A3630 vs A3640 change or not?

[SpeedGeek]

I tend to favor the A3640 but I'm biased since my A3640 now gives a WarpEngine a good run for the money. (Some extra memory would be nice though).

[SpeedGeek]

@mechy
 
Yes, motherboard fast memory is still slower than WarpEngine memory. I managed to get 19-20 MB/sec with PAL/GAL mods.* (I have plans for a motherboard SRAM project which will be faster than WarpEngine memory). My A3000 does 3-3.5 MB/sec SCSI I and I'm satisfied with that. The PAL/GAL's may run a little warmer but they can tolerate it since their rated to run at 50 Mhz min.
 
The Delay Line hack does not delay memory access it just compensates for higher clock speeds. What actually delays memory access (and other motherboard resources) is the sloppy state machine logic which adds 2 wait states to every bus cycle. *I fixed this and thats how I get the results above.
 
My A3640 is a rev 3.2 and has the caps installed correctly. It has the original 25 Mhz MC68040 overclocked with cooling fan. The WarpEngine will also add synchronization delays to motherboard resources if it runs at a different clock speed than the motherboard. My A3640 runs synchronous the motherboard and avoids these delays

[SpeedGeek]

@mechy
 
Synchronization delays add wait states when the WarpEngine (running at a different clock speed) would access any motherboard resource. This is not a factor in WarpEngine local fast memory but eventually the CPU has to access motherboard resources right? I already have the WD rev 8 SCSI chip in my A3000. The SCSI II command set does not change the SCSI I max transfer rate of 5 MB/sec and you still have a SCSI I hardware interface on this chip. Even if you have a 5 MB/sec hard drive SuperDMAC and SuperBuster will limit transfers to approx 3.5 MB/sec. This could be increased a little with a 30 Mhz overclock but I have to run them at 20 Mhz to maintain a sychronous relationship.
 
Most accelerators do not support burst memory access to motherboard fast memory either because it requires a sychronous relationship to the motherboard (Ramsey) or a more complex state machine design. In any case Ramsey would only support Burst with static column RAM installed. So A4000 owners and A3000 owners with page mode memory could never use it anyway. The original rev 04 Ramsey does not have a "page detect bug" it simply lacks page mode support of any kind but works with page mode memory if burst and page mode are disabled. "SkipRamsey" doesn't work with an overclocked motherboard and you still need 60-70 ns DRAM. Rev 04 Ramsey does not support software skip mode but A3000's have a motherboard speed jumper to set this mode.
 
I never intended to hijack this thread and Q&A on my A3640 speedup mods should really be on this or the EAB thread:
http://www.amiga.org/forums/showpost.php?p=597580&postcount=9
Thanks!

[SpeedGeek]

@mechy
See this thread:
http://eab.abime.net/showthread.php?p=777332#post777332
 
Do you still think I should give up my A3640 for a WarpEngine?

[SpeedGeek]

@mechy
I had an even faster benchmark with 60 ns DRAM and Ramsey @ 4 clock cycles (24MB/sec fast memory). But I had to disable my 16MB bank select PAL leaving only 4MB of fast memory. Also only 1 of my 60 ns 16MB SIMM's was fast enough. I'm currently looking for a 50ns 16MB SIMM my young apprentice!
 
As far as the A4000 64MB hack Dave Haynie said exactly the same thing I said on comp.sys.amiga.hardware. But it won't work on the A3000 and he did not address the issue of where to map the extra 16MB banks. (Probably in RAMSLOT memory space if you want auto config or above RAMSLOT space with addmem config if you want to use WarpEngine local memory).
 
@Jeff
ZorRam is only 1/2 to 2/3 the speed of motherboard fast memory!