home *** CD-ROM | disk | FTP | other *** search
-
- Copied From The CyberNET BBS CLINK 911:6230/2.0 714-378-0085
-
- item: 2
- subj: 50mhz FPU on A3000
- from: Oliver Rokoff
- on : Tue 1-Feb-1994 6:43p
-
- I recently read the following post on Usenet, though i haven't got the guts to
- try it out, it does sound interesting:
-
-
- *** Area: USE_HARDWARE Date: 11 Jan 94 13:15:50
- *** From: Mike Noreen (92:901/231.14)
- *** To : All
- *** Subj: A3000 FPU run at 50Mhz successfully!
-
- I just thought I'd tell all A3000 owners something they might find
- interesting:
-
- I right now run my FPU at 50Mhz, and it works!
-
- I've been rendering a 30sting:
-
- I right now run my FPU at 50Mhz, and it works!
-
- I've been rerame animation for the last 24 hours to test
- the FPU, and sofar there's been no problems. The temperature in the room
- is 21 degrees centigrade, and the temperature of the FPU is still under
- 40 degrees (according to a thermometer I left in the immediate vicinity
-
- of the FPU), so there doesn't seem to be any heat dissipation problems.
- I'll continue to render for a few days more to make absolutely
- sure the FPU can handle the doubled speed.
-
- According to AIBB I'm 11-96% faster than an ordinary 25Mhz A3000 on
- floating point (I assume the differences depend on the amount of memory
- accesses done by different types of tests). In the beachball test (the
- interesting one) I'm 40% faster than a 25Mhz A3000.
-
- SysInfo claims that I'm doing 1.1MFlops, as compared to the 0.66 I did
- before the hack.
-
- The beauty of this hack is that it didn't cost me a penny.
-
- This is the way it was done (thanks to Stefan Johannesdal for telling
- me how to do it!):
-
-
-
- (continued next post)
- --- CNet/3
- * Origin: Amiga's Alley (911:5050/1.0) Washington Hub Kent, WA 206-859-0525
- 14.4 HST (911:5050/1)
-
- End of item. Respond or Pass>
-
- item: 3
- subj: 50mzh hack on 3000(cont.)
- from: Oliver Rokoff
- on : Tue 1-Feb-1994 6:45p
-
-
-
-
-
- 1) If you look at the motherboard from the front of the computer, you
- will see the FPU close to the front/middle of the computer. Locate
- pin 11 (it's the second from right on the far side of the 68882).
- From this pin a copper wire goes out then right, and then disappears
- down through the card on the right side of the FPU. This wire is the
- FPUCLor CLK30), and has to be cut (preferrably on the underside of
-
- the motherboard. It resurfaces one cm to the right of where it went
- down, and me and my more soldering experienced friend cut it between
- these two points).
-
- 2) Locate resistor R120 (22 ohm). It's just left of the 50Mhz crystal,
- and the far side of this resistor is to be connected to the FPUCLK
- pin of the 68882. We did it this way: we soldered a wire on the underside
- of the card from the point where the wire after the R120 goes through the
- motherboard, to the point where the FPUCLK wire from pin 11 went through
- the motherboard the first time.
-
- That's it. We tried to do it neatly, and to make it as hard to spot as
- possible for service technicians (I've still got warranty...maybe), but
- if that's not a concern you can simply solder a wire between R120 and
- the FPUCLK pin, after cutting the FPUCLK wire. The greatest problem was
- actually dismantling/reassembling the A3000...
-
- ***************************************************************************
- If anyone got any programme which tests the reliability of FPU's, I would
- GREATLY appreciate it if you'd UUencode and mail them to me! PLEASE!
- ***************************************************************************
-
- UUencoded AIBB module available on request.
-
- DISCLAIMER: I am not responsible for any errors which might have gotten
- into this post, or for any mistakes you might do when trying to repeat
- this hack, nor for any damage it might cause. It works for me, on my
- A3000, but there's no guarantee it will work for you. Don't try this
- unless you're absolutely sure that you can handle it. This hack pushes
- the FPU beyond it's specs, and might shorten the FPUs lifespan, and
- cause computational errors. I have no indication this is the case, but
- it is a definite possibility.
-