Contents:Prozessortuning si
Quelle:Abgelegt im Internet: IISI-25Mhz-upgrade-FAQ-txt
From: eepmatt@casbah.acns.nwu.edu (Matthew Friedman)
Subject: The 25MHz IIsi
Date: Sat, 27 Feb 93 0:49:16 CST
Here's the FAQ sheet I compiled on the 25MHz upgrade for the IIsi. A couple of people requested that I send it to you for archiving in your reports section. This FAQ was compiled from comments, letters, and posts to comp.sys.mac.hardware. I apologise for the many instances where I have not been able to include attributions and sources. Maybe in version 2.0.... :)
Contents:
Q: What does the "upgrade" do?A: The upgrade involves swapping the 40MHz oscillator in your IIsi for a 50MHz one, thus increasing the CPU speed by 25%, from 20MHz to 25MHz. Some systems speeds, like displaying 8 bit graphics, have been noted to improve even more than 25%. All this for around $4.11.
Q: What are the stats on the performance increase?
A: Numbers below are speed relative to a Mac classic, so bigger is better:
| BEFORE | AFTER | ||
|---|---|---|---|
| CPU | 5.24 | 6.69 | |
| Graphics | 6.17 | 7.64 | |
| Disk | 1.30 | 1.34 | |
| Math | 5.45 | 6.91 | |
| Overall Performance | 4.75 | 5.92 | (25% faster) |
| KWhetstones | 6.33 | 7.97 | |
| Dhrystones | 5.11 | 6.62 | |
| Towers | 4.42 | 5.69 | |
| Quicksort | 5.01 | 6.43 | |
| Bubble Sort | 5.88 | 7.50 | |
| Queens | 5.83 | 7.33 | |
| Puzzle | 5.61 | 7.22 | |
| Permute | 5.33 | 6.55 | |
| Fast Fourier | 4.27 | 5.39 | |
| F.P. Matrix | 4.53 | 5.70 | |
| Int. Matrix | 4.85 | 6.09 | |
| Sieve | 6.53 | 8.35 | |
| BENCHMARK AVG. | 5.31 | 6.74 | |
| Graphics: | |||
| 1 bit (mono) | 1.72 | 2.15 | |
| 2 bit | 1.83 | 2.31 | |
| 4 bit | 1.92 | 2.47 | |
| 8 bit (256 colors) | 1.22 | 1.89 | (50% FASTER!!!!) |
| Avg: | 1.67 | 2.21 |
Q: Why does the upgrade work?
A: Well, a caveat first. Most of what follows is conjecture, so while it _does_ make sense, take it with a grain of salt rather than a shovel full of earth. Only Apple's engineers know for sure, and they ain't saying. The main difference between the IIsi and the more expensive IIci is expandability and speed. But aside from these differences, it's been asserted that the guts of the two machines are more or less identical.If you think about it, it does make engineering sense to reuse as much of the IIci's design as possible. With chip prices falling these days, it might be cheaper to use essentially the same board and chips in the two machines. Putting a 20MHz CPU in the IIsi would then be a smart marketing decision -- would you buy a IIci for $3500 if you could buy an equally fast IIsi for $500 or $1000 less and sacrifice only expandability? So the IIsi may have been "crippled" for marketing reasons, slowed down to allow the price of its faster sister to be raised. Again, the above paragraphs may have no basis in fact, and are really only rumours and whispers made over the net. For all we know, the IIsi chips were designed to run at 20MHz which resulted in considerable savings which was passed on to the consumer. Its up to you to draw your own conclusions from the reasoning and testimonials of those who've made successful upgrades (given below).
Q: Won't this void my warranty?
A: Yes, absolutely. In fact, some Apple service technicians will refuse to work on any machine that shows any evidence of user tampering at all. If you attempt this upgrade, you may need to find a new service outlet.
Q: Where do I get what I need, and how much will it cost?
A: You'll need a soldering "pencil" (preferably >30W), and its strongly recommended to have a desoldering iron or a solder sucker, such as the Soldapullt Model DS 017, as well. Thanks to James MacPhail for pointing out that most soldering guns are step-down transformers that generate heat by passing a large AC current through the tip. While this generates lots of heat, there is also a large magnetic field at the tip. Some kinds of electronics are destroyed by strong magnetic fields (similarly with electro-static discharge) so in most cases, a "soldering gun" is a no-no for this kind of job. Use something that does not build up a large magnetic field. You'll also need something small to pry with. These tools you can get at any local Radio Shack. You'll need to install a heat sink as well. While this may not be mandatory for every single machine, some people have reported that the upgrade worked only after installing a heat sink on the CPU, so better safe than sorry. Nothing fancy is necessary -- just about anything the size of the 68030 will do. You'll also need something to attach the heat sink with: thermal tape and thermal glue have been reported to do the job handsomely. Also required is a 50MHz TTL oscillator package and a 14-pin DIP IC socket for the oscillator (you may need to go back to 20MHz if your upgrade fails). Fry's electronics in Palo Alto, CA is one place that sells the oscillator package and socket. If you walk in off the street it costs around $4.00 for the oscillator, and the socket will put you back about 11 cents. Mail order from Fry's is considerably more expensive, possibly by a factor of two to three times. Their phone number is 415-496-6000. Fry's has another outlet in Fremont at 510-770-3797. Or, you can fax them: 408-735-6800 in Sunnyvale CA, 415-496-6060 in Palo Alto CA, and 510-770-3700 in Fremont CA. Other suppliers include B.G. Micro: $1.49 for the oscillator and $3.25 for postage. $10 minimum for MC/Visa. P.O. Box 280298, Dallas, TX 75228 (214) 271-5546. Digi-Key will also take MC/Visa, the part number is CTX121, $3.44/ea + $5 handling and actual shipping charge for orders under $25 only. 1-800-Digi-Key. I have no info on whether these two sell the sockets also.
Q: So how to I do it?
A: Here's the procedure. It's an amalgam of the information posted by Jim K. H. Yu and George John, who attributes the pioneering of this procedure to "(forgot his name) at CalTech and Paul A. in Australia."
Warning: The soldering iron stuff comes next. If you have never used a soldering iron before, DON'T START NOW! Virtually all of the people who have attempted this upgrade have reported the soldering to be extremely difficult and dangerous. Have an experienced solderer with you to help. The IIsi's motherboard is a multi-layered board and can easily be damaged by excessive heat or force. If you pull too hard, you can ruin the contacts between the crystal and the motherboard, and then you might as well buy a new mac.
To help with desoldering you might consider using a solder-sucker, or a soldering wick to help draw away the old solder. One person suggested cutting the crystal's pins, so you can remove them one by one. Of course, if you plan to use this method of attack, order a new 40MHz crystal along with your faster one in case the upgrade doesn't work for you.
If you're a little more ambitious (or daring) you might try this variation suggested by James MacPhail:"I tried hooking up the 57.2832 MHz on-board osc. Initially I found it did not work with my 80ns 4M simms, but did work with 80ns 2M simms. A while later, the 2M simms didn't work either. I found that the bottleneck was the startup test. I wired in a switch on the cover for the expansion slot card connector opening which selects the 40 or 57.3 MHz output. I start-up at 40 MHz, drop into MacsBug, switch to 57 MHz (the machine crashes when the clock is switched), and Reset.
"After installing the 40/57 MHz switch, the 2M 80ns and 4M 80ns SIMMs worked fine. (I speculate that the startup test uses the 'test mode' of the SIMMs, as it has different timing requirements) The machine works fine thereafter at the higher speed. (At leer two terminals connected to that pin of the 40 and 57 MHz oscs. I used wire-wrap wire, trying to keep it as short as possible (each piece is about 5" long).
"Since one doesn't want to introduce static-electricity inside the case, it is best to keep the wires inside the case, and have the switch toggle outside the case. ast it has worked fine for about 60 hours continuously).
"I used a small SPDT toggle switch, with the center terminal connected to the motherboard at pin "8" of the 40MHz osc position, adn the other two terminals connected to that pin of the 40 and 57 MHz oscs. I used wire-wrap wire, trying to keep it as short as possible (each piece is about 5" long).
"Since one doesn't want to introduce static-electricity inside the case, it is best to keep the wires inside the case, and have the switch toggle outside the case.
I also have added interrupt and reset buttons to the back of my IIsi (these work even when the keyboard doesn't). I don't know if the keyboard reset will work when the clock is switched..." (email faq author for these instructions....)
Q: My CPU's only rated at 20MHz. Can the upgrade still work?
A: Possibly. In fact, virtually all of the reported successes have been on machines with CPU's rated at only 20MHz. However, it's possible that a percentage of those CPU's won't work at 25MHz. The practice of engineering products with margins of safety to cope with random variations in component characteristics has been normal practise for many years. Very few CPU's rated at 20MHz will work _only_ at 20MHz. Without this buffer zone, it would be possible for a CPU rated at 20MHz to have occasional errors when the operating environment (such as temperature) fluctuated. Without this safety zone, a CPU operating at its rated speed might occasionally fail. This means that in some instances the upgrade might be pushing the CPU to its absolute limit, removing some if not all of this safety margin. Computers that appeared to work flawlessly in the colder winter months might possibly begin to exhibit failures on warmer summer days, or when other environmental variables begin to change. On the other hand, your particular chip may have a safety margin that can comfortably handle the increase. Your 20MHz 68030 might share the same design as the 25MHz 68030, but due to irregularities in the silicon wafer and fabrication process, just didn't pass a parameter tests at 25MHz (plus the 25MHz safety allowance). Such chips are then retested at 20MHz (or even 16MHz) and sold as such. Marketing may also play a roll here. If demand for 20MHz CPUs is higher than fabrication yield, some chips that might have achieved a 25MHz rating could simply have never been tested beyond the 20MHz range. What this means is that some people may find that the upgrade won't work at all, while others may find that a heat sink solves the problem (if perhaps the only failed parameter test is related to excessive heat) and still others will have no problem at all. To check and see what speed your CPU is rated, pop the hood and look at the square chip near the bank of four vertical plug-in memory modules. The CPU will have a label like MC68030fe20b, where the 20 represents the highest speed rating that particular chip was successfully tested at. A couple of people have reported that their si's have actually come with CPU's rated at 25MHz installed.
Q: Why donUt I just replace my CPU with one rated at 25MHz?
A: (From James MacPhail) The answer is that it is a much more difficult and expensive operation, and almost certainly requires expensive surface-mount rework equipment. (And will probably destroy the old 68030 in the removal unless the right equipment is available).
Q: Can I go higher than 25MHz?
A: There were two reports that people had gone higher. One reported 27MHz off of a 54MHz crystal, and another claimed to be running off of a 59.4MHz crystal. However, do remember too that the assumption this upgrade bases its validity on is that the IIsi is essentially the same as the 25MHz IIci-- going past that speed may be really pushing an uncertain thing as it is.
Q: Do I have to have SIMMs with a certain speed for this to work?
A: Nobody has tested whether a successful upgrade can be foiled by putting in slower SIMMs, however the logic is appealing. The SIMMs in the IIci are 80ns chips while the SIMMs in the early IIsi's are 100ns. This seems to followalong with the 25% increase in speed the upgrades gives. 80ns = 25% faster than 100ns. More recent si's have included 80ns chips. Keep in mind that even though you may have upgraded your ram, there are still SIMMs soldered directly onto the motherboard which may be rated at only 100ns.
Q: What if I have a NuBus card or a FPU.
A: These seem to be the wild cards in most people's upgrade attempts. Some successful upgrades have been made to systems with NuBus cards, but on the other hand, most of the failures have been experienced on systems with NuBus cards or FPU's. Nobody really seems to know why. Note that there _is_ a 40MHz oscillator in the Apple NuBus card, but this does not need to be replaced. That oscillator apparently generates signals for the NuBus only and the FPU runs off the motherboard. The fact that some third party PDS cards contain an FPU and no oscillator supports that conjecture. Also, one successfully upgraded machine with an FPU tested increased FPU performance, which would again seem to suggest the FPU takes it timing off the motherboard. But again, this is only conjecture.
Q: What's the fail/success ratio?.
A: A poll was taken by Tom Savard over comp.sys.mac.hardware. Here were the results he posted:
| RAM | RAM Speed | Slot | Result |
|---|---|---|---|
| 17 | na | NuBus w/FPU | success |
| na | na | empty | success |
| na | na | empty | success |
| 17 | na | NuBus w/FPU | success |
| 9 | na | empty | success |
| na | na | na | success |
| 5 | 80 (ns) | NuBus w/FPU | success |
| 9 | 80 | Quicksilver w/FPU | success |
| 5 | na | empty | success |
| 17 | na | 030-direct w/FPU | success * |
| na | na | NuBus w/FPU | success # |
| na | na | cache card w/FPU | fail $ |
| na | na | NuBus w/FPU | fail & |
| 17 | 80 | NuBus w/FPU | fail ! |
*__only to 22 MHz-didn't have correct crystal
#__computer locked up after 5-10 Min. w/o Heatsink
$__possibly a bad oscillator
&__did not have a heat sink installed
!__failed with and without NuBus w/FPU installed
It is important to note that any conclusions drawn from the log may be irrelevant because the data sampling is very small and represents only computer owners willing to post news to the net or email me. And, ironically, computer owners with permanently damaged systems may no longer have the means or the desire to access the net.
In all the reported failures to the net, no hardware damage occured, and the machines were restored to operable condition by reinserting the original 40MHz crystal back into the socket. (This is the reason for not soldering the new crystal directly onto the motherboard.)
Q: Why doesn't someone just check the chips' speed ratings?
A: Timing calculations must be done for each and every signal pathway in the entire machine because if _any_ signal pathway is too slow for the faster clock, then you cannot guarantee that the upgrade will work reliably. Furthermore, some of the chips are almost certainly Apple-proprietary, e.g. custom PALs and/or ASICs, so a data sheet won't be readily available for them. That makes it even harder to guarantee that all timing is within spec.
Q: Does the upgrade increase the amount of heat produced by the si, and is this a bad thing?
A: Yes and yes, if untreated. Running the si at a faster clockspeed will increase the heat produced. P=(I^2)/R (power equals current squared divided by resistance) and the upgrade increases the current by 25%, which will increase the power by... well... at least 25%. Is this a problem? Untreated it could shorten the life of the CPU. Also, the increased heat might cause more processing errors, depending on the quality of your CPU. But the general net consensus seems to be that no permanent damage could befall your hardware if you install a heat sink. Whether you will encounter software errors or not is another question altogether.
Q: Are there any other drawbacks to the procedure?
A: Yes. One area that really hasn't been adequately tested is regarding machines using LocalTalk (this is an area where accelerator boards usually cause difficulties.) So far there has been one report that the speeded up si works fine on LocalTalk, and none claiming problems. Another area where accelerator boards often have problems is formatting floppy drives, but again there have been nothing but positive reports from those with 25MHz machines in this area as well. Probably the largest drawback, though, is that altering the system clock may have unexpected and disturbing effects in some instances. Worse still, its effects may be completely unpredictable. There's no worse problem than an intermittent one. This alone is a good reason to keep a 40MHz oscillator close to the computer. But the biggest drawback seems to be basic uncertainty about the process. In a nutshell, its a gamble, and on top of that no one really knows the long-run ramifications yet.
Q: What does Apple say about this?
A: Nothing. Apple probably won't ever comment on the effects of unauthorized modifications to the logic board (other than that it voids your warranty).
Q: Who should try it?
A: The general opinion on the net seems to be that people should try this upgrade if:
Q: Don't you, as the compiler, want to include a disclaimer?
A: Yes. I haven't tried this process -- just followed the discussion with interest. I haven't even made up my mind about whether or not I think it's safe. For all I know, your computer will burst into flames, taking your house, apartment building, block, city, etc., with it. But at least you'd get on the evening news.... All the people whose comments I compiled here expressed more or less the same sentiment. If you try this, its your own decision and you have no one but yourself to blame/congratulate if it fails/succeeds. But whatever the result, if you have any corrections or additions you'd like to make to this FAQ list, please email me at eepmatt@casbah.acns.nwu.edu and I'd be glad to include them in the next revision.
Good Luck,
Matt