NETWORK@FRSAC11.BITNET.UUCP (10/01/87)
Date: 1 October 1987, 10:45:15 GMT From: NETWORK at FRSAC11 To: INFO-ATARI16 at SCORE Facts: 1. the ST line is a 68000 hard and soft box (no '10, no '20, etc) 2. We would all love to have efficient and safe multitasking. 3. The new memory chips are easily available in 100ns to 80ns variety. 4. The problem with the 68851 (?) MMU chip is the 50ns or so delay. 5. We dont want wait states. 6. What is called MMU in the ST is not what the rest of the computer world call a MMU, whose functionnalities are needed to implement safe multitasking (segment protection and translation). 7. It should not be very hard to insert a MMU between the 68000 and the rest of the chips. Question: How to stuff fast memory, MMU chip in the ST, and slightly modify the &|"'@)$#% GEMDOS call to have the necessary process/memory call cleaned up ? May be an answer is OS9, or a modified Minix, just keeping the graphic/window part of the then defunct GEMDOS ??? If you are not harware interested, just forget it, If you think I said something totaly brain damaged, well, let me know. Jean-Pierre H. Dumas network@frsac11 (bitnet) network%frsac11.bitnet@mitvma.mit.edu (arpanet) dumas@sumex-aim.stanford.edu (arpanet)
dyer@atari.UUCP (10/02/87)
> 4. The problem with the 68851 (?) MMU chip is the 50ns or so delay. It is possible to do reasonable MMUs with *no* delays. But you have to be clever, and you have to do it right on the chip. It's hard to build chips in your garage.... > 6. What is called MMU in the ST is not what the rest of the computer world > call a MMU, whose functionnalities are needed to implement safe > multitasking (segment protection and translation). Correct. Atari US doesn't call it an MMU -- we call it a Memory Controller Unit (MCU) -- but our people in Japan decided to change its name on all the schematics. > 7. It should not be very hard to insert a MMU between the 68000 and the > rest of the chips. It is. Very hard. There are NO nanoseconds available -- even one gate delay would kill memory access. -- -Landon Dyer, Atari Corporation {sun,amdcad,imagen,hoptoad}!atari!dyer The views expressed here do not necessarily reflect those of Atari or the AI software that has taken over my brain. YOW! I am waiting for my warranty-expired interrupt!
bammi@mandrill.CWRU.Edu (Jwahar R. Bammi) (10/06/87)
In article <858@atari.UUCP> dyer@atari.UUCP (Landon Dyer) writes: >> 7. It should not be very hard to insert a MMU between the 68000 and the >> rest of the chips. > >It is. Very hard. There are NO nanoseconds available -- even >one gate delay would kill memory access. > It seems to me that un-grounding DTACK and attaching it to a *real* MMU should give you as many nanosecond as you would want, or am i missing something. I will take a few nanoseconds penalty for a real MMU. -- usenet: {decvax,cbosgd,sun}!mandrill!bammi jwahar r. bammi csnet: bammi@mandrill.CWRU.edu arpa: bammi@mandrill.CWRU.edu compuServe: 71515,155
dclemans@mntgfx.MENTOR.COM (Dave Clemans) (10/07/87)
As I understand it the reason for it being difficult to impossible to add a full MMU to the current ST's relates to the video screen refresh. Specifically, memory accesses made to refresh the screen happen over the SAME bus as memory This is normally invisible to you, as on the ST the cpu only needs 50% of the available memory bandwidth, and the video system also only needs 50%. But nothing is left over for extra or longer "cpu" type accesses. The best way to add an mmu would probably be to speed up the memory system bandwidth by some fraction, and probably also speed up the cpu clock. dgc