, including these Performas. If RAM access timing strictly dictated bus speed then they would all have had 12.5 MHz system buses.

According to Motorola's user manual for the 603, this is a non-standard and unsupported configuration: the 603 is designed to only run at 1, 2, 3, or 4x bus speed, with a variety of supported bus speeds. However, 37.5 is not one of them; the 75MHz part is designed to be run at 3x a 25MHz bus. while the fastest 1:1 speed available for the 603 is 66MHz.

Running a 75 MHz part (or an 80 MHz part @ 75 MHz) at 2:1 for a bus speed of 37.5 MHz is a non issue. Just because Motorola didn't list it as an example does not mean it couldn't be done or was outside their design specs. I can think of plenty of examples of part manufacturers listing common configurations and system manufacturers tweaking the speeds up/down a little bit.

Having said that: I find it curious that the 603e manual also lists the highest external bus speed as 66 MHz. I could dismiss the 603 manual as being written before newer parts, including newer versions of the 603, were available. But the 603e manual suggests 66 MHz was a 'hard limit' at the time for these processors. If that's true then naturally the external 603 bus could not run at 1:1 with a CPU running at 75, 100, or 120 MHz.

I'm going to do a little more research, but my gut feeling is that you are correct on this point and that both the 603 and 68040 buses ran at a multiple of the core CPU speed. I'll add notes to the blog posts if that's the case. Good catch!

Also: if anyone can document any other technical errors in either of my articles, please let me know. I will gladly correct them.

 

[dtaylor]

I have a 6220/75 at home, as I've said, it's "fine" with PPC native code -- I've run 7.6.1 and 9.1 with a couple different apps, which TBH almost everything I happen to use is, I'll pull out some older 68k versions of things and poke around with it at that point.

I nearly grabbed a 6200 on Craigslist one time because I very much wanted to test it against the 6300 and the 6100/66 on PPC and 68K code. Lost the chance and haven't seen one locally again.

TBH I'm very curious about how the 6200 did perform in general (since most of Mac OS at the time was still 68K) and with 68K applications.