A very late 040

[jessenator]

jessenator tells me 9 isn't great on the 7200/75

I'd be interested to see the this test's results on it! But yes, on a full, non-clean install of 9.x , the 7200/75 was a bit of a dog, with and without L2 cache, but I didn't run the numbers formally. Frankly I got sick of how friggin long it took to boot, so I wiped it and put 7.6.1 on it.

I think I'll test my 603ev StarMax that can underclock to 120 (125) and see how it does as well.  So that's Norton Utilities Speed Test v2.0 that we're using? IDR, is that plopped onto vTools?

 

[Cory5412]

yeah, there's a folder for it on vtools. It's in a disk image and just on a folder. I copied the folder to the desktop of my 6200 to run it, my own additions to the bench pool are outside the folder but I think they should probably be inside it. You'll see like "coryw-6200" and "6200-SD8".

There's a notes field too, so it might be possible to standardize on a shorter format and then put meta/context information in there.

The Starmax at 120 should do okay, I expect to come out around the middle of the pack, the main question basically being how much faster it is than the NuBus machines and the 6300. I'd expect it to maybe beat out a 7200/120 + cache  by a bit and be probably a bit slower than a 604 though, depending on the exact test/sub-test.

I'm collecting some benches here: https://doku.stenoweb.net/doku.php?id=macdex:nortonsysteminfobenches

I need to come up with a good way to document the configurations.

If you'd like to have something added, the text format is generally:

^ System    ^ System Rating  ^ CPU    ^ FPU    ^ Video  ^ Disk  ^
| Q700      | 100            | 100    | 100    | 100    | 100   |

The exact number of spaces doesn't matter. Just sub in the Q700 for your results.

 

[jessenator]

I'm collecting some benches here

Other specs in the system description shown for disclosure purposes. Feel free to include or omit...

| StarMax 5000 (603ev@125 MHz + no L2 + CF + Rage128) | 184 | 200 | 19.1 | 142 | 305
| StarMax 5000 (603ev@125 MHz + 256k L2 + CF + Rage128) | 245 | 267 | 19.9 | 203 | 424
| StarMax 5000 (603ev@125 MHz + 512k L2 + CF + Rage128) | 253 | 277 | 19.9 | 207 | 439

 

[Unknown_K]

Quadra 950 with 68040/50+ cache

Video is a Nubus cards so the bench sucks.

HD is a native 50 pin SCSI drive.

CPU: 186

FPU: 203

Video: 65.6

Disk: 177

 

[Unknown_K]

Radius 81/110

CPU: 83.9

FPU: 15.7

Video 89.4

Disk: 170

 

[johnklos]

What version of Norton System Information are you running? I have version 3.5.3 here, and by default the PowerMac 6100/60 is the reference system.

 

[NJRoadfan]

I think 3.5 is PPC native (I have it on my 6100 too), so benchmarks aren't going to be directly comparable.

 

[johnklos]

I think 3.5 is PPC native (I have it on my 6100 too), so benchmarks aren't going to be directly comparable.

3.5 is PPC and m68k, so m68k compared with m68k will be, well, comparable

I just figured out that System Info version 3.2.1 shows the Quadra 700 as the baseline and will open benchmark results made with 3.5. My Amiga 1200 when the m68060 was running at 57 MHz:

System rating: 247

CPU: 255

Video: 210

Disk: 243

FPU: 429

 

[Cory5412]

Hmm, I thought I wrote it down, but, I must not have!

*boots 6200*

I'm running Symantec System Info 3.1, which MrFahrenheit uploaded to vtools. Please PM me if you'd like a vtools account, there's folders for norton system info bench collection and MacBench 4 bench collection.

System Info I'm using exclusively to collect information on relative 68k emulation performance. MacBench 4 is where we already have more results and does a much better job comprehensively measuring and comparing system performance with 1994-1998-era FAT software.

To be honest, it wouldn't bother me not to get the disk performance, I don't really know if it means anything, but the test runs it by default and it's not really that much of a bother so, no big deal.

Basically, the thing I wanted to see is, exactly how bad did the 16k L1 on the 6200 hamper 68k emulation performance and how much can it be saved by speed doubler. The answer is "a fair bit" and "quite a lot" respectively. I'd be interested in seeing a 6300's results, and, perhaps this weekend I'll have the energy to pull out my 7200 and compare that as well.

The other results I'd be seeing are some slower NuBus PPC Macs in various configs, an x100 at 60, 66, and 80, perhaps. Ideally with some different cache configs if possible, and with/without Speed Doubler 8.1.2. I have a 1400c/166, I'll run it on there as well.

MacBench also allows you more granularity with what it does and how and does things like repeating benchmarks and has fairly good built-in comparison tools. I find the disk benchmarks comparable with modern disk benches in terms of comparing different file sizes and with random/sequential writing, although it doesn't create the pretty speed graph that some of the vintage disk bench tools do.

 

[Cory5412]

I just added LibMoto back to my 6200/s 7.6.1 system folder and only got one or two points of difference, with LibMoto being slightly better.

Also, if found out that if you go to the Benchmarks, then click on Show Detailed Ratings, then choose "Absolute' In the menu you can get a tabular output, which will make transcribing results way easier.

I might re-arrange this but for now, the benchmark results are going into a common folder on vtools:

Public (share) -> NortonSysInfo-Benches -> Norton System Info -> Benchmark Results

Sync the Benchmark Results folder with your own and when you open System Info (in the Norton System Info folder) you'll be able to see what people have put up so far. You can select a specific system to see configuration details, as well.

 

[Franklinstein]

Basically, the thing I wanted to see is, exactly how bad did the 16k L1 on the 6200 hamper 68k emulation performance and how much can it be saved by speed doubler. The answer is "a fair bit" and "quite a lot" respectively.

The original 603 from the first 75MHz 52/62xx machines used 8k+8k split caches (the later 603e was 16k+16k) which was too small for the Apple 68k emulator to fit into, meaning it had to be run from slower off-chip memory instead, hence the poor performance with 68k code. If the chip had used a combined cache like the 601 did they may not have had that problem. However Connectix may have been able to either squeeze their emulator into 8k or split it effectively across the two stores, or it may just have been that much more efficient (they were excellent when it came to fixing Apple's shortcomings).

There were three different logic boards for the 5x/6xxx series machines: the original 37.5/75MHz 603, the faster 40/100MHz (and possibly also 120MHz) 603e-based versions, both always with a 256k L2 cache built into the ROM SIMM, and a final 40/100 or 120MHz variant with soldered ROMs and optional L2 cache. The latter's L2 cache slot is physically identical to the x100 series Power Macs but their cache SIMMs don't work in these for some reason (either no boot or just unrecognized), nor do the earlier ROM+L2 cache cards work. I would be interested in finding an authentic late-model 5x/6xxx L2 cache or figuring out what to change to utilize an x100 model's cache here (I assume some pins need to be modified).

 

[Cory5412]

If the chip had used a combined cache like the 601 did they may not have had that problem.

Yeah, it's interesting, because the 6100 isn't actually much faster at 68k emulation than the 6200 is.

There's some results in here of a 6100/66 that got 53 at CPU and 9.5 at FPU and my 6200 (before SD8 and libmoto) got 45 and 8.7. I don't know what cache that system had, though. I think this is one of @MrFahrenheit's systems.

When you say "final" 120MHz board, is that a PCI version like the 5400/120? I had thought the 6300/120 had the 256k of L2 cache as well. A 6300/100 or 6300/120 should, overall, be way more performant than the 6200, although, any 100-120MHz 601 is also way more performant than a 60-66MHz one as well, so some of it's architecture improvements (cache, bus) and some of it's just better CPU.

There are known dedicated cache modules for the PCI architecture 5000/6000 machines, some of which have PowerPC G3 processors on them, so that's a knowable and buildable format, it's just a matter of finding one of those cards, since I bet they weren't very common on their own, and by the time you feel like it's important to buy an L2 cache to augment your cacheless 5400 or 6400, you're most of the way to being able to justify a new imac, anyway, unless it's still 1997 and you bought a cacheless model for availability reasons.

The original 603 from the first 75MHz 52/62xx machines used 8k+8k split caches (the later 603e was 16k+16k) which was too small for the Apple 68k emulator to fit into, meaning it had to be run from slower off-chip memory instead,

Yes, this is a good, accurate, detailed long-hand of "not enough L1 cache". My apologies for the confusion.

In general with architecture on these machines before The Mythical Road Apple | Taylor Design (taylordesign.net) (old version of SSL certificate warning on some browsers, chredge let me click through when I clicked on advanced) (I think this person joined us on here but I forgot the username) which has some more information about this as well. There's a couple different posts about this overall.

 

[Franklinstein]

When you say "final" 120MHz board, is that a PCI version like the 5400/120?

No, there was a late run of the 5x/6xxx that used 120MHz processors. Supposedly the Performa 5320 was one of them but there was a thread I created a while back that had the box (and the CPU itself) clearly listed as 100MHz instead. Some spec pages on the internet also list them at 100MHz yet there are resources elsewhere that claim 120Mhz, so who knows what the deal is with those things. It was a confusing time at Apple.

Anyway, these weird 100 and 120MHz soldered-ROM boards are Apple p/n 820-0751-A so you can do an image search or possibly buy one for yourself and you'll notice the cache slot is empty. I bought a mess of them on eBay once because they had a mishmosh of chips over the various boards (same chip, different suppliers) and I wanted one with the Chips and Technologies-produced Valkyrie video controller (most are AT&T or Sierra Semi). They seemed to have used color-coded heatsinks: aluminum for 100MHz, anodized blue for 120MHz (though this may not be a rule, just something I noticed with the lot I bought, so don't get mad if you buy one and find a blue heatsink on a 100MHz chip). They also typically used a Fujitsu-made Capella, a Samsung-made Primetime, and occasionally used IBM's apparently uncommon blue epoxy encapsulated 603e, which I always thought looked really pretty. It's just too bad about the weird L2 cache slot.

 

[Franklinstein]

In general with architecture on these machines before The Mythical Road Apple | Taylor Design (taylordesign.net) (old version of SSL certificate warning on some browsers, chredge let me click through when I clicked on advanced) (I think this person joined us on here but I forgot the username) which has some more information about this as well. There's a couple different posts about this overall.

Yes I'm pretty sure I'm in many of them. In a nutshell: I think the original 75MHz 5200/6200 was a bit of a misstep, or at least feels that way when compared to the outgoing 61xx (according to Everymac, the first 6200 versions were priced only about $100 less than a 61xx at introduction and the 6200 performed no better or was even slightly slower in some cases; you may have got a better deal buying a discontinued/refurbished/used 61xx bundle), but the followup models were really not terrible; I rather like my Performa 5270. Also the LEM guy who wrote the original error-filled screed against them had virtually no idea what he was talking about whereas between us that Taylor guy and I have at least a minor understanding of how these things were built and operate. However the general lack of accurate technical information available on the internet at large leaves a few gaps and makes it difficult to cite sources for anything outside of the 603 UM.

 

[Cory5412]

W/re the 6200 as a misstep:

I disagree. In the US at least, all the evidence i have been able to find indicates that the 6200 cost a lot less than the 6100 did. Especially considering that most 6200s were sold as Performa bundles. At introduction of the 6200, in the USA, pricing for a 6100 was roughly $1800 for a bare system (BYODK) and $1200-1400 for a Performa 6200 with software, keyboard, monitor, and often a printer or a modem.) In addition, a 6200/75 is, in daily operation, about the same speed as a 6100/60 is. They have very similar graphical capabilities, with the benefit slightly going to the 6200 for having had separate VRAM and a couple acceleration tricks.

(Price: MacWorld June 1995 says the 5200 runs a bit under $2000 and a 6100 will run around $2600 so that price difference is there, even if I"m misremembering the specific numbers.)

The L1 cache was absolutely a misstep, but in general, they were "fine" computers and their only crime was having been cheap.

(MacWorld also says that once you add a cache to a 6100 it becomes faster, and, I believe that, but I don't have one handy to bench, and even so, implicitly Apple was "fine" with that level of performance, and, to be honest, it works well enough for "basic computing" kind of stuff. The graphs in this article form 1995 show a 6100/60 with cache slightly beating the 5200/75.)

It would absolutely have been better if Apple designed a new platform for the 6200, even if it were still not a PCI platform, but a lot of what goes into making a product, or at least what did in that moment, was compromising what you can do for pretty cheap to meet a need and what will be performant, and so you end up (in Macs as well as PCs) with solutions that aren't the most elegant. This is comparable to the Yikes/PCI Graphics PowerMac G4, except that the blue-and-white PowerMac G3 was a better platform up front than the 630 was, since it was the high end machine.

The other thing is, Apple was really bad at naming products and I think that it's a misnomer to suggest that the 6200 replaced or succeeds the 6100. I think that in reality, the 7200 succeeds the 6100, the 7500 succeeds the 7100 and the 8500 of course succeeds the 8100. (the 8200 is a vertical 7200 and so it sits in the same slot below the 8500, beside the 7200, in markets where it existed). What the 6200 replaced was the 630, a machine that was still a 68k based computer and introduced after the 6100, to serve as a low end and nominally to wrap up the lives of several 68k models that had continued existing because Apple couldn't manage its product line to save its life from 1987 to 1997.

That there were Performa 6100s and not Performa 7200s is mostly a side-effect of, just, again, Apple being bad at this whole game. Either the 6200 was late, causing the Performa 6110 series or they had some demand for a performance machine and decided the 6100 was good enough, even though it technically lacked some things the 630 and 6200 had, like the TV/FM system.

An interesting speculative question might be whether or not it would be reasonable/worthwhile for apple to have just tried to reduce cost on the 6100, but, the 6100 really was originally intended as a "professional" computer. A low end one, but a professional one nevertheless, and the 630 and 6200 get their origins in being a cheap consumer-oriented design, and basically until 1998, Apple (and most of the computer industry, some still does) believed that it was okay to provide older and explicit cost-saving designs to consumers. This even carries through in the 6400 (versus the contemporary 7300/7600/8600/9600) where the 6400 uses IDE and doesn't have onboard ethernet and has a lower-end graphics system that supports a couple fewer legacy formats/options.

That, or, as with '030s in the early '90s (and, as with the other '040s out until 1996) is there an argument to be potentially made here that Apple should have not built the 5200/6200 and continued selling the 640 as-is for another year.

I think that's a toss-up, and, while Apple is incompatible with discontinuing products in a reasonable time, they're were attracted to introducing new products like moths to a flame, so, I feel like the 6200 was a little bit inevitable, in that sense.

 

[Cory5412]

Also the LEM guy who wrote the original error-filled screed against them had virtually no idea what he was talking about whereas between us that Taylor guy and I have at least a minor understanding of how these things were built and operate.

Yeah. Low End Mac, a number of different contributors, absolutely poisoned the well on this machine. It's a fine, if cheap computer, that does poorly at a specific thing that might have been commonly needed in the mid '90s and people spent several years not wanting to touch them because of their status as "the worst Mac" which only barely made sense in the original of Low End Mac, which was as a used Mac buying guide in the late '90s when some of these things would still have been with their first owners. (Hell, DK only got started writing LEM a year after the last of the 68k Macs had finally stopped being built.)

At some point, I've been told I should just go rewrite the wikipedia page on the 6200/5200, and, I should, but that's a big undertaking. I've been meaning to do my own page about 6200 performance on my own wiki, to at least have something to send people, but i haven't had time to author it yet.

 

[Unknown_K]

Well during the later 90's when the internet was more wide spread and used by the masses a 6200 would have been obsolete anyway.

Sure maybe the article kept resell values lower then they might have been but big deal.

 

[Cory5412]

The big deal isn't what happened to the 6200 in 1997 (at only two years old) -- it's what happened to it from 2003 to now. Even among people who don't have any reason to think about this, you see all the time on social and in discussions "oh, those are bad, avoid them" when the 6200 gets mentioned, from people who read LEM once a decade ago and never thought that "left 32, right 32" was literally nonsense. 

A 6200 would have been "fine" online. Not great, merely fine. To be honest, they accept bigger (cheap, IDE) hard disks and use fairly obtainable RAM so in 1997-2002 or so a 6200 would have been fine for basic Internet-faring computing too.

(I mean, I was using an 840av with 1998-era software at varying times between 2001 and 2005, a 6200 should run all of that stuff that's available as PPC native better.)

Now, if it's 1999 and you're in a Computer Renaissance and a 6200 and 7200 are sitting next to each other and they have the same config and cost the same, you probably should get the 7200, but I bet that wasn't really a very common situation.

Professionally speaking, Dan Knight, LEM's founder and editor styles himself as a technical writer and allowing the Power Macintosh 6200/5200 pages to sit on the site for as long as they have, uncorrected until a month or two ago, is a huge oversight. If the dates on this are correct, the machine was only two years old when these accusations were lobbied against it, and then they sat, being reinforced by not getting updated (and, more people wrote further articles with further outlandish claims) for 23 years.

It's not about resale values, it's about a shared perception of reality being damaged by people who didn't understand what the things they were writing meant.

 

[Franklinstein]

From about 2001-2003 I alternated between using a Performa 475, PM 6100/60, and a 6200/75 on the internet (with a dial-up modem, mostly with NetZero). They were still usable, especially for AIM and light web use, though the G3MT and blueberry iBook I bought at the end of '03 were quite a bit better, especially once Flash and JavaScript pretty much took over everything. If I could still use the AIM client for 68k Macs, I would.

On topic, I recently bought a Performa 640 DOS because a; it has the crazy DOS subsystem that plugs into the '040 socket and PDS slot, and b; it's one of the good dual-slot RAM models. The Pioneer MPC-LX100 also used these dual-slot RAM boards (though AFAIK they didn't offer a DOS variant) but they used a full '040 instead of the typical LC chip found in most 6xx-based machines.