QL 34th Anniversary new hardware.

A place to discuss general QL issues.
User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: QL 34th Anniversary new hardware.

Post by Dave »

Under guidance from Nasta, I learned recently about how the 8301 generates /DTACK before the data is actually ready, knowing that it'll take most of a clock cycle for the CPU to latch it. Given the memory in the QL was pretty marginal in the first place, this may be the reason the QL runs at 7.5MHz instead of the 8MHz that was originally intended.

Sooo, since I have a few various stage 4MB RAM prototypes, including a couple that just have the CPU on them, I set to work.

Idea: Can I increase the clock rate of the CPU if I delay /DTACK a little?

Method: Use an old 7400, chaining all the gates together, to make a delay. Use this to invert the /DTACK signal two or four times to obtain a slightly delay it.

Execution: I scoped it out and found the delay on the 74HCT00 was only real world around 7ns per gate. This gave me two valid options: 14ns and 28ns. I dug out an old 74LS00 and found the delay was around 16-17nS per gate, giving me options of 33ns or 66ns delays. With no expansions in, and going through the xtals I had, I happily ran the CPU at 8, 10 and 11.0592 MHz. I couldn't get it to run reliably at 12 MHz, but it's an older 10 MHz part so that isn't really surprising. However, it wouldn't run reliably if a Trump Card was plugged in.

Conclusion: If the CPU runs faster, it could respond to /DTACK before the data is ready on the bus. Delaying /DTACK a little allows the CPU to be clocked a little faster. However, this does not provide performance gains linearly with clock speed. Manipulating /DTACK can have unpredictable results. On a base machine, it could yield useful gains, though. If the QL had much faster RAM, there would be much more that could be done with this manipulation.

Future experimentation: Make up a board with the VERY over-clockable 68SEC000. Replace the QL internal RAM with very very fast RAM.

I'm waiting for my new logic analyzer to arrive. It's about 8x faster than my current one. Then I'd be able to get a much better insight into what's going on in there. The current one, once you're at a high enough sample rate to provide useful info you are down to four channels - I need 16 channels all up in this.... I also need to make a doohickey to convert sine to square waves because my signal generator can produce 25 MHz sine waves but only 6 MHz square waves.

If there's a #1 target to replace with a more capable equivalent, it's the 8301.


User avatar
Peter
QL Wafer Drive
Posts: 1953
Joined: Sat Jan 22, 2011 8:47 am

Re: QL 34th Anniversary new hardware.

Post by Peter »

Pr0f wrote:68030 offers better pipelining and faster burst fill memory access that the 68020 doesn't offer. It also supports faster clock rates than the 68020, and still offers the dynamic bus sizing, which makes it significantly easier to connect to a QL bus, than say a 68040 or later variant.
If you have fast memory, the 68030 practically only wins something from clock frequency. For example, both 68020 and 68030 need 2 cycles for a best case primitive register-register operation. Whereas the CPU inside the Q68 needs just 1 cycle, if run from SRAM or (not yet implemented) cache. So I would not expect wonders from a 68030.

When you say "connect to a QL bus": What makes sense to connect, except the QL internal I/O, which can be done by less than 8 address lines, all other lines fixed?
QL video output is a pain (at least on all three QLs I own), and using QL internal RAM does not make any sense.

I'm asking this, because reduding the number of QL side address lines to less than 8 allows to simplify level shifting or signal conditioning a lot when connecting to 3.3V circuitry.


User avatar
Pr0f
QL Wafer Drive
Posts: 1298
Joined: Thu Oct 12, 2017 9:54 am

Re: QL 34th Anniversary new hardware.

Post by Pr0f »

Peter wrote:
Pr0f wrote:68030 offers better pipelining and faster burst fill memory access that the 68020 doesn't offer. It also supports faster clock rates than the 68020, and still offers the dynamic bus sizing, which makes it significantly easier to connect to a QL bus, than say a 68040 or later variant.
If you have fast memory, the 68030 practically only wins something from clock frequency. For example, both 68020 and 68030 need 2 cycles for a best case primitive register-register operation. Whereas the CPU inside the Q68 needs just 1 cycle, if run from SRAM or (not yet implemented) cache. So I would not expect wonders from a 68030.

When you say "connect to a QL bus": What makes sense to connect, except the QL internal I/O, which can be done by less than 8 address lines, all other lines fixed?
QL video output is a pain (at least on all three QLs I own), and using QL internal RAM does not make any sense.

I'm asking this, because reduding the number of QL side address lines to less than 8 allows to simplify level shifting or signal conditioning a lot when connecting to 3.3V circuitry.
Well - all good points - so let me counter

1) Q68 'core' isn't available as a processor option, the 68EC030 is. If you made it an available core it could be the basis of either an add on, but the Q68 is already a ready to run computer.
2) If I build a 68030 board, it will be to work with an existing QL - because I have one I can test with - it would need to make use of the ROM and screen memory and I/O at least - if I get it to work with my DIY design as well - it could make more extensive use of internal ROM and a better video circuit.
3) Aside from the size of a transceiver chip, converting from 16 pins or from 8 is just more lines on a pcb - for either 5v or 3.3v


User avatar
Peter
QL Wafer Drive
Posts: 1953
Joined: Sat Jan 22, 2011 8:47 am

Re: QL 34th Anniversary new hardware.

Post by Peter »

Pr0f wrote:1) Q68 'core' isn't available as a processor option, the 68EC030 is.
No, the TG68K.C processor core is available under the terms of the GNU Lesser General Public License. Now that the years of debugging work for the QL are done, it is not so difficult to use anymore.


User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: QL 34th Anniversary new hardware.

Post by Dave »

Peter wrote:No, the TG68K.C processor core is available under the terms of the GNU Lesser General Public License. Now that the years of debugging work for the QL are done, it is not so difficult to use anymore.
Did you feed your bug fixes back to the main source tree, and were your patches accepted?

Do you plan to ever open up this work under a public license, once commercial activity is exhausted?


User avatar
Peter
QL Wafer Drive
Posts: 1953
Joined: Sat Jan 22, 2011 8:47 am

Re: QL 34th Anniversary new hardware.

Post by Peter »

Dave wrote:Did you feed your bug fixes back to the main source tree, and were your patches accepted?
Yes and yes.
Dave wrote:Do you plan to ever open up this work under a public license, once commercial activity is exhausted?
The CPU is available under the terms of the GNU Lesser General Public License.
Last edited by Peter on Sat Mar 10, 2018 10:43 pm, edited 1 time in total.


User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: QL 34th Anniversary new hardware.

Post by Dave »

I meant the entirety of Q68? Not just the 68K core.

I mean, it's been years since Q40/Q60 were economically active and you haven't released that into the wild, so it looks like it might be a long time, or never, before Q68 goes into the pool.

With my projects, some will be open source right away. Others, which require more development work and expense, at the time of release we'll start either a 5- or 10-year clock, after which the logic becomes public domain. This gives us a bit of time to exclusively sell the items to recover the development costs (or some of them) without selling our homes, but ensures that the knowhow does become public at a known date in the future. This way, if something happens to us, our wishes are already publicly stated. We'll ensure that someone trustworthy within the community has the .pld and jedec files, schematics, etc.

We are all getting older, and we've lost enough people to know better than to not plan ahead for it and make our wishes known. :)


User avatar
Peter
QL Wafer Drive
Posts: 1953
Joined: Sat Jan 22, 2011 8:47 am

Re: QL 34th Anniversary new hardware.

Post by Peter »

Well, as you know the QL-SD design is free and open source. It is a simple design, where I did not suffer a lot from time consuming questions. Designs like the Q60 are at a level of complexity where publishing them would not make the machine available - simply because expertise is needed to build. And I'm not in the mood to spend time and teach that expertise, nor to answer the questions or problems that would arise.


Derek_Stewart
Font of All Knowledge
Posts: 3932
Joined: Mon Dec 20, 2010 11:40 am
Location: Sunny Runcorn, Cheshire, UK

Re: QL 34th Anniversary new hardware.

Post by Derek_Stewart »

Hi Dave,

Your expansion plans for the QL sound good. But why not goto the 68040 or 68060. I am not really intetested in the 68030...

I built many Q60 boards in 2002. A lot of work to solder many joints and there are some construction issues that if care is not taken failure could occur.

As Qbranch found out, there requires alot of attension to detail. Which they did not have, ad thet suffered failures due bad component choice and useless solderong skills. Only the 1st batch of Q40s built by Tony Firshman were consistantly good.


Regards,

Derek
User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: QL 34th Anniversary new hardware.

Post by Dave »

From the PCB design and assembly and CPLD programming point of view, Q60's an extremely simple design. The hard work was in creating the custom logic, and in porting SMSQ to the new hardware.

If I had the CPLD contents, ROM contents and a detailed schematic, I think I could be churning those out within a couple of months. That is the power of your good design. The Q68 even more so. It's trivial to assemble. Programming the FPGA? Plug in the JTAG, press [PROGRAM], let it verify, plug in the test SD, let it check all the systems, done.

That shouldn't be used as a reason to keep the IP proprietary long after you're gone. It actually creates a bit of a problem, because you're essentially doing an Apple and saying "I am taking away your ability to repair this, or to understand it. There should be spare CPLDs available for repairs, etc.

And yes, I would clone the Q60 in a heartbeat. It's the pinnacle of QL development and it is a shame it is not available.


Post Reply