View Single Post
Old 17 February 2012, 13:28   #2320
Join Date: Jun 2010
Location: out in the wild
Posts: 1,132
Originally Posted by SpeedGeek View Post
The A2630 has many design bugs (it just barely works @ 25 MHz) including the 1K pull-downs which are sometimes installed as pull-ups! You will spend 20x the time fixing the bugs that you spend replacing the oscillator and 68030/68882 chips.
I figured that much from all the different versions of the PAL chips that Dave Haynie has published (I have rev.3, rev.4 and rev.9 here). Having all those signals travel over a huge board and through several chips makes it practically impossible to simulate the design (not speaking of the temperature drift of the delay-lines). Like I wrote: I thought about it, but won't do it. It's more work than making my own accel for the 2k, and that will automatically happen when I finish the ACA520 design.

Originally Posted by SpeedGeek View Post
I managed 8 + 2 + 2 + 2 clock cycles with burst @ 52 MHz on the DKB2632 and I may be able to get 6 + 2 + 2 + 2 with faster address buffers. I would not try to run SDRAM @ 2x this clock. If I could get 4 + 1 (or 2) etc. and eliminate expansion board memory I would be happy with that. you can appreciate the 2-1-1-1 burst on the ACA1231 :-)
I've thought about other optimizations, for example using the ECS (external cycle strobe - nothing to do with the chipset) signal to initiate opening the SD-Ram row. ECS comes one cycle earlier than AS, and if I read the 68030 manual right, then AS will *always* come if caches are disabled. Cache=off probably sounds bad, but with a 0-waitstate STERM-terminated fastmem access, it might just give the 68030 another kick over a 2-1-1-1 burst.

Have you seen the write-performance of the ACA1231 to fastmem? It's just under 45MByte/s, where the ACA1230-56 gets to 41.8MByte/s. Note to interested readers: There is no burst in write access, so write will always wait for a full "open-row-write-close-row" cycle.

I'll publish an update later today that should also work with the 30MHz A600 accelerator.

Schoenfeld is offline  
Page generated in 0.05368 seconds with 9 queries