A sneak peek at the ACA1240/1260 cooling system

The ACA1240/1260 is taking shape: Our internal prototypes have been equipped with all custom-made parts of the cooling system for the first time this week.

The heat sink will replace the trapdoor cover of the A1200. On the final version, a rubber foot will make sure that the fan will always take in enough air that is blown against the fins of the heat sink.

The fan has been custom-made for us, and it's state-of-the-art: Speed can be controlled, and the current speed can be sensed (PWM and tach signals). This ensures that the required amount of air is available without generating unnecessarily high noise levels.

A high-precision temperature sensor is mounted in the heat sink. Especially the 68040, which does not have an integrated temperature sensor like the 68060 does, profits from well-regulated cooling: In power saving mode with 4.8V core voltage and 20MHz speed, the fan can always run almost without noise at lowest speed. Full overclocking with a core voltage of up to 5.5V is only allowed while the heat sink temperature stays within the specified range.

Quote from DanyPPC

Does it mount always 128MB ram ?

No, it's 192MByte fastmem (non-shared, so full speed ahead for the CPU). The GFX expansion gets it's own memory.

Quote from DanyPPC

I hope this board doesn't have slowdown to the memory chip access like the Blizzard 1260.

The bus speed of the 68040 and 68060 CPUs can't be changed - we do have some measures to get the most out of these CPUs, but there's limits that the CPUs have internally. We do run at full bus speed (not half bus speed like the B1260 does).

Quote from temisu

1. Since the fan is separate from the heatsink, I hope that these things can then be ordered separately as well and user replaceable? (since they are consumables essentially)

Yes, we'll have spares.

Quote from temisu

2. Will it support the IDE-speeder as the new ACA1233n?

Yes, the IDE-speeder will even have finer granularity of timings. There's more surprises that we'll have on the next prototype

Quote from KKR75

will the ACA1240/1260 support the ACA500+ on an A500 ?

Yes, it will.

No details yet - the prototype is "only CPU and RAM" for now. Next prototype will have a lot more peripherals.

Quote from Akiko

Any plans for a CD32 or A3000/A4000 version?

Those are entirely different computers. The CD32 lacks the CIA chips and mass-storage, so a concept that adds CIA chips and an A1200 acceelrator connector would be what I'd aim at. I even started a design a few years back, but shelved it due to lack of time.

The A3000/A4000 have a totally different timing, as the bus is 68030-like, not 68020-like. Further, DMA from Z3 cards must be supported, so the complexity is a good deal higher than an A1200 card. While I believe that it's possible to make such a card, it certainly depends on the financial success of the A1200 version.

Quote from DaveP

Any idea on what this will roughly cost - specifically the 1260 version.

That will largely depend on the CPU mask - with the extremely low quantity of available Rev.6 CPUs on the market, these will probably be in the "may afford next month"-domain, while I still have the prime goal to offer the "smallest" 68040 version for under 500,- EUR.

The 68040 is much easier to get on the market, and therefor we have started development with the 68040. My goal is not only to make the "small" version affordable, but also to make the faster versions (33 and 40MHz) not only viable, but desirable alternatives to the much-more-expensive 68060: The old Apollo and Blizzard 060 accelerators clocked at 50MHz gave about 37000 Dhrystones in Sysinfo, which is of course awesome: It's about 3 times faster than the fastest 68030 that we can offer.

However, even with our first prototype, which does not have all the optimizations that the next prototype will have, we get well over 30k Dhrystones for a 40MHz 68040, and memory access performance that has never been shown on a 68040 before:

Note that the Chipram performance is something that we intend to improve on the next prototype - I hope that read performance can be improved on that. Write performance is already at the top of what can be done - the chipset just does not have any more time slots that the CPU can hit.

The attached AIBB module file is just a development snapshot of what we have achieved on the current prototype. This is not the final performance, but already a nice demonstration of what fast memory can do, possibly even a demonstration of the advantages of SD-Ram over DDR memory.

Quote from DaveP

When is it likely to be available?

We're pushing hard - the next prototype board is already in production and that has all the hardware features that I want to see for the release version. Verifying all the peripherals will take a few weeks, so we will definitely miss this Xmas. If prototype verification runs as smooth as it did for this current prototype, we're definitely talking Q1/2020 for the mass-produced units.

Quote from W3DRK

will there be a possibility for the customer to supply their own CPU?

The ACA1240/1260 comes with a CPU socket as standard. However, there are two hurdles exchanging the CPU. First of all, there's the mechanical part, as removing the CPU requries a PGA puller tool. Prying it out with a screw driver is possible, but will damage the socket and maybe even the CPU, as it sets very tight in the socket. Re-inserting the new CPU is also not the easiest part, and I fear that the amount of force required to push it in will result in bent pins if they are not perfectly aligned with the socket's pins.

Further, if you buy the 68040 version, it will of course come with the 68040 memory controller, which does not work with the 68060. The license for the 68060 memory controller must be purchased separately, so "on a budget" won't be just for the price of the CPU.

If you decide to make changes to the card, you must keep in the back of your head that legally, you become the vendor of the whole thing. Warranty and support ends if you make such changes, as we can't possibly provide assistance in getting a possibly-defective (or even fake) CPU to work. We may introduce a "DIY-corner" with voluntary assistance in this forum, as I don't want to leave you completely alone. In that separate area, it'll be very clear that we assume no legal obligation to help out making the product do what it was not sold as.

Over the past seven years, I have bought many 68040 and 060 CPUs, and I have shipped back way more because they were either defective or fake. The worst fakes were 486 CPUs with 68060 marking printed on them, other bad fakes were 68040s with Freescale-68060-Rev6 print on them, and the better fakes were Rev.5 CPUs re-labelled to Rev.6 - these were hard to identify, as even the mask codes and date codes were very plausible, and there were virtually no scratches on the ceramic package. At first glance, they looked genuine to me, and probably to many others as well. I fear that many of these are sitting on Amiga user's shelves, waiting to be debunked as fakes. I know that some of them already ended up in CT060 cards (the Atari ST 68060 project).

All this fear and warnings aside, it'll be possible, and probably the best chance of serving the demand for 68060 cards with the extreme shortage of CPUs that we have in the market.

Quote from W3DRK

Also, I should add I'd happily whip out the credit card for a modern 3000/4000 040/060 accelerator in an instant.

The ACA1240/1260 is the single most expensive project in the history of iComp, even beating the C-One project by an order of magnitude. We'll have to see how it goes financially before I can commit to making an A3000/A4000 card, which will be even more complex due to Z3 slots requiring a DMA path into the accelerator's memory. I am not excluding this possibility, but a decision needs to be based on the financials of the ACA1240/1260.

Component-wise, the key thing is the 200-pin CPU connector for the A3000/A4000. We have good quantities of them in stock.