SuperBuster Zorro2 Speedup Mod

[SpeedGeek]

Here is the mod I originally described in the "PicassoII burned resistor" thread. Hopefully, the schematic is self explanatory. The only issue not explained is where A4000 owners get the 14 Mhz clock needed for this mod. It should be available on the Alice Chip (see A4000 schematics). I'm using a PAL in my A3000 for this mod but anything similar to the 74LS157 would work just fine. I will try to anwser any questions as they arise. Enjoy!

** NEWS UPDATE **
See post # 7 & 10 below regarding SuperBuster rev 11 issues.

[SpeedGeek]

Originally Posted by chiark

That sounds like a neat hack! Does it not need to keep 7MHz clock for compatibility's sake on the data transfer?

It keeps the 7 Mhz clock between Zorro2 cycles and for the first clock of a Zorro2 cycle. The Zorro2 board still runs @ 7 mhz. So it speeds up cycle termination and data transfer from the CPU's point of view.

[Cosmos]

Overclocking possible from 7 to 10 or a bit more ???

[SpeedGeek]

Overclocking 7 to 10 Mhz is possible but any Zorro2 board which relies to much on the 7 Mhz timing is likely to fail. A few weeks ago I experimented with replacing the motherboard 28 Mhz clock with a 32 Mhz clock. The RGB port has the Xclk and XclkEN signals for a Genlock supplied clock so this is easily accomplished.

This provided an 8 Mhz clock to the Chip and Zorro2 buses. My A3000 booted and ran well except that the A2091 memory failed and the serial port failed. Apparently the Baud rate calculation is also dependant on the 7 Mhz timing.

However, my PicassoII worked fine @ 8 Mhz and was a little faster. Chip memory and blitter were also faster. My 1950 monitor synced up well to the approx. 17 Khz horz scan rate and Amber quite happily scandoubled it to 34 Khz.

It's to bad that someone at C= failed to realize that with the ECS and AGA chipsets programmable scan rates that the chip bus was no longer dependant on the 7 Mhz clock!

[DarrenHD]

Originally Posted by SpeedGeek

Here is the mod I originally described in the "PicassoII burned resistor" thread. Hopefully, the schematic is self explanatory. The only issue not explained is where A4000 owners get the 14 Mhz clock needed for this mod. It should be available on the Alice Chip (see A4000 schematics). I'm using a PAL in my A3000 for this mod but anything similar to the 74LS157 would work just fine. I will try to anwser any questions as they arise. Enjoy!

Hi, nice mod!

Do you have any benchmarks to see how much faster this actually makes things?

[SpeedGeek]

Originally Posted by DarrenHD

Hi, nice mod!

Do you have any benchmarks to see how much faster this actually makes things?

Here are my (Cpu nodatacache) Bustest results*:

PicassoII
BusSpeedTest 0.19 (mlelstv) Buffer: 262144 Bytes, Alignment: 32768
========================================================================
memtype addr op cycle calib bandwidth
user $00300000 readw 925.5 ns normal 2.2 * 10^6 byte/s
user $00300000 readl 1817.9 ns normal 2.2 * 10^6 byte/s
user $00300000 readm 1819.3 ns normal 2.2 * 10^6 byte/s
user $00300000 writew 705.8 ns normal 2.8 * 10^6 byte/s
user $00300000 writel 1407.9 ns normal 2.8 * 10^6 byte/s
user $00300000 writem 1408.4 ns normal 2.8 * 10^6 byte/s

A2091 w/2MB
BusSpeedTest 0.19 (mlelstv) Buffer: 262144 Bytes, Alignment: 32768
========================================================================
memtype addr op cycle calib bandwidth
user $00500000 readw 567.8 ns normal 3.5 * 10^6 byte/s
user $00500000 readl 1117.9 ns normal 3.6 * 10^6 byte/s
user $00500000 readm 1127.9 ns normal 3.5 * 10^6 byte/s
user $00500000 writew 563.8 ns normal 3.5 * 10^6 byte/s
user $00500000 writel 1126.8 ns normal 3.5 * 10^6 byte/s
user $00500000 writem 1126.9 ns normal 3.5 * 10^6 byte/s

* Note: These results are with SuperBuster rev 6 and a 40 Mhz 68040 CPU.

[SpeedGeek]

** NEWS UPDATE **
I recently upgraded my SuperBuster from rev 6 to rev 11. After upgrading I just discovered that this mod no longer works with my A2091 memory! (SCSI controller is OK). However, it still works fine with my PicassoII!

This one caught me by surprise since AFAIK these late rev chips had only Zorro3 changes. This does not mean the mod is completely unusable for rev 11 users but it's another case of "Your mileage will vary" since some boards may work and some may fail.

I am going to try some timing changes and see if this will help. I wish I had more Zorro2 boards to test, but these are all I have. So if anyone has done this mod please post your results. Thank you!

** 2ND NEWS UPDATE **
I found the problem! CCS timing differs on rev 11. However, READLT timing is the more fundamental problem. See new schematic which fixes READLT timing.

[8bitbubsy]

Does this work with Amiga 2000 (latest buster chip)? Or is it already fast?

[SpeedGeek]

Originally Posted by 8bitbubsy

Does this work with Amiga 2000 (latest buster chip)? Or is it already fast?

It's already fast. If you click on the link from chiark's question in post #2 you can see the reason why SuperBuster is slow.

[SpeedGeek]

** 3RD NEWS UPDATE **
Since upgrading to rev 11 I found that Zorro2 cycles were running a little slower than rev 6. But they are still noticeably faster with this mod! Rev 11 is also a little slower on Zorro2 DMA than rev 6. So you take the good with the bad when you upgrade to rev 11.

Also, I recently aquired a GVP Spectrum graphics board which works fine (in Zorro2 mode) with this mod. So that brings the total number of successfully tested boards to 3!

[billt]

I'm curious why something like this wasn't done decades ago.

[SpeedGeek]

Originally Posted by billt

I'm curious why something like this wasn't done decades ago.

Well, several things needed to happen first:

1) Technical info on SuperBuster needed to become available to Amiga hackers and not just developers. (i.e. the Dave Haynie archives)
2) At least one Amiga hacker had to do some benchmarking to verify SuperBuster Zorro2 was at best 1 MB/sec slower than A2000 Buster Zorro2.
3) At least one Amiga hacker needed to spend some time trying to fix the problem rather than just upgrading to Zorro3 boards whenever possible.

[SpeedGeek]

** 4TH NEWS UPDATE **
The last mod worked fine with a CPU up to 40 Mhz, but at 50 Mhz problems occured with the A2091 memory again*. The A2091 is a fickle beast but a good benchmark and reliability test with this mod.

*This may be slightly slower than the previous version with a slower CPU speed but it's better than trying to have more than one mod for different CPU speeds. Also, it would be possible to install a jumper or toggle switch on the 74LS157 select input to disable this mod.