My Apollo 1260

[wXR]

i must say how i do love the idea of these superfast cards... but this sounds dangerous to them

[Photon]

Hehe, just wait till someone makes an 060 card with SDRAM or Rambus RAM, or finds the weak points of common cards and puts a water cooling system on them I bet some Polish dude has already...

[Photon]

I made sure I had no contact problems with the expansion connector and tried 100 MHz rebooting 20+ times and only two insertions. No luck.

This leaves me with a clue for the two RAM sticks... since I can make it reach the bootscreen only if I remove both sticks, and the RAM sticks can handle an Apollo @ 140 MHz, the only reasons left are 1) not clean enough connection to RAM sockets, and 2) not enough power @ 100 MHz for both the CPU and RAM.

The 120W PicoPSU outputs 8A peak and 6A average @ 3.3V - or about 20W. That's not a lot of heat, but I have seen a small correlation between cooling and bootability in 80 MHz. Maybe it's just pure luck that the times I've not been able to boot @ 80MHz, turning up the fan instantly made it boot. 3 out of 3 is not much statistics though, and now I run @ 80 MHz with minimum fan and it boots fine. As usual it took a few boots the first time (at minimum fan) after closing the case and now it boots fine every time, even cold boots in the morning at minimum work fine last 2 days.

If we can find out power consumption for two 32MB EDO 60ns RAMs @ 50 MHz, we can add that to the 6A and get a total Ampere count on the 3.3V for those looking to do this 100 MHz mod and wanting to buy a proper PSU.

After this independant testing I agree that it makes sense to have decent cooling and clean the weak points (edge and RAM connectors), and above all: juice up the PSU!

Some sketchy calculations from a 1Mx16 EDO RAM spec sheet and power consumption increase from 50 to 66 MHz from the 68060 manual puts the power consumption for RAMs at 32x110mA=3.5 + 6.5-7.7A for the CPU = 10A @ 3.3V minimum on the PSU! To this is added a little power for the card itself. This is about what we get from the even more sketchy "68060 works @ 100MHz - without RAMs", ie. 6A+3.5A=9.5A.

(No diagram available in the manual, but projected 100 MHz power consumption from 2 preliminary datapoints is 6.5 falloff, 7.0 linear, 7.7 exponential.)

I glanced at the 160W PicoPSU specs which has 8A @ 3.3V but has a note that says it requires forced air flow when 3.3V @ 8A (which is insufficient anyway).

So it seems you can't have the Apollo 1260 stable AND silent at the same time, at least not if there's no other 10A ATX PSU that is silent.

The only other option I can see is at least trying with some really low-power EDO RAM... they would have to at absolute maximum consume 2.1A for both, for them to even be worth trying. If someone have some or know where to buy, let me know It's very unlikely it would work, unless 100 MHz power consumption is lower than projected. It's an easy test tho, the hard part is finding the exact RAMs.

This is indeed sufficient and apparently silent, as fan PSUs go. But it would be such a cumbersome and ugly solution... I've started to really like the looks of my rig now.

This "discovery" (basically repeating the conclusion by Roy Bates ) was an eye-opener, it's all very obvious now but it was unexpected to find what I think is the real power consumption of just the RAMs and the ol' 68060 which is so slim, innocent-looking and cool running - putting the card at at least 35-40W @ 100 MHz.

With the linked PSU and 50ns RAM and some space-age cooler, maybe one could go substantially above 100 MHz - until the Mach chips give up, probably

In Imaginary Land and with some super-mod involving space around the card to fit the big cooling-blocks, 168+ MHz could perhaps be achieved. I envision some alien Youtube video featuring Peltier elements, freak memory imported from Japan etc, filmed in a meat storage freezer room

[roy bates]

@photon

i think you have came to the same conclusion as me about power consumption at 100mhz and im glad you calculated in the power needed for the simms,plus i also think you are right about a pre-charge of the card at startup well done.this is the reason why im using a 400 watt supply.it sounds ridiculous but its true.what i tried to do was factor in the minimum load for all the parts times two.
this was so i wasnt running the supply at 100 percent load.
and yes its very noisy indeed.
also out of interest im afraid this mask wont go much further than 100mhz trust me ive tried.
oh before i forget the entire load for the card is 6.5 amps.cpu plus ram plus logic.

funny you mention peltier the problem with that. is it would require even more power to run and you have a chance of producing condensation.
water cooling hmmm that is something ive thought about

i thought for a while that it was a fluke at 100mhz but ive since done it with a second card with the exact same result.
and it also fails at 30 degrees celsius room temp oh well so much for a silent apollo with low power consumption.
and its the exact same result on my atari falcon at 100mhz with pc133 ram.350 watt supply and i have to have the case open.
ive found that all of the above are fine upto 90mhz.but when ran at 100mhz the power and cooling needs go through the roof.

but then again its no more noisy than my pc.

regards roy

[Photon]

And noise I like to avoid (in a case with built in keyboard at least, where I can't put the computer/PSU in another room)!

Stachu correctly pointed out that the RAM runs at 5V, which makes my above phantasm/rant a bit silly... eh, what can I say, was a long time since I learned the exact spec to hunt down...

To be honest I don't see how the Apollo @ 100 can boot with 0 RAM but not with RAM then. It should be as simple as "CPU draws ~5.6A @80 and ~7 @ 100, and I only have 6A in my PSU. Maybe it gets less hot when there's no memory access, but it's a stretch since it accesses chipmem.

I might check 5V and 3.3V with my oscilloscope to see if there's stress on the PSU at startup.

Even if a 160W PicoPSU could save the day, I'd probably still have to have cooling on the PSU. So this wonderful almost noiseless machine I have now seems to be incompatible with >80 MHz and perhaps I should settle for that.

[delshay]

Blizzard PPC 68060@92Mhz. screenshot http://www.amiga.org/gallery/index.php?n=3562

Blizzard PPC card can also boot @100Mhz,but this speed is not part of my project,so is not used just like both my screenshots in this thread.

[Photon]

Oh. Well, this is about mucking about with overclocking Apollo cards so you can't really transfer much knowledge from luck/experience with overclocking other cards. If you've made an Apollo go faster it would be nice to hear how you did it tho

Where did you get your 92 MHz oscillator? That would be nice to know Also how many Ampere on 3.3V you have on that PSU. Cooling used? Those kinds of things could be useful actually

[Amigafanboi]

Maybe he has a variable oscillator

[Photon]

btw Roy, here are 2 106.25 MHz oscillators for the price 135 SEK and 103 SEK, respectively, if that's interesting. SMD but should be np I think.

[roy bates]

Originally Posted by Photon

btw Roy, here are 2 106.25 MHz oscillators for the price 135 SEK and 103 SEK, respectively, if that's interesting. SMD but should be np I think.

sorry photon it wont work at 106mhz,the max this cpu works at is between 100 and 105,plus are those oscillators 3.3 volts and ttl i dont know if they will work or not.

also i wont boot any card without memory ,it seems pointless to me,it wont prove the card is working or not as the memory interface is not in use(please dont be offended by this coment)as i see lots of peaple say "it booted with no ram"so it works.
the cpu is not addressing any memory so its a false reading.this goes for all cards on any amiga.

i think the problem is with the 74FCT16543TPV or 74F257AD parts,due to indirvidual tolerances,
i think the card starts up with no ram because theres no load on the cpu.which makes me think its the above parts.

i think delshay got the blizzard to run at 92mhz with the help of 50ns simms which is just about within range.blizzards run the simms 1 on 1 timing.

you can have oscillators built at any frequency(between 20 and 120mhz) but its expensive(for what it is).i think a minimum order is 25 pieces.for any mhz at 5volt 50ppm ttl.

also can you tell me what revision motherboard your using.i only tested this with a 1.A , 1.B , 1.D.1 worked ever time on 1.A and 1.B

stachu knows his thing i respect his knowledge.

the thing is running these cards at 100mhz is un-known territory.so its going to be a bit of a hit or miss affair.also i got hold of 060's known to work at 95mhz to have more of a chance at overclocking it.where i got it from i only managed to get two.before he ran out of them.

ive got to ask, why did you add an extra wire for the 3.3volt line it wont make any difference.it will only draw what it needs.
and it cant draw more than the psu can output on that line.

also i have another theary that maybe it has something to do with how the card takes over the system at startup,when it turns off the 020.
not proven though.but there you go.

regards roy

[Photon]

Originally Posted by roy bates

sorry photon it wont work at 106mhz,the max this cpu works at is between 100 and 105,plus are those oscillators 3.3 volts and ttl i dont know if they will work or not.

Forget the one for 103 SEK, see Tekniska Data - runs on '2.5V'. The other one is 3.3V+-0.3V HC=High-speed CMOS. These come in 5V too, but Elfa.se doesn't have any above 80 MHz.

Originally Posted by roy bates

ive got to ask, why did you add an extra wire for the 3.3volt line it wont make any difference.it will only draw what it needs.
and it cant draw more than the psu can output on that line.

also i have another theary that maybe it has something to do with how the card takes over the system at startup,when it turns off the 020.
not proven though.but there you go.

It's just standard practice for me to solder thick wires when the Ampere draw isn't known. I wouldn't have had to connect a thick GND wire either, there are plenty of GND lines via the expansion connector. The extra wire was in case there was some 'deficient contact' between the pins through the through-holes and the 3.3V layer (such as pins too thin for the current draw or through-hole ring has bad contact with 3.3V layer internally). Redundant, but for the reason of redundancy.

Re: the theory - this is how I see it now: RAMs run at 5V so they could never run out of juice at 100 MHz. With a PSU with 6A on 3.3V the CPU was on the edge at 80 MHz and went over the edge at 100 MHz, simple as that I think. It did guru long before using fastmem many times.

When buses open on a CPU, it draws more current, probably much more, but with the humble speed of the chipmem it stays on the edge. At full speed though it goes over the edge. Something like that.

I'm pretty sure with a more capable PSU there would be enough current for the CPU to operate properly for longer than .5 seconds of memory access There could also be specific 68060s of the same mask as yours and mine that can't go to 100, and individual differences in chips on the card that can't go to 100. This CPU has worked fine at 80 for years, but has never been tested at higher speeds like yours were.

[roy bates]

@photon

i have just had a look at some of the parts on these cards,do you or anyone here have the specs for the 74F257AD ic.as the "ls" version only functions at 40 mhz?.

i think i may be onto something but i need to be sure.

these parts connect to the simms in 16mb blocks.i really need to know what frequency they are rated at.

i know the "propagation delay" is 4.3ns.i need to know its maximum operating frequency in mhz.

regards roy.

[Photon]

The full operating cycle is a chain of such propagation times. It's possible to know if you know the rest of the memory r/w circuitry. (For circuitry that follows an established standard, you don't need to know anything because the components in the circuitry meet the standard or else. This is not the case here, I think. Custom circuitry to use memory readily available at the time.)

Here's a datasheet and another brand, but in order to read the middle column of the timings of Waveform 1, one must 1) know what it does and 2) at what frequency it operates right now (in the circuitry it's used).

A guess for 1) is the multiplexer is for normal dynamic RAM row/column addressing, and a few ns later the _OE goes low and bus opens up to CPU. But you kinda sprung this on me so I don't know for sure.

For 2), measure the S pin with a multimeter set to 'frequency'. If with a 100 MHz oscillator it reads 50 MHz, it seems to me timing specs should be plenty good. But I don't know the 68060 memory access cycle timings. (=Too lazy to look them up...) If you read 100 MHz, that could be it. If the rest of the circuitry copes. 1000/10.5=95.23 MHz, so going higher makes it unreliable.

If you suspect it, I think you'd save time to simply get another 74257 with just a better Max value for Select to Output (if you can, F series are about the quickest) Just skimmed through it but it seems to operate at Vcc=5V with a 3V clock pulse.

[roy bates]

cheers that is what i was thinking,as ive heard peaple having trouble running these cards with a 50mhz cpu and 50mhz bus.(ive never seen it)but there you go.

thanks photon so 50mhz must in fact be in range(if its a philips part)which would explain why some cards are factory set to clock half on memory bus sometimes(with other eqevelant parts fitted)maybe, but i cant confirm it myself.as both mine are philips.

i.e: 1240@40mhz factory set to clock half to 20 mhz,when the memory can run at 40mhz easely.,i have seen this configuration many years ago.