Prozessortuning Q700
Liegt als Textfile (quadra-700-clock-mod-145.txt) im Internet
Date: Sun, 6 Jun 1993 01:10:15 -0700 (PDT)
From: Guy Kuo guykuo@u.washington.edu
Subject: quadra-700-clock-mod-145
quadra-700-clock-mod-145.txt
This procedure should only be attempted by users experienced in circuit board fabrication and repair. You must desolder a component on a multi-layer motherboard. Such boards are fragile and expensive to repair. This is NOT a good first soldering project. The process voids your warranty. If you are not dissuaded by the above, read on and see how a $5.00 part can bring your Quadra 700 up to Quadra 950 performance.
The Quadra 700 has been successfully accelerated by exchanging its CPU clock oscillator for a higher frequency unit. The original oscillator is a 50 MHz unit from which is derived the 25 MHz and 50 MHz clocks used by the 680RC40 processor. Speeds up to 33 MHz are usually attained with new oscillators. Higher speeds are attainable by some individual motherboards. One can reasonably expect to attain 30 MHz. Higher speeds are likely but not guaranteed. There is a small chance that your particular motherboard is incapable of higher speed.
Of the 33 Quadra 700's reported to or modified by myself the results are:
| 35 MHz | 2 machines (custom Fox electronics 70 MHz clocks) |
| 33 MHz | 17 machines. Two of which reported as unstable until cooling fan added |
| 32.5 MHz | 12 machines (had 65 MHz clocks available at low cost) |
| 31.5 MHz | 1 machine required cooling fan to operate. Higher speed testing pending. |
| Failed | 1 machine failed at 33 MHz. Lower speeds not yet tested |
| 1 machine with a clip on clock oscillator failed at 32 MHz to properly access floppy drive |
Motherboard destructions: 0
Motherboard damages: 2 episodes of plate through hole damage which the users managed to solder through.
Unusual problems: 1 motherboard shorted against the case during reassembly. A piece of paper between the motherboard and case solve the problem.
33 MHz exceeds the manufacturer expected performance of the Newer Technology Variable Speed Overdrive. Newer guarantees 30 MHz as attainable by all. Of the five VSO's in Seattle I know about, 31 MHz has been the recommended speed. Personally, I have run a motherboard with full stability (except 24 bit video) at over 35 MHz. 24 bit video instability at accelerated speeds prompted me to swap motherboards. My second motherboard has a top stable speed of 30.5 MHz with a VSO. 24 bit stability was not attained until I slowed to 30 MHz. This very same motherboard is running rock solid at 33.3 MHz & 24 bit video is working very well. Apparently, the machine finds a true clock oscillator more palatable than the VSO's synthesized clock. This suggests that speeds over 30 MHz will be more easily attained using true clock oscillators. This observation is confirmed by two other user reports.
The VSO is more expensive but does not require warranty voiding board modifications. I also market a clip on clock modification which uses a true clock oscillator which has a solder free installation. However, I recommend a true soldered installation for highest reliability. Clip on mounts are primarily for users wary of voiding their warranty. One user of my clip on had floppy access problems. It is unknown if that machine would have done better with a soldered installation, but I would not be surprised.
As new user reports arrive, I shall add them to this info file.
Although machines will run more rapidly, this modification pushes the circuits beyond normal operating speeds. Although no reported, long term failures have occurred on Quadra's due to higher speed operation, there is a possibility of shortening the lifespan of components on the motherboard. I have run my Q700 at accelerated speeds for almost 10 months and have not had a board failure. Others have run their VSO's longer. Given this, I doubt this is a very high risk. On the other hand, the machine may be more prone to system crashes. In my experience, this has not been the case. For safety reasons, do not perform this on any mission critical Macintosh.
PARTS
Clock Oscillators: (4 pin TTL or CMOS in 14 pin DIP form factor) Check that you are not receiving the half size package! Obtain speeds beginning at 50 MHz and higher as you wish to attempt. Reasonable values would be 55, 60, 62, & 66.66 MHz units. You might try higher speeds as well. The oscillators are typically less than $5.00 each mail order.
You should include the 50 MHz clock in case you damage the original oscillator or wish to plug a 50 MHz unit into your socket. The original Apple clock oscillator has an Output Enable control on pin 1. The units listed below always have output enabled. This is not a problem unless you try to use a Variable Speed Overdrive with one of the below clocks.
I obtained my clock oscillators from DigiKey 1-800-344-4539 Some of their part numbers are:
| 50 MHz TTL Clock Oscillator | (part # X121) |
| 55 MHz CMOS Clock Oscillator | (part # SE1509) |
| 60 MHz CMOS Clock Oscillator | (part # SE1510) |
| 64 MHz TTL Clock Oscillator | (part # X136) |
| 66.66 MHz TTL Clock Oscillator | (part # CTX137) |
Speeds above 66.66 MHz and up to 70 MHz are difficult to find. If you wish to try higher speeds, Fox Electronics 1-813-693-0099 can make 70 MHz clock oscillators on their "Fast Fox" program in 15 working days. The cost is about $12 each.In ten weeks their factory can make any value you wish. Reasonable values to try would be 69, 70 and 71 MHz. Ask for TTL in a 14 pin DIP sized four pin can.
Socket: Obtain a 4 pin socket which is in the same form factor as a 14 pin DIP package. If you try cutting the extra pins off a regular 14 pin socket, be absolutely sure no remnants of unused pins can short traces on the motherboard. Marc A. Tamsky helpfully suggests using a needle nose plier to push the pins out of a machine pin socket. It tried it and it work well.
I used 4 machined socket pins cut from a screw machined socket. This allows easier removal should need arise.
Cooling Fan?: A small, 12 volt muffin fan can be mounted on top of the CPU heat sink. Obtain one which has dimensions about 40 mm square for easier mounting.
Newer Technology's Variable Speed Overdrive includes a cooling fan. However, most machines with modified clock oscillators have survived without a cooling fan. James MacPhail measured a 4 degree increase was noted at 33.3 MHz. See later in this document for more thermal information. Additional cooling may not be absolutely necessary, but one Quadra 700 which I upgraded to 33 MHz failed at that speed after two hours of operation. Adding a cooling fan allowed that machine to operate reliably. If your machine crashes or locks up after several hours of operation, you may improve reliability by adding a cooling fan.
WARNING: Do not let your Quadra 700 run too long with the cover off. It needs the cover on to properly direct air past the CPU heat sink. One Apple source stated that the motherboard is known to die after 20 minutes of open air operation.
PROCEDURE
For your information the pin assignments are:
--------------
| 14 8 | 1 Ouput Enable
| | 7 Gnd/Case Gnd
| | 8 Output
| 1 7 | 14 V dc (+5)
L--------------
Tip: Put a small amount of soldering flux on the joints before desoldering. This can greatly speed the process, especially on pin 7.
WARNING: If you use plain socket pins, leave the leads of your clock oscillators long enough to keep the can from touching the pins and shorting out the circuit board!
Again, watch the orientation of the oscillator when you plug it in. It goes in the same orientation of the other clock oscillator next to your new socket. Reversal will destroy the clock oscillator.
Thanks to Rainer Menes, whose comp.sys.mac.hardware article prompted me to try this modification. Thanks also to the following for submitting reports allowing me to summarize the success rate of this procedure:
Guido PaccagnellaGood Luck to all who attempt this modification. There is a small but real risk, but you will likely reach Quadra 950 speeds or higher with less than $50 in parts. My personal Q700 at 33.3 MHz with an external memory cache benchmarks faster than a 950. I pass this information along as a very pleased techie.James MacPhail Charles Grosjean "Stuart R. Harper" Rainer Menes "Eric D. Kemp '94" Dan Winkler Rick Botman Mark Newman Holy Smokes! Doug_Steinfeld@vos.stratus.com Dustin Boyette "Marc A. Tamsky" Yushi Kaneda
Guy Kuo (guykuo@u.washington.edu)
BTW: This same type of mod works for the IIsi, IIfx, Q900, Q950, Centris 610 and Centris 650.
And now an important caveat from Rainer Menes
From: menes@statistik.tu-muenchen.de (Rainer Menes)
Subject: WARNING: Q700 clock upgrade to 33MHz
Keywords: test your machine very carefull!!!!
Date: 10 May 93 08:43:41 GMT
Organization: Technische Universitaet Muenchen, Germany
Hi all,
Yesterday I have encountered on the Q700 of a friend of mine some very strange problems. It looks as if the RAM on the motherboard (80ns DRAM) isn't able to do 33.33 MHz) Sometimes it fails and may damage your hard disk or what ever. This problem varies from board to board. Mine works perfectly under any tested condition with 33.33 MHz.
So here my warning:
TEST your upgraded Quadra 700 very carefully. Run it under full load a minimum of 24 hours with, for example, a POVRay picture, which uses most of the components in your Quadra 700 (CPU, FPU, SCSI, DRAM, VRAM ...). After that run a memory test for another 12-24 hour. If your Quadra doesn't show any problems you can be 99% sure that it runs at 33MHz. 1% is left over sorry, but no risk no fun.
To be 100% on the safe side make your room a little hotter than normal when you run the tests. This gives you more security and the 1% probability of problems is now smaller than 1%. Another tip, if you have not done a backup of your harddisk do it now before you upgrade. This will give you a better sleep with out nightmares. Good luck for all how have or think about upgrading to 33MHz
Rainer
I include an informative temperature & performance report from James MacPhail
Date: Thu, 22 Apr 93 21:53:49 -0700
From: James MacPhail (jmacphai@cue.bc.ca)
Subject: Quadra 700 osc upgrade temp results
I have done some Q700 CPU temperature measurements using an HP 34401A DMM with a Fluke 80TK temperature adapter. I placed the temperature probe on the case of the CPU where it is exposed beside the heat sink, as this is certainly not the place to measure the temperature most accurately, the actual conditions are probably a bit hotter than indicated.
Synopsis: The Q700 has a large design margin for CPU heat dissipation. Increasing the clock speed by 33% increases the CPU case temp about 4 degrees, indicating an increase in power dissipation of 25%.
Machine configuration: Base machine + four 1M SIMM=D5s, Quantum LP52 drive. (no additional cards or VRAM).
With the box closed in the upright orientation, the equilibrium CPU temperature was 37 degrees C at 25 MHz, and 41 degrees C at 33.3 MHz.
The temperature rose rapidly when the cover was removed: it was up 8 degrees in 5 minutes (and still rising). Room temp was 21 C (69 F).
The 68040 designer's handbook predicts Theta(JC) as 2.7 degrees C/Watt. Their calculations use figures of 3 and 5 watts dissipated (typ). Using 5 watts gives a junction temperature of 55 degrees C, so we have oodles of margin (max rating is 110 degrees C).
For those of you who also asked about performance numbers, I did more accurate comparison runs with Speedometer 3.1:
Synopsis: Except for SCSI I/O (the Disk test), 33.3 MHz machine is 33% faster than 25 MHz as expected (including on-board video).
Extensions off, 24 bit mode, 1 bit two page display, AppleTalk off, System 7.1, Quantum LP52, after quitting Finder (no other applications running):
| KWhet | Dhry | Towers | QSort | Bubble | Queens | Puzz. | |
|---|---|---|---|---|---|---|---|
| Q700 @ 33.3 | 205.479 | 27.247 | 24.960 | 22.391 | 23.823 | 24.105 | 28.804 |
| Q700 @ 25 | 155.078 | 20.390 | 18.909 | 17.166 | 18.000 | 18.320 | 21.721 |
| Perm. | FFT | FPMM | MM | Sieve | BMAve | PRGraf | |
| Q700 @ 33.3 | 27.170 | 172.661 | 154.452 | 30.250 | 25.602 | 63.912 | 28.347 |
| Q700 @ 25 | 19.892 | 127.619 | 115.839 | 22.891 | 19.071 | 47.908 | 21.489 |
| PRDisk | PRMath | PRCPU | PRAve | FPUMM | FPKWhet | FPUFFT | |
| Q700 @ 33.3 | 1.897 | 136.210 | 21.898 | 31.264 | 13.416 | 14.827 | 7.636 |
| Q700 @ 25 | 1.807 | 101.943 | 16.320 | 23.530 | 10.733 | 10.750 | 5.600 |
| FPUAve | Mono | 2Bit | 4bit | 8bit | ColorAve | ||
| Q700 @ 33.3 | 11.960 | 6.134 | 6.527 | 6.785 | 0.000 | 6.482 | |
| Q700 @ 25 | 9.027 | 4.615 | 4.878 | 5.036 | 0.000 | 4.843 |
32 bit mode (with 8 Mb) tests about 5% faster for video MacsBug, AppleTalk, and a bunch of inits hits video about 20%