A sneak peek at the ACA1240/1260 cooling system

Quote from NicoPPC

Would it have yuv to rgb conversion for video ?

Is colour space conversion anything that the Amiga would really benefit from? If so, what software and what API?

Jens

Quote from OleK

Any updates on progress on the ACA

1240/60? I have money all saved up and ready to go 🤩

We're getting closer - it's still a LOT of steps until we can launch the (paid) beta phase. The mere fact that the hardware is already in production might paint a wrong picture of "available soon", though.

Quote from OleK

Also, any idea on the PSU-requirements?

Yes: Something that powers the Amiga

Part of the complicated design is that the requirements especially on the 5V rail (which is not properly regulated on 99% of all PSUs out there, especially new ones) are very low. All main power (3.3V logic/RAM voltage, 2.5V PLL voltage, 1.2V FPGA core voltage and 3.3V/5V CPU core voltage are all generated from the 12V rail, which has plenty of reserve power on pretty much every PSU. Usually, Amiga power supplies can deliver 1A on 12V, and the A1200 itself only needs maybe 100mA of that. That leaves well in excess of 11 Watts of power available to the ACA1240/1260, even on the worst power supplies out there, with very low requirements on regulation, as all regulation is happening on the ACA1240/1260 locally. The one thing I wanted to avoid is to have support cases because of known-bad PSUs that are sold as "new and great" on Amazon or other shops. I've been trying to educate the Amiga community about power supply problems far too long, largely unsuccessful.

Jens

I'll verify that separately, but expect it to be fine with the stock PSU, yes.

Yes, the plan is still to have a few select beta testers, and I already wrote (I believe in the German part of this forum) that I'd prefer to have testers close, like max. 200km from my place in Germany, which includes parts of Belgium and the Netherlands. This is mainly to limit shipping cost in case the boards need to come back for a re-work.

Right now, we're working on a peculiar problem: The board is not stable when it's cold, and this will only show on a few computers. Naturally, it only shows on Timm's and Peter's computer, but not on any of my A1200s. When the computer(s) have warmed up, everything is super-stable and can run for days, even at elevated temperatures.

Peter and I had a hunch earlier this week and he's working on a solution - might be a few more days until we have resolved that. It's the last showstopper before Timm can start working on his part of the software.

Jens

Nothing worth reporting. Well, the "doesn't start when host machine is cold"-problem is resolved if you wanted to know that.

Jens

This week's three-day meeting with Peter revealed that the 68040-40 CPU violates it's datasheet in terms of "when does it take over data". As a result, we've decided to re-factor the one thing that's at the core of the design: The SD-Ram controller. Tweaking the card for top performance is currently too tedious, and going beyond 40MHz for the 68040 might stretch that "violating" timing even more, requiring modifications on SD-Ram timing that are currently too difficult to make.

Don't worry, this is not "back to the drawing board" - but it's definitely a delay in releasing the cards to beta testers. May still happen this year, but it may also be January 2024.

Jens

These past two days, we've exclusively been working on the ACA1240/1260. First goal was to overclock the 68040, because I've read that Apple guys have cranked the 68040 up to 50MHz, but that didn't work, even on a carefully-selected 40MHz unit with 0.65µm mask. Also tried a 0.5µ mask (V-version), but that performed even worse. We gave up when even 42MHz wasn't stable.

It got more interesting with an MC68060-Rev.6. We've worked up in 10MHz-steps and got 100MHz to work reliably. However, when stepping up to 105MHz, things don't work any more: As soon as we execute setpatch, the system crashes. I assume that the MMU just can't run that fast. It is definitely not a memory problem, as we have memory working up to 120MHz with a 40MHz 68040 (triple memory clock for shortest-possible first-access penalty).

Here's a few screenshots from benchmark tools and an AIBB module. Caution: This isn't optimized out yet! It's still possible that we can eliminate another wait state, getting even higher memory performance. However, even with the current performance, we are at the top end of what's been available for the Amiga so far.

Jens

There is no "newest" Sysinfo. The source has been released to people who have re,branded it, but to my knowledge, didn't really make significant changes. There are at least two forks now, and it's not entirely documented where changes are.

AIBB is of course the patched version that doesn't require the 68000-setting in tooltypes.

Chipram access speed maxes out the A1200 bus. One thing this test does not show is the available bus time on the 68040/68060 side. More on that when the full specs are released.

Jens