Preview: BigRAM 2630, 112MByte for Commodore 2630

[SpeedGeek]

Originally Posted by Schoenfeld

I've thought about overclocking, especially because my PLL will generate a clean clock up to 80MHz (so 40MHz CPU clock in theory), but I'm afraid of capacitive load and rise/fall times being too slow to work reliably. I am therefore developing this with a standard 25MHz clock (50MHz memory clock) and will only guarantee proper function that speed. Anything faster than that is "free bonus"; if it works for a specific setup: Good for you. If it doesn't: Please reduce the clock rate.

Jens

Most A2630 owners will be limited to 28-30 MHz if they want to try overclocking (and I already warned them it will take a huge amount of work to go beyond that and be very risky unless they really know what they are doing). So your Big RAM 2630 probably has enough timing margin for a little overclocking.

I know the capacitance load issue is the main issue on the Big RAM side of the expansion port but on the A2630 side the resistor load issue is really a bigger problem (especially for the address and data buses). There are also some skew problems on the Clocks, logic arbitration delay problems, etc.

The DKB2632 originally used 74F244s for it's address buffers which place a full TTL load on the address bus. The problem is the Address bus was already overloaded (especially after the FCC guys switched the 1K pull downs to pull ups)... and there is much much more!

It's truly amazing that with all the design bugs this combination worked as well as it did!

[Schoenfeld]

Originally Posted by SpeedGeek

The DKB2632 originally used 74F244s for it's address buffers which place a full TTL load on the address bus. The problem is the Address bus was already overloaded (especially after the FCC guys switched the 1K pull downs to pull ups)... and there is much much more!

It's truly amazing that with all the design bugs this combination worked as well as it did!

Address is not a problem, as the address bus is valid a full cycle before /AS (during /ECS), so there is plenty of time for all lines to settle, and for logic to trickle to a stable state before /AS triggers.

1K pulldown on the data lines is a little more challenging, but the LCX16245 drivers that I'm using are strong enough to drive the load - just not really fast due to the high capacitive load with FPU and all the RAMs unbuffered on the bus. My prototype has some trouble doing a -1-1-1 burst, but since the first access works fine, I can always revert to a -2-2-2 burst. I'd like to try fiddling with shifting clocks first, and only if that's not successful, I'll slow down.

The 30MHz limit is only given if you use the 2M or 4M on the 2630. If you can live without that, it should be safe to clock a lot higher. Haven't checked the 68000-access state machine yet, so I don't know if that gives a limit. Did you ever try running at higher speeds without local memory? I know it doesn't make sense without a mem expansion and it will only work properly without DMA devices.

Jens

[SpeedGeek]

Originally Posted by Schoenfeld

Address is not a problem, as the address bus is valid a full cycle before /AS (during /ECS), so there is plenty of time for all lines to settle, and for logic to trickle to a stable state before /AS triggers.

1K pulldown on the data lines is a little more challenging, but the LCX16245 drivers that I'm using are strong enough to drive the load - just not really fast due to the high capacitive load with FPU and all the RAMs unbuffered on the bus. My prototype has some trouble doing a -1-1-1 burst, but since the first access works fine, I can always revert to a -2-2-2 burst. I'd like to try fiddling with shifting clocks first, and only if that's not successful, I'll slow down.

The 30MHz limit is only given if you use the 2M or 4M on the 2630. If you can live without that, it should be safe to clock a lot higher. Haven't checked the 68000-access state machine yet, so I don't know if that gives a limit. Did you ever try running at higher speeds without local memory? I know it doesn't make sense without a mem expansion and it will only work properly without DMA devices.

Jens

Actually the address bus is valid 1/2 cycle before /AS. On the A2630 /AS is delayed by current limiting and pull up resistors so the address is probably valid for 2/3 cycle before the logic sees /AS. Another bug which needs to be fixed to support much higher clock speeds.

I already have the 68000 state machine running at much higher clock speeds with 4MB local fast memory. Did you see the link in post #17 ?

[Schoenfeld]

Originally Posted by SpeedGeek

Did you see the link in post #17 ?

I did, but don't have an Amibay account (and don't want to make one either). Would be nice if you uploaded the parts you want me to look at here, or to eMail them to me.

I also looked at the details of a 8MB/16MB hack that floated around the Aminet in the late 90s, created by two German guys. It appears to be compatible with external memory like the DKB expansion or BigRAM2630, but the final answer can only be given if someone with such a hacked 2630 will try&report.

Jens

[klx300r]

@ Schoenfeld

Hi Jens, since you're here please just answer this quick question about your X-Surf 100 and Rapid Road USB:

question is: do I need to install the EasyNet /ethernet software first to get the ethernet end working on the X-Surf first BEFORE getting the RR going???

[Schoenfeld]

wrong thread - I answer where and when I choose to. If you need support, please follow the usual support chain (reseller first).

Jens

[wmsteele]

@klx300r, no you don't but please be aware RR has multiple issues, see separate thread http://eab.abime.net/showthread.php?t=75396

[SpeedGeek]

Originally Posted by Schoenfeld

I did, but don't have an Amibay account (and don't want to make one either). Would be nice if you uploaded the parts you want me to look at here, or to eMail them to me.

I also looked at the details of a 8MB/16MB hack that floated around the Aminet in the late 90s, created by two German guys. It appears to be compatible with external memory like the DKB expansion or BigRAM2630, but the final answer can only be given if someone with such a hacked 2630 will try&report.

Jens

The A2630SpeedFix archive has a simple modification for the 68000 state machine (to support much higher clock speeds) but you will probably need to disable local fast memory to use it. I don't have your email address so send me a PM with your email address and I'll send you the archive.

Please read the WARNING (on the EAB thread) before you use it.

I haven't had time to study the 8/16MB hack but it would have to map the extra memory into 32 bit address space higher than expansion port address space to be compatible.

[klx300r]

Originally Posted by wmsteele

@klx300r, no you don't but please be aware RR has multiple issues, see separate thread http://eab.abime.net/showthread.php?t=75396

thanks & yes I'm sadly very aware of that thread

[Schoenfeld]

A German member of a1k.org has borrowed me his A2630, which is equipped with the expansion connector on the component side (AKA the "wrong side"). So far, all these models were considered "impossible to expand", but I had no idea why. The attached picture explains it: The memory expansion's connector would conflict with the metal cage of the CPU.

For the special version of BigRAM2630, we'll have to do some cutting on the connector. Luckily, mechanical stability of the connector won't be affected (after all, it's just a pin header and we'll be cutting on the outside box only).

Jens

[Schoenfeld]

Not a preview any more.
Started taking pre-orders today. Circuit boards are already here, but I don't want to assemble them all in one go. Components are expensive, and I prefer to only produce what the market really needs. For this product, it's a bit hard for me to estimate.

Jens

[dirkies]

Well done Jens! Hopefully several Amiga stores will be able to list this as pre-order to spread the word and get some viable sales volume.

[Schoenfeld]

No prob - dealer pricing is available on this one, and my resellers can also place an order before start of production.

Jens

[Cylon]

Is there any indication on the pcb itself to identify the different versions from one another? Otherwise secondhand units might cause problems in the future i'm afraid.

[Matt_H]

Ordered! Thanks.

[Kitch]

Ordered mine as well :-)

[Schoenfeld]

Originally Posted by Cylon

Is there any indication on the pcb itself to identify the different versions from one another? Otherwise secondhand units might cause problems in the future i'm afraid.

One of the two connectors will be cut on one end, otherwise it would not fit with the metal cage. Also, most of the PCBs I have are blue, but a tiny quantity is green. If the "reverse" version stays as rare as it looks from the current amount of orders, I'll simply do "green" for reverse and "blue" for standard, so there will be two indicators that show the "reverse" property.

Jens

[SpeedGeek]

@Jens

I sent you the A2630SpeedFix archive to an email address I found on your icomp site. I see that your user preferences don't allow you to have access to PMs so I can't verify this via PM or even be certain you received the email.

[Schoenfeld]

Got the archive, didn't have the time to look at it. I'd still recommend to keep the 2630 as "stock" as possible, because I really don't want to do support for butchered cards.

If an A2630 has been successfully overclocked, and the CPU frequency is not above 40MHz, the memory expansion *should* work. If there's any trouble, clock down to 25MHz and use original PALs/GALs - that's how I'll instruct anyone who approaches me with stability problems.

Jens

@Jens
Does the Bigram also work on the rev. 6 2630's?

Only the DKB 2632 rev. E (1994) did work on the rev. 6 2630. The more common (in my experience) rev. C of the DKB 2632 only works on rev. 9 2630's, hence the question (as I have 2 rev. 6 A2630's).