Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!dali.cs.montana.edu!uakari.primate.wisc.edu!caen!sol.ctr.columbia.edu!ira.uka.de!fauern!math.fu-berlin.de!grasp1!stephane From: stephane@grasp1.univ-lyon1.fr (Stephane Guillard) Newsgroups: comp.sys.amiga.hardware Subject: Re: 14 Mhz Hack Message-ID: <1991May17.071500.15130@grasp1.univ-lyon1.fr> Date: 17 May 91 07:15:00 GMT References: <1991May15.005045.3401@news.iastate.edu> <1991May15.103225.30756@kuhub.cc.ukans.edu> Organization: INSA Departement Informatique, Lyon, France Lines: 78 In article mmm@reaper.Chi.IL.US (Michael Marvin Morrison) writes: >In article <1991May15.103225.30756@kuhub.cc.ukans.edu> markv@kuhub.cc.ukans.edu writes: >>>>I would also suggest using a 16mhz 68010. >>> I was told by a man at Motorola in the MC680x0 conference that a 16 MHz 68010 >> >>Yes, the fastest 68010 available is a 12MHz part. However, the >>[...] >should generally not overdrive a processor, it *should* work (I run my 25mhz >XC '030 at 28mhz with no problems, but this chip was designed much later). >-- >Michael M Morrison /| |\ >mmm@reaper.chi.il.us | | Cash, for Cache.. | | >reaper!mmm@miroc.chi.il.us \| Hmm.. sounds fair. |/ I run the 881 on my vanilla A2620 at 28Mhz although it's a 16Mhz part without any kinda problem ; many of my friends do the same on the same board (A2620). It seems to us, after a certain time spent in dealing with RotoMola components, that the '20, '30, '881 and '882 seem to be able to run at much higher freqs than what they're supposed to. We even managed to run a 16Mhz '20 at 40Mhz on a LUCAS board. Let me mention that furthermore, there ain't any kind of a temperature raise on the 881 between 14 and 28Mhz. Most benchs let us think that this combination is equivalent to a 68881 clocked at 25Mhz in terms of absolute performance, which ain't bad. But the 'plastic' parts (68000, 68010 and 68008) don't seem to be treatable that way. In terms of a 14Mhz hack, our first experience was long ago (5years), when we plugged a 16Mhz 68000 instead of the 68K on the A1000, running a 14Mhz clock from a 7474 in the 'metal shielded littel box' (you have 2 phases of 14Mhz in that box, and ONLY ONE seems to run, which to my opinion is a matter of S1/S2/S3/S4). So the hack was reduced to plugging a 16Mhz 68K w/pin 15 bent out, and a lead from this pin to a 7474 pin, and WORKED FINE with any software was available to us at this time. (for those who may ask, it was a CERAMIC packaged MC68000/16, from the french Thomson Semiconductors, and we never could find this part again). Unfortunately, we managed to burn this chip (we had only one) in another design, and had to try with a zillion plastic pkged 68000/8, and none worked. We then tried zillions 68010/12s, with no more success. Hope this helps ! PS : all of us also run our vanilia A2620's with 4 meg of 100nsDRAMS (the standard 2620 RAMS), and the jumper for 80ns/100ns selection on the 80ns position. This causes an average 10-15% performance increase again, BUT one of us had problems when his machine reached some temperature point, at which it caused random RAM read/write errors. The solution was to run the 2000 without it's cover, but this is not very good in terms of air flows between the boards. PPS : Another idea about the 2620 : one could take the 68851 (PMMU) away and replace it with an Apple-style PMMU cap (in the former 68020-based MacII, there was a socket for the 68851 (for A/UX), but there was no chip (in a $xxxyyy machine, it's a bit 'mesquin' like we say in french, and the chip was presented as a customer option :-))) Anyway, for the address bus to be translated from the 020 to the rest of the machine, there was a cap in the socket which intrinsics were... leads. This could be (to my opinion, never tried) used in 2620's instead of the 851 (I love it) fur suppressing the uncompressible 1 wait state which ALWAYS occurs on a 2620 because of the 851 (this is extracted from the Roto doc about 851 : even if the mapping is Identity (no mapping), there is 1 wait state, which is suppressed in the 030 with the integration of a subset of the 851 in it). This could lead to another 10% perf increase, I think. Problem : no more SetCPU FASTROM. Anyway : the RAM-release of KS2.0 does NOT need 851 to run in any machine (if it's still assembled at $20xxxx). I'm gonna try this in the next few days ; let me know if you're interested in the results. PPPS : Dave, I know all this AIN't respectfull to all your working hours, but I could not resist trying such things... I promise I'll sacrifice (with pleasure) a marketroid. -- \\\ Only Amiga | Stephane GUILLARD - GRASP INSA Lyon FRANCE \\\ makes it | Tel: +33 72438383 poste 5546 \\\/// possible! | eMail at : stephane@grasp1.univ-lyon1.fr \XX/