wna@milton.u.washington.edu (Warren Nagourney) (12/21/90)
I have noticed a fair amount of interest in this newsgroup in accelerator boards containing a 68030 (and possibly a 68882). I think some of you may be interested in my experiences with a Gemini 1 board (made by Total Systems, Eugene , Oregon) that I have had for a little more than a year. I will anticipate my conclusions by stating at the outset that, while the CPU performance of the 4 (!!) boards I tried was fine, the SCSI port (on the board) proved to be idiosyncratic (and unreliable) to the point of being, in my opinion, of little use. The system I am using it on is a homemade 512KE (sometimes called a "hackintosh") which has worked very well for several years with a Novy 16 MHz 68020 (and 68881) board using Novy's SCSI port driving a 43 meg Rodime drive. About a year ago, a PAL failed on the Novy board (it has since been repaired) and I decided to sell the Novy and get a Gemini (now called a Gemini 1). My reasons were a desire to use virtual memory and to get the faster SCSI performance supposedly provided by the Gemini board. I also bought 4 megs of SIMMs for the board. It runs at 20 MHz with one wait state. I was very chagrined to find that the Gemini would not boot from the Rodime drive (the Novy had booted without any trouble). Numerous discussions with TSI and Rodime and experiments with various formatting programs led me to the conclusion that this was a hardware incompatibility which would never go away. I accepted it and swallowed the inconvenience of booting from a floppy (the Rodime could easily be "mounted" after booting from the floppy). I then found that I was getting frequent crashes which were connected with hard disk activity (there were no bombs when I used the floppy). TSI sent another board which cured the problem. Unfortunately, a few weeks later, I found that the hard disk would occasionally reset itself; this didn't cause any problems if there wasn't disk activity when it happened. Another board cured this and the system has been working fine for about a year (except for the inconvenience of booting from a floppy). About a month ago, I decided to get a larger drive and purchased a 105 meg Quantum (I had heard that the Gemini worked fine with Quantum drives). After connecting the drive, I noticed that it would make spasmodic seeks, which occurred more frequently as the drive was warming up. I noticed once during boot that the system crashed immediately after one of these "spasms". I then ran a HD test program and found that immediately after each of these spasms, there would be about a 50% probability of the program registering an error (more often write than read). I subsequently found that the spasms were some form of head-positioner recalibration which occurred more frequently when the measured temperature gradient on the drive housing was "high". With the help of the local seller of the Gemini board, I made the following tests. We tried: 3 Gemini boards (including a 4th from TSI to help solve this problem), 2 Quantum drives (the second was an 80 meg), 2 Motherboards, 2 HD formatting programs, 2 HD test programs (including Apple's), Numerous cable combinations, Numerous power supplies. We tried as many combinations of the above as was feasible. Without exception, they all failed! I found that the only way to make the system run the test for an hour without failure was to turn both '030 caches off (this slowed the SCSI port down about 20%, as determined by the time it took to do the tests). Otherwise, when the system was cold, it would almost always fail the test within 5 minutes (when warm, the recalibrations and possible failure would occur with much longer intervening periods of time). In conclusion, I think the board has definite hardware timing problems. It is possible that specially written SCSI drivers could work around these problems without the (unacceptable) necessity of turning off the processor caches; I don't know of any drivers which work reliably with this board. Deciding that life was too short to waste any more time on this unreliable product, I recently sold the board. Again, I should say that the CPU performance was fine and the TSI people were quite cooperative in helping me try to solve the SCSI problem. The board might be a good choice for those using a mac with its own SCSI port. I hope this is of help to those people interested in buying an accelerator board. Warren Nagourney University of Washington Dept. of Physics