Looking for a 68060 with FPU & MMU

[SpeedGeek]

PM sent.

We'll see if SpeedGeek can make a slow A3640 fast .

Thanks guys, I really appreciate your generosity but there were a few reasons why I never built Mozart's adapter:

1) It's to tall to fit in an A3000D
 2) The clock circuit needs to be modified to buffer  PCLK so it can be used with the stock A3640 clock divider (74Fx803)*
3) It required some software patching (on a Zorro2 PCB) to disable the FPU (but Cosmos probably has exec.library patched by now).

Apparently, you haven't seen my 50MHz A3640 + 68030 state machine mod thread on Amibay?

P.S. Yes, a 50 MHz 060 is faster than a 50 MHz 040 and yes the 060 can run at higher clock speeds than 50 MHz but my A3640 outperforms even the fastest 3rd party 060 accelerators in access speed to all motherboard resources.

*My A3640 has a custom clock circuit which now only generates 50/100 MHz clocks but 99.99% of all A3640 owners still use the 74Fx803.

[SpeedGeek]

No problem. So you don't need the 68060 then?

That's between you and Cosmos but he really wants me to have this adapter which is useless without an 060.

You would also need a voltage regulator for a 68060. The 68060 is probably not worth it if you have to turn off the FPU (although you could use a cheaper EC/LC 68060 that overclocks more). Why doesn't the FPU work?

The 68040 is stronger in memory than the 68060 in some ways and it's very consistent in integer performance with any 68k code. The 68060 has twice the caches, higher clock speeds, better instruction and addressing mode timings (but missing int 32x32=64), superscalar (but needs optimized code for best performance), lower electrical draw and a better FPU. If you take away the 68060 FPU, the 68060 still wins easily in int speed but the 68040 is closer in no-hassle versatility and 68k compatibility.

How much faster are your access speeds to motherboard resources with a 68040 A3640? My CSMK3 68060 SysSpeed memory results:

Cache Read 283.44
ReadROMb 35.15
ReadROMw 65.62
ReadROMl 85.75
ReadFastb 38.34
ReadFastw 53.04
ReadFastl 65.41
WriteFastb 30.96
WriteFastw 48.18
WriteFastl 48.16
Fast2Fastb 17.99
Fast2Fastw 26.46
Fast2Fastl 29.93
Fast2Fastm 27.58
Fast2Fast16 37.04
ReadChipb 1.11
ReadChipw 2.22
ReadChipl 4.44
WriteChipb 1.69
WriteChipw 3.38
WriteChipl 6.75
Chip2Chipb 0.67
Chip2Chipw 1.33
Chip2Chipl 2.66
Chip2Chipm 2.68
Chip2Chip16 2.67
Fast2Chipb 1.68
Fast2Chipw 3.36
Fast2Chipl 6.72
Fast2Chipm 6.20
Fast2Chip16 6.70

The voltage regulator is included with the adapter. You only need to  disable the FPU for early startup then 68060.library enables the FPU at boot time.

My A3640 benchmarks are posted on the Amibay thread:

http://www.amibay.com/showthread.php?19993

P.S. The results are "Pure Bullsh*t" if you have the data cache enabled regardless of using SysSpeed or Bustest (unless your trying to determine the speed of the data cache).

[SpeedGeek]

Many thanks to my good friend SuperCosmos (and thanks to matthey also) for the 060 adapter and Rev. 1 CPU!!!

After many hours of work (and some hair pulling, head scratching and saying a few 4 letter words) I finally got the adapter working! :cool:

First, was the hardware fix (getting my A3640 back to 25 MHz) and that was the relatively easy part. Then adding the PCLK skew compensation mod (again relatively easy).

Next, building a new Kickstart ROM compatible with the 060. Should have been relatively easy but very difficult due to some corrupt files in my Remus build directory. (Also, complicated a bit by the 2 separate 040 and 060 exec.library's needed). :furious:

Anyway, the end result is in A3660.jpg (for you curiosity seekers).

[SpeedGeek]

@matthey

I've been very busy with software patching and debugging so I don't know when/if I'll find time to do a full writeup (Cosmos want's me to work on overclocking the adapter next). But thanks again for your assistance...

@NorthWay

Motorola made significant changes with each new generation of 68K CPU so code written for earlier generations usually needed some patching to make it work. Anyway, I now have a patched exec.library (45.20b) working with 030, 040 and 060! I'm sending it to Cosmos for additional beta testing, but as far as I'm concerned the Zorro2 PCB and Diag code @ $F00000 are obsolete...

[SpeedGeek]

I know. I patched my own Exec years ago to properly detect the 060, fix its status, and set the flags for it.

So I didn't really see the need for having more than one version.

I wasn't aware that you (or anyone else) had previously patched exec for the 060. Otherwise, I certainly would have been willing to try yours. What version of exec is your patch based on?

If it's and older version than BB2 some users may still prefer a newer version.

[SpeedGeek]

OK, as promised (to Cosmos) the first episode of A3660 overclocking has begun!

I attached a small cooling fan to the Rev. 1 CPU + Ramsey's CLK90 is modified for the A3000*.

Hopefully, as I acquire the additional components needed and time permitting clock speeds will increase with future episodes.

*Note: There is no practical overclocking limit with the adapter itself.      

[SpeedGeek]

OK, episode #2 of A3660 overclocking is completed!

This is the upper limit for this Rev. 1 CPU... Also, SDMAC is working at it's maximum CLK speed. I made a custom oscillator and added an extra 10 uF cap to the top PCB for this project.

Higher clock speeds are still possible but will require a late Rev. CPU and a custom timing hack to keep SDMAC happy....

[SpeedGeek]

OK, episode #3 of A3660 overclocking is completed!

Once again, many thanks to SuperCosmos for the rev. 6 060 (and for his patience with my stupid mistake delaying this update for another 2 months)
 
The custom oscillator I made used a dual frequency crystal so changing  the frequency was quite easy. The more complex part of this episode was  the CLK D4 GAL designed as the solution to keep SDMAC happy.

The CLK D4 GAL switches the mobo to a 1/4x clock during DMA so SDMAC is  supplied with clock speed it can handle and the mobo runs at the 1/2x  clock speed when the CPU masters the bus. See attached images.

This also works quite well with the 040 but I was really saving it for this episode!

Higher clock speeds are still possible but I will have to source some oscillators or crystals for the next episode.

[SpeedGeek]

OK, episode #4 of A3660 overclocking is completed!

Unfortunately, SysSpeed becomes less accurate at higher clock speeds so I have included a Scout result as well.

The CLK_D4 GAL signal BGACK_040 is now connected to VCC. This always  runs the mobo at the 1/4x clock speed. Also, more A3000 mobo and CLK90  timing mods were made to keep the following clocks skew compensated and  stable:

68060 CPU = 80MHz
A3640 logic = 40MHz*
A3000 mobo = 20MHz

Note: Considering that the A3000 is a timing intolerant beast  overclocking the mobo to 40MHz was not a practical consideration here.

*A3640 state machine mod required

[SpeedGeek]

Fantastic !

Will works on the A4000 ?

It should work with the A4000 but I don't have one to test it on. I would expect the A4000 to have more flexible timing options for CLK90 since I'm lucky to find one which keeps Ramsey and SDMAC working at non-standard clock speeds. :rolleyes: