Cosmos' PIO2 chips for A4000

[alenppc]

So I noticed the other day that Cosmos was selling these replacement GALs for the A4000 which are supposed to upgrade its IDE to PIO2. Has anyone else tried them yet? I installed them yesterday, but really the performace is exactly the same as with the original chips with the A3640. Plus my A4000 crashes during boot with a SanDisk 30x 4GB card, yet works fine with a US Modular (Samsung) one (2GB).

Anyway, the plus side is that the gal chips are 10ns as opposed to Commodore's 15 ns so they might work better with a 33 Mhz A3640 (the original ones do not work at that speed, at least for me).

[Turran]

Isn't CPU usage at 100% even without PIO2 when transferring large files? I doubt PIO2 will help then as its the CPU that is the bottleneck.

[alexh]

They are supposed to but as COSMOS explains, your CPU isn't powerful enough :

http://leblogdecosmos.blogspot.co.uk...00d-pio-2.html

[hooverphonique]

You should probably ask cosmos.. afaik, the screen shots on his blog for this mod are made running a 060 (I don't know on which type of cpu board, though)..

[SpeedGeek]

The A3640 is fast enough but you will only get a small improvement from PIO2 by itself. If you add the state machine mod you get much better results:

http://www.a1k.org/forum/showthread.php?t=37954

Cosmos also has an updated scsi.device which gives better performance with PIO2. I recommended a GAL speed of 15ns for U901 and 10ns for U902 with the IDEU901M update. But CF cards are always a case of your mileage will vary.

The PIO2 state machine assumes 25 MHz timing so overclocking to 33 MHz is another (serious) case of your mileage will vary.

[alenppc]

Thanks everyone for the feedback. I was surprised that this mod wasn't discussed in more detail here.

SpeedGeek, I am very interested in your A3640 mod as I wrote on Amibay, but unfortunately I lack the necessary skills to properly understand what I need to do. I should be able to program pals/gals and solder them but that's pretty much where my knowledge ends, I can't make heads nor tails out of the oscilloscope graphs.

I have attempted to overclock the A3640 to 30/33 mhz with the delay line hack, but the old IDE chips refused to work properly at that speed, kept giving me various loading errors, constant system crashes etc despite me cooling the 68040 well. So I put it back the way it was.

Today I have received a 40 Mhz 68040 so I am going to try again pushing it to 33 Mhz; let's see what happens.

[SpeedGeek]

@alenppc

I thought I made it clear in the A3640 state machine mod documentation overclocking is optional but not required...

[alenppc]

Ok, so the card will work fine as-is at 25 mhz with the updated logic.

What I have no clue how to do is to provide "skew compensation between the 2 clocks" in case of a 50 mhz CPU clock.

K, I will do some more testing.

[alenppc]

I received the new MC68040RC40A, K63H mask and fitted it to the A3640 to replace the old 25 MHz D50D mask cpu supplied by Commodore. It runs a lot cooler!

Unfortunately the Amiga would not boot at 33 Mhz even with the delay line hack, it is possible that new IDE chips might be responsible.

However, it runs stable at 30 MHz, with a nice 20-22% performance increase except that some zorro cards refuse to work at this speed. Not sure if the delay line hack might have something to do with this, or is it just a clock issue. I should try reversing it and running at 30 Mhz to compare.

[hooverphonique]

my a3640 runs at 33MHz without delay-line hack or any other mods (ide is working), but your mileage will vary depending on your motherboard and a3640 - there's no way to know other than trying :-)

[SpeedGeek]

AFAIK most A4000 owners could overclock to 33 MHz with an A3640 with original IDE PALs installed. But I don't know how many 33 MHz machines were successfully upgraded to PIO2?

You might want to check the speed of your Fast memory and be sure it's at least 70ns (or preferably 60ns) but you can always pull the Fast memory and see if works OK without it.

There should be no problems with Zorro cards since they operate from different or independent clocks from the mobo. I would look for other problems here... low voltage on the +5V supply or bad capacitors on the A3640 or mobo, etc.

[alenppc]

Progress update:

I have realized that even 30 Mhz is unstable after a little while. The reason? Commodore GALs get incredibily hot after a few minutes (too hot to touch for more than a few seconds), which is funny cause the 040 stays really cool. I think I need faster GALs. This should be solved once I apply SpeedGeek's 030 machine state mod. I have obtained all the necessary 10ns, 7ns parts. I am just about to get some 5ns ones for experimentation as well. Also I am waiting for a programmer socket adapter to arrive from China. If it ends up taking too long, I will build my own.

SpeedGeek, I have reread the entire documentation that you wrote once again, and some things are a bit clearer to me. Applying the machine state mod should be fairly straightforward.

I will get a 74F74 chip also, and try to figure out how to use it for a 25 mhz clock to feed to the motherboard, not sure exactly where to start on that one but let's first do the state machine mod and go from there.

My entire A4000/A3640 and daughterboard have fresh brand new capacitors. I have even fixed the reversed cap polarity issue on the 3640. I am also using 70ns ram as I can't find 2/4 mb modules in 60 ns flavour.

[SpeedGeek]

GALs should not overheat at 30 MHz regardless of speed ratings. You may have some TTL PALs on your A3640.

Apparently, you did not look at the A4000 schematics and notice the 74F74 on the mobo is already set up to make a 25 MHz CLK?

[alenppc]

Yes, there might be some TTL PALs, I haven't peeled off the labels yet. But I will find out soon enough since this weekend I will be adding sockets.

I know that there is a 74F74 on the mobo but I have no clue how to connect it to the A3640, nor what the implications are. So I got a spare chip just in case it turns out to be a simpler solution using it instead...

[hooverphonique]

Originally Posted by alenppc

I am also using 70ns ram as I can't find 2/4 mb modules in 60 ns flavour.

If you short (the currently floating) dram address line A10 to gnd or 5V, you can use 16mb simms (acting as 4mb), which are easier to find in 60ns.. that's what I use on mine and that works nicely at 33MHz mobo clock..

[SpeedGeek]

alexh once asked me about writing an "A3640 Upgrade Guide" and I told him I didn't think a guide which works for everyone was a practical idea. I guess this thread speaks for itself...

[alenppc]

@SpeedGeek:

My goal is to figure out how to do this for the A4000 and then write an A4000 specific guide, since there seems to be a lot of difference between A3000 and A4000.

Update: commodore used a mix of 15ns PALs and 10ns GALs on my A3640.

[alenppc]

I have ended up removing the Cosmos chips and putting back the original Commodore ones. I was having random data corruption issues while using his chips, sort of like what used to happen with the early versions of the FastATA from Elbox.

As for SpeedGeek's 030 state machine mod, I am having problems with it being stable at 25 Mhz, although that might be cause I am using 10ns 16V8 gals (basically I just reprogrammed the commodore ones). Adding some cooling to the GALs makes the machine last longer before crashing but only just.

The memory access speed has indeed increased by about 17% and the overall system performance has increased as well. I am hoping to be able to find 7ns GAL chips compatible with my programmer and give it a try again. If that doesn't work, I will reprogram them with the original Commodore logic.

[SpeedGeek]

@allenppc

I doubt the 10 ns 16V8s would cause any problems @ 25 MHz. (There were a couple of guys on A1k.org which used them without problems). Check the + 5V supply, Rev 3.0 A3640s have a STERM sampling bug, check solder connections and socket mounted chips, etc.

[alenppc]

Originally Posted by SpeedGeek

@allenppc

I doubt the 10 ns 16V8s would cause any problems @ 25 MHz. (There were a couple of guys on A1k.org which used them without problems). Check the + 5V supply, Rev 3.0 A3640s have a STERM sampling bug, check solder connections and socket mounted chips, etc.

Okaaay... After looking into the cut & jumper mod described here: http://wonkity.com/~wblock/a4000hard/a3640ref.html
that is supposed to deal with the STERM sampling bug I followed the instructions and done it here on U200 as described. This appears to have stabilized the system a bit. I am still experiencing sporadic crashes, although I am starting to suspect the SIMM modules. Anyway, it appears that I am on the right track.

Here is the pic: