Thread
Wanted: Advice on performing 68040 upgrade for Q605
I'm finally getting around to swapping the 68LC040 in my Q605 for a full 68040. I have a 68040 donor machine ready to go, put I have no experience pulling chips. Any advice on where I can get the right kind of puller, or on how to do without one? Keep in mind I've never done this before, and I'd be very unhappy if I damaged anything 
A broad (1"+) wood chisel can be used to pry the processor out of the socket, working slowly one side at a time. You don't hit it, you turn it like you might a screwdriver, and you keep it out at the edge just under the rim.
I have done several like this with no CPU fatalities. Presumably a knife could be used as well, but a chisel gives good control. (Mind you, I am a woodcarver, so am used to the actions involved.)
After the processor begins to come up, you can basically pull the rest of it up with your fingers. The pins are maybe 1/4" long, so you don't have much to pull.
I have done several like this with no CPU fatalities. Presumably a knife could be used as well, but a chisel gives good control. (Mind you, I am a woodcarver, so am used to the actions involved.)
After the processor begins to come up, you can basically pull the rest of it up with your fingers. The pins are maybe 1/4" long, so you don't have much to pull.
Unless you have a proper IC puller, or some other device that can get under and lift both sides at the same time, then the trick is to ease each side up a very small amount at a time, and then go to the other side. That should help prevent you from bending the pins.
I've heard the point of the plastic shirt pocket clip on a Bic biro lid can be good for this. Being plastic rather than metal you run less chance of destroying pins if your hand slips. Otherwise maybe a painter's putty knife?
If you do end up with any bent pins, a retractable pencil is often suggested for straightening them. Remove the lead and use the pencil's lead gripper in the tip to surround the pin and gently bend it back to vertical.
I've heard the point of the plastic shirt pocket clip on a Bic biro lid can be good for this. Being plastic rather than metal you run less chance of destroying pins if your hand slips. Otherwise maybe a painter's putty knife?
If you do end up with any bent pins, a retractable pencil is often suggested for straightening them. Remove the lead and use the pencil's lead gripper in the tip to surround the pin and gently bend it back to vertical.
I usually use a thin screwdriver. I know, not the recommended way of doing it, but as long as you are extremely gentle, you should be fine. I've been doing this for years, and have never had any CPU or board fatalities.
Just lever the the cpu with a thin knife, SLOWLY. I would suggest no more than 1mm or 1/32" before moving on to next side and repeating, the pins are only 3mm / 1/8". Ok you cant remove it flat but incrementally on a small angle is just fine.
The worst thing I've done with this method is hit the circuit board with the knife and injured one trace. Fixing that with a circuit pen was far harder than straightening a pin. The cpu pins are robust and handle a little straightening.
With insertion make sure they are straight and aligned before putting on any pressure. The cpu should sit down a little when ready to be placed in the socket.
After that you could swap resisitors for a better speed. http://homepage.mac.com/schrier/q605.html The 33 works for me, the 40 never did, even with a resistor from another quadra.
The worst thing I've done with this method is hit the circuit board with the knife and injured one trace. Fixing that with a circuit pen was far harder than straightening a pin. The cpu pins are robust and handle a little straightening.
With insertion make sure they are straight and aligned before putting on any pressure. The cpu should sit down a little when ready to be placed in the socket.
After that you could swap resisitors for a better speed. http://homepage.mac.com/schrier/q605.html The 33 works for me, the 40 never did, even with a resistor from another quadra.
Most sockets have little plastic tabs which are meant for a metal clip which holds the heat sink on. Take a broad flat headed screwdriver and use the plastic tabs to pry against and edge each corner up a little at a time. The whole CPU will eventually start coming up after half a dozen rotations or so.I'm finally getting around to swapping the 68LC040 in my Q605 for a full 68040. I have a 68040 donor machine ready to go, put I have no experience pulling chips. Any advice on where I can get the right kind of puller, or on how to do without one? Keep in mind I've never done this before, and I'd be very unhappy if I damaged anything
Are you going to up the speed to 33 MHz?
http://homepage.mac.com/schrier/q605.html
You can go up to 40 MHz if you can find a 20 MHz oscillator.
nb: 40MHz is not always stable though. A heatsink/fan for the 040 would be a good idea.
http://homepage.mac.com/schrier/q605_40.html
Output Enablers sell some useful parts.
http://homepage.mac.com/schrier/q605_40.html
Output Enablers sell some useful parts.
That would make no difference whatsoever.
We'll see how the CPU swap goes before I decide on overclocking. I actually have a Q605 and an LC475 as well as two 68040 donor machines, so I have some room for error. My end goal is to have a Q605 with a full 68040, possibly overclocked, 128Mb RAM, and a new HD (whenever I can scrounge up an IDE-SCSI or (preferably) SATA-SCSI bridge). This machine will host my Apple IIe card, and dual boot 7.5.5 and 8.1.
Don't try to pry one side out completely before the others. You'll bend some pins. You need to work around the CPU and try to pry all sides up a little at a time until it comes loose. Your risk of bending pins is greatly reduced. I have a full 040 in my Performa 631CD and it makes a difference. I wish I was brave enough to change the clock resistor on the motherboard to get it up to 40mhz.
I can confirm it - at 40 MHz heatsink is a must and you should use at least 33MHz processor if overclocking that baby!
My Q605 had processor socket with notches which permitted fitting a heatsink from Q650/700 variety. I also installed real 40 MHz 68040 and changed Motorola PLL chip. After these mods machine have been stable for something like 8 years or so.
ojfd
If you can't find a socket with clips, you can get something like Arctic Silver thermal adhesive:
http://www.newegg.com/Product/Product.aspx?Item=N82E16835100005&cm_re=arctic_silver_thermal_adhesive-_-35-100-005-_-Product
(not specifically endorsing Newegg, but they are a pretty no-fuss company)
I've used this to adhere the heat sinks on a number of QuadDoubler-type CPU cards. You can see a little bit between the CPU and the heat sink in this picture:
Just FYI - almost all m68040s in Quadras were made with the exact same mask and feature size, so there's no manufacturing difference between most 25 MHz m68040s and 40 MHz m68040s besides the fact that the 40 MHz parts were tested at 40 MHz in the factory. I've overclocked many 25 MHz CPUs to 40 MHz without problems, and I wouldn't be surprised if they worked fine at 50 MHz in QuadDoubler-type situations.
What'd be nice would be to find a source of Freescale post-2002 m68040s which are made with a smaller feature size, take much less power, and can probably run much faster. The latest (and perhaps last) mask from Freescale is L88M at .57 microns.
I imagine one in a QuadDoubler-type card on a 33 MHz motherboard would be sweet. It might even be possible to run such a CPU at 80 MHz... One of these days I'll find one!
http://www.newegg.com/Product/Product.aspx?Item=N82E16835100005&cm_re=arctic_silver_thermal_adhesive-_-35-100-005-_-Product
(not specifically endorsing Newegg, but they are a pretty no-fuss company)
I've used this to adhere the heat sinks on a number of QuadDoubler-type CPU cards. You can see a little bit between the CPU and the heat sink in this picture:
Just FYI - almost all m68040s in Quadras were made with the exact same mask and feature size, so there's no manufacturing difference between most 25 MHz m68040s and 40 MHz m68040s besides the fact that the 40 MHz parts were tested at 40 MHz in the factory. I've overclocked many 25 MHz CPUs to 40 MHz without problems, and I wouldn't be surprised if they worked fine at 50 MHz in QuadDoubler-type situations.
What'd be nice would be to find a source of Freescale post-2002 m68040s which are made with a smaller feature size, take much less power, and can probably run much faster. The latest (and perhaps last) mask from Freescale is L88M at .57 microns.
I imagine one in a QuadDoubler-type card on a 33 MHz motherboard would be sweet. It might even be possible to run such a CPU at 80 MHz... One of these days I'll find one!
Did you do the 40 MHz modification on a Q605? If so, did you use one of the two methods listed on Marc Schrier's page with wires and such, or did you just do the resistor swap for 40 MHz?
Hi trag,
Yes, I've modified Q605, not LC475, althoug I've put a jumper on the MB after I went to 40 MHz, which changed its gestalt ID to LC475.
At first I swapped resistors to go to 33MHz and afterwards I installed 40MHz oscillator, as per instructions on Marc Schrier's page.
http://homepage.mac.com/schrier/q605_40.html
I also changed MC88920 PLL to MC88916DW70 which I removed from that useless IIvx MB, installed "real' 40MHz MC68040 and put a heatsink on it. Heatsink came from Q700 that had damaged video output.
Cheers,
ojfd
Yes, I've modified Q605, not LC475, althoug I've put a jumper on the MB after I went to 40 MHz, which changed its gestalt ID to LC475.
At first I swapped resistors to go to 33MHz and afterwards I installed 40MHz oscillator, as per instructions on Marc Schrier's page.
http://homepage.mac.com/schrier/q605_40.html
I also changed MC88920 PLL to MC88916DW70 which I removed from that useless IIvx MB, installed "real' 40MHz MC68040 and put a heatsink on it. Heatsink came from Q700 that had damaged video output.
Cheers,
ojfd
Would you still need a heat sink/fan if you were using an 040 that was actually rated for 40mhz? I have two of them installed in 33mhz machines that I want to kick up.
Thanks. I did the MC88920/88916 swap, and pull pin 9 high (?) or something like that.
http://homepage.mac.com/schrier/q605_40.html
on one machine but could never get video. But I think my VRAM is too slow. There's a note about sometimes needing 80ns VRAM after the 40 MHz upgrade. I've set it aside for several months though (hmmm, I guess it's been almost two years actually. Sigh).
But I've always wondered why the straight resistor swap won't do the trick. Schrier's page says it doesn't work, but it also doesn't say that anyone ever tried it with the clock generator/buffer swap. It seems to bear more experimentation. One of these days...
Hmmm. I just looked over the 88916DW70/80 datasheet and reread the speedup instructions. I wonder if I pulled pin 9 of U17 high, or if I pulled pin 9 of the DW88916 high? I can't remember now, but I think I may have gotten confused and done the latter....
Yes. The 40 MHz part is made exactly the same as the 25 MHz part, so you'd definitely need a heatsink at minimum. On the other hand, if you can get your hands on a Freescale made m68040, they will run almost cool at 40 MHz without a heatsink.
I'd be interested to hear the results.
There's another simpler way to get to 40 MHz which used to be available from here:
http://gabezing.sytes.net/LC475-40Mhz-E.html
However, that site's been unavailable for a while and archive.org doesn't have copies. Does anyone have a copy or know of a mirror? We need to put those up.
The process is simpler than sourcing a MC88916, and simpler installing, too. Just get a 20 MHz oscillator, take the power for it from the 31.3344 MHz oscillator on the corner of the motherboard by the video connector, remove R93 from the backside (also close to the video connector), and run the 20 MHz oscillator's output to the inside pad where R93 used to be (by inside I mean towards the center).
I've done this mod on a number of motherboards, and it's never not worked at 40 MHz.
Here's one of the machines to which I did this modification something around ten years ago:
http://www.mac68k.org/~john/boobookitty/
Yeah, like that's gonna happen. They are like $250-$300 each in bulk according to Freescales' site. 8-o
You never know - companies are still buying them, so it's just a matter of time before some older product becomes "obsolete" and the chips work their way onto eBay. Or if a company goes titsup, their stock could get liquidated. I'm hoping...
Hi guys,
Although we often mention Marc Schrier's page, we all should give a credit regarding this mod to that person from Japan, namely
Hiromasa Yamaoka
who was first to do such mod.
In all Quadras + IIVx this PLL works as Frequency doubler. That means, you put 12.5MHz in and out comes 25MHz. You put 20MHz in -> out comes 40MHz.
Now, if You read Motorola's MC88920 and MC88916 data sheets very carefully, you will see that:
1. MC88920 has been designed specifically for the 20 and 25Mhz 68040 microprocessors.
2. MC88916DW70 will work at max 35 MHz processor frequency and has been designed for faster than 25MHz processors.
3. MC88916DW80 will work at max 40 MHz processor frequency and also has been designed for faster than 25MHz processors.
If the manufacturer says what we should use, it would be good engineering practice to follow his recommendations. In our case we all should be using DW80 version, not even DW70!
Running 35 MHz part (DW70) @ 40 MHz is already 5 MHz too much, but running 25 MHz part at that frequency is living on the edge.
Regarding heatsinks on "real" 68040/40 - just look how its done in Q840 - full 68040@40MHz, regular pin-grid heatsink. DW80 PLL chip. And it keeps going.
ojfd
Although we often mention Marc Schrier's page, we all should give a credit regarding this mod to that person from Japan, namely
Hiromasa Yamaoka
who was first to do such mod.
There could be numerous reasons. Gazelle clock generator chip (Apple p/n 343S1135 -a) is not documented anywhere (anyone care to ask Apple's engineering department ;-) ), hence nobody really knows what's going on inside that chip. It might be that it can't generate "clean" 40MHz due to its internal limits. It might be that long traces + extra capacitance at 40MHz distorts signal waveform. Nobody knows... I have absolutely no motivation to poke around with my Tek TDS724 scope inside that poor Q605 trying to improve it or demistifying Gazelle chip's inhards . If it works with extra oscillator, why bother? Period.
"johnklos", _You can not make it work without the help of PLL chip in any way_! (PLL stands for Phase Locked Loop, check Google or Wiki how it works, in caseYou don't know.)
In all Quadras + IIVx this PLL works as Frequency doubler. That means, you put 12.5MHz in and out comes 25MHz. You put 20MHz in -> out comes 40MHz.
Now, if You read Motorola's MC88920 and MC88916 data sheets very carefully, you will see that:
1. MC88920 has been designed specifically for the 20 and 25Mhz 68040 microprocessors.
2. MC88916DW70 will work at max 35 MHz processor frequency and has been designed for faster than 25MHz processors.
3. MC88916DW80 will work at max 40 MHz processor frequency and also has been designed for faster than 25MHz processors.
If the manufacturer says what we should use, it would be good engineering practice to follow his recommendations. In our case we all should be using DW80 version, not even DW70!
Running 35 MHz part (DW70) @ 40 MHz is already 5 MHz too much, but running 25 MHz part at that frequency is living on the edge.
Regarding heatsinks on "real" 68040/40 - just look how its done in Q840 - full 68040@40MHz, regular pin-grid heatsink. DW80 PLL chip. And it keeps going.
ojfd
I'm sorry, but how can you tell me that I can not make it work when I've been running at least one of my Quadra 605s at 40 MHz this way for a decade? That doesn't make any sense.
If that were true, nobody would overclock since overclocking is almost always running parts faster than what the manufacturer says.
We're talking about a simple PLL clock driver, not a multiple-million-transistor device like a CPU. Also, the implementations and specifications are identical for both with the only exceptions being a recommended change in the anti-jitter resistor for applications below 40 MHz (20 MHz CPU) and the guaranteed minimum working frequency:
http://www.mac68k.org/files/MC88916.pdf
http://www.mac68k.org/files/MC88920.pdf
The added oscillator's output simply goes to pin 8 of the MC88920 which is disconnected because of the removal of R93. How is that not working with a PLL when we're precisely working with the MC88920?
OK, if you're using 20 MHz oscillator in your mod, would you mind telling me where that 40 MHz signal for the processor comes from?
:-D
ojfd
Well, apparently each MC88920 on every Quadra 605 motherboard I've tried will take 20 MHz in on the SYNC line and produce something close to 40 MHz. Granted, the speeds from each differ a little - Clockometer shows 38.something on a couple, the "cpu: delay factor 1593" line is a little different on each (1593 is from a 50 MHz m68040; a true 40 MHz is around 1280, and some of the 605s are in the low 1200s and high 1100s which would put the actual clock between 36 and 39 MHz).
The reason I brought this up is that most people are not capable of desoldering and resoldering surface mount chips, so if you can get most of the speed with a lot less risk, it's worth sharing.
On the other hand, I just bought a lot of seven MC88916DW70 on eBay, so if anyone would like one or two, email me - I'll sell them at cost ($4.50 USD plus shipping).
Yes, I noticed that when I reviewed the datasheets last week. However, folks seem to run at 20 - 25 MHz even after they've swapped in an 88916. There might be problems if the 88916 gets hot or otherwise loses margin, I imagine.
The 88916s I have on hand are the '80, but I got mine as samples from Freescale over ten years ago. I wish I had somehow gotten more now, as I only have two. Freescale sent the part to IDT who does not appear to offer it any more, or if they do, they sure don't make it easy to figure out on their web page. Maybe it's a Flash thing.
The easiest place to get the 88916 now, is probably from Marc Schrier. He still lists them for $20 on his Output Enablers website. Assuming that info is still valid, that may be the only remaining source short of ordering 50 of them them Arrow. The problem is that Marc lists the DW70, not the DW80, so while the DW80 would be better, teh DW70 is probably all that is realistically available.
I guess I should go add MC88916* to my list of Ebay searches...
I don't quite follow you.. are you saying that people swapped 88920 for 88916 and are wondering why their computers aren't working at 40 MHz or are you saying that they swapped 88920 for 88916, changed resistors to go to 33 MHZ, added 20 MHz oscillator to go to 40 MHZ and then backed back to 20 - 25 MHZ because their Q605's were unstable?
What? Isn't that enough? How many Quadras are you going to overclock?
Cheers,
ojfd
There are two clocking issues here; one is running everything on the motherboard at a different speed (33 or 40 MHz), which is why some people have problems with the speed of the video memory, and which is also why people don't just swap the resistors to give them 40 MHz. The memory on all the Q605s I've seen have always been 70ns, whereas 60ns SIMMs are pretty common. My guess is that if the motherboard memory were faster, the 40 MHz resistor change would work.
The second clocking issue is bypassing the motherboard speed settings and sending a separate clock signal to the MC88920 (what I mentioned before). The page which discussed this also recommended doing the resistor swap to 33 MHz, probably so that the memory will run a bit faster. In this case, an unhappy MC88920 might give a different clock rate (somewhere between 36 and 40 MHz), but I can't imagine even the most unhappy MC88920 giving 25 MHz.
So the question is whether anyone who has swapped their 88920 for an 88916 has done a proper resistor swap, or inserted a 20 MHz oscillator signal, or both, and if they have, that'd be VERY strange... Where did you hear about these cases?
The datasheet says that the 88920 is for slower 20 - 25 MHz operating frequencies. And that the 88916 isn't suited for the slower speeds. However, when doing a careful modification, the first step is to remove the 88920 and install an 88916. Then retest to make sure the board still works. At that point, one is operating at 25 MHz and it works fine, even though it's not recommended with the 88916.
That is what I mean.
I have two Q605s. When I'm done, I'll have no 88916s in my parts supply.What? Isn't that enough? How many Quadras are you going to overclock?
Although I am happy to see John's offer. If only I can remember it when I eventually get back to this project. I've implemented a new rule for myself, where I buy nothing until the moment I'm ready to use it. No more, "I'll buy this. It will be useful in that project three years from now and might be hard to find then." That's how I ended up with a living room (soon to be attic) full of electronic crap....
Just curious - where in the datasheet does it say that they're not suitable for slower speeds?
Ha ha ha... If only we all followed those rules... Then again, I don't know if my Diesel Chevette would be finished and running now if I didn't hoard parts over the years... and I finally got a latest-mask version of the MC68060 (71E41J) which is a .32 micron device (compared with the original, which is a .6 micron device), and if I hadn't kept around my CyberStorm Mark III until I found someone who could repair it, I wouldn't have a working 80 MHz m68060 system right now... So it all depends.
When I lived in NYC, I always had to prefer smaller machines to larger ones, and I suppose I've tried to keep that philosophy. But if only we didn't hoard everything, there'd be one less stereotype about us, I suppose.
These chips are small, so I'll keep them for when you need a few, trag.
Well, they did eventually make a 40Mhz 68LC040, which is now at half that price, and as a bonus these has the FPU emulation bugs fixed.
Bargain! [
)] ]'>