Having recently acquired a Color Classic, I decided that my first task, after making sure it had survived the rigors of the Postal Service's package abuse, was to start hacking it apart as much as possible. No, I don't mean literally, but it felt like that about 30 minutes into disassembling it for the first time.
Why was I disassembling a perfectly functional computer? Why, to perform the famed High-Res modification, of course. The whole 512x384 thing just wasn't enough for me, and besides, I'm an incurable hardware hacker at heart. After successfully performing the upgrade and making the necessary adjustments to the CRT, I had a nice Color Classic (CC) with a 640x480 screen.
That was all well and good, but the stock CC was just too slow for my taste, even after I dropped in a 16 MHz 68882 FPU in the FPU slot. How slow is too slow? I decided to run Speedometer 4.02 on it. (CC, 10 MB RAM, stock 80 MB HD, System 7.1 Update 3.) It wasn't pretty:
KWhetstones: 0.156 Matrix multiply: 0.093 Fast Fourier: 0.065 Average: 0.105
The 68882 is ridiculously weak compared to the 68040's built-in FPU, and the CC is further inhibited by its 16 MHz speed and 16-bit bus, so this wasn't entirely a surprise.
CPU: 0.250 Graphics: 0.224 Disk: 0.911 Math: 1.941
The bright spot here is Math, which ranks at almost twice the speed of a Quadra 605. What you don't notice right away, though, is that the Q605 didn't have an FPU while this CC does. Ouch. The disk test can't be directly compared to a Q605 since the stock hard drive differs between the two, but I will come back to this later.
KWhetstones 0.644 Dhrystones 0.130 Towers 0.129 Quicksort 0.212 Bubble sort 0.252 Queens 0.225 Puzzle 0.237 Permutations 0.212 Integer Matrix 0.214 Sieve 0.278 Average 0.253
As I noted in my article about clock-chipping the Q605, the Whetstones benchmark appears to be heavily FPU-dependent, which explains why this CC scored about three times higher than its average for the other tests on that particular benchmark. The rest of the scores are pretty dismal — but expected for a 16 MHz '030 on a 16-bit bus.
1-bit: 0.247 2-bit: 0.245 4-bit: 0.257 8-bit: 0.264 16-bit: n/a (not enough VRAM with video mod) Average: 0.253
As expected, graphics scores were nothing spectacular either.
Where it gets interesting, however, is when you do the easiest CC upgrade there is — installing a motherboard from an LC 575 in place of the CC board. You need a new back panel (I made one from an old LC 575 panel), and if you don't do the VGA mod, you'll need to modify the board. System Enabler 065 v1.1 is required for using an unmodified LC 575 board.
You won't be able to use the CC's old RAM — the 575 uses 72-pin SIMMs — and the VRAM needs to be 80ns or faster to work in the 575, but that's a small price to pay for this seemingly minor upgrade. Since I had a spare 25 MHz 68040 chip at home, I decided to go ahead and do the total package, swapping out the 33 MHz LC040 for the 25 MHz full 040 and putting a heat sink on the new CPU.
Note that there aren't any heat sink clips on the LC 575's CPU socket, so you'll have to either epoxy the heat sink on (with appropriate heat-conductive epoxy) or do as I did — put a thin layer of heat sink grease on the bottom of the heat sink and let gravity hold it in place. I don't plan on moving the Mystic much once I've finished with it, so I'm not too worried that the heat sink isn't fully attached. Note also that not all 25 MHz 040s will work at 33 MHz; the same cautions apply here as applied in my Chipping the Q605 article. After running a very CPU/FPU-intensive stress test this afternoon, however — I ran 1000 iterations of the FPU benchmarks in Speedometer 4 — I'm convinced this one works fine.
Speaking of benchmarks, let's take a look at how this thing performs. Note that I've got 2 MB more RAM (12 MB vs. 10 MB), but everything else is the same as before.
FPU, 10 iterations
before after change KWhet 0.156 1.038 6.65x Matrix 0.093 0.987 10.61x Fast Fourier 0.065 0.981 14.09x avg. 0.105 1.002 8.54x
There's a 10x performance increase right there, and is now on par with a Quadra 650 (not really a huge surprise since both the CC/040 and the 650 are 33 MHz 040s). Next?
before after change CPU: 0.250 1.163 4.65x Graphics: 0.224 1.299 5.80x Disk: 0.911 0.971 1.08x Math: 1.941 19.711 10.16x
This is at worst a 5x increase (excluding Disk) and in the Math, again, almost a 10x increase. I like these numbers; keep 'em comin'! Even the Disk benchmark showed about a seven percent increase. Remember, this is the exact same hard drive as before. I'm impressed.
KWhetstones 6.499 Dhrystones 1.432 Towers 1.348 Quicksort 1.322 Bubble sort 1.327 Queens 1.298 Puzzle 1.344 Permutations 1.297 Int Matrix 1.323 Sieve 1.255 avg. 1.844
It just keeps getting better and better. These are, again, anywhere from 6x to about 10x increases in performance. Bring on the graphics tests!
1-bit: 1.233 2-bit: 1.207 4-bit: 1.213 8-bit: 1.232 16-bit: 1.302 avg.: 1.238
I can't say how this compares to a stock LC 575, but I do know it's about 6x faster than the stock CC was.
All in all, I'd say this was about the best Mac upgrade I've ever done. It cost me nothing (Thanks, Jeff H.!) except 5 minutes of my time to swap in the full 040 and then swap the boards. If you've got a Color Classic and you're wanting to improve its performance, this is by far the easiest and fastest way to do it. Cost-wise, you can pick up a used 62xx or 63xx Power Mac for less money than an LC 575 board (in most cases) but you'll spend hours fitting the Power Mac board into the Color Classic.
For more information about Color Classic upgrades, check out The Colo(u)r Classic Upgrade Mega-FAQ.
copyright ©2000-2004 by Chris Lawson