(Please note: Overclocking has been known to damage computers! Proceding in such an endeavor is entirely at your own risk. We will not be responsible for damage you cause. Please compute responsibly.)
Once I settled on a motherboard I/O voltage of 3.3, overclocking was relatively simple. (Before changing the I/O voltage from 3.48, overclocking was impossible -- the system wouldn't POST (Power On Self-Test).)
Only one jumper on the PL unit has to be changed to increase the CPU multiplier from 5.0 to 5.5. I tried core voltages of 2.2, 2.3, and 2.4 -- with slightly more stable results at the higher voltages. (Trying to get screen shots of Mechwarrior2: Mercenaries, I did have some lockups at 2.2 and 2.3 volts core. Since then, I've been working on BIOS RAM settings. Somehow, the most agressive RAM settings in the BIOS just don't cut it for the faster AMD K6-2/333... I had to reduce the DRAM read burst to 333/444. All other settings remain the same as for the Cyrix 6x86/133. The reduced DRAM read burst does lower video scores by a few MP/s, but the system does not seem overly slow now, and 366 MHz (66.67x5.5) is rock stable at 2.3v core.)
Changing the core voltage is only a matter of a few more jumpers -- a very simple task with long skinny fingers or tweezers :) The CPU never got noticeably warm even at 2.4 volts. The old Cyrix was uncomfortable to touch when running at rated speed. The K6-2 runs very cool.
The chart below shows overclocked vs. normal Landmark scores.
The next two charts show Wintune97 and Wintune98 results for the overclocked CPU. Note: I was not able to get Winbench98 to run with the overclocked chip and the original DRAM read burst settings. I kept getting a message that interrupts may have been disabled during the test. I am clueless as to what this means or what caused the problem, but Winbench98 would not run at 366 MHz with the system settings as they were. When I get a chance to reload Winbench98 I'll try again with the reduced DRAM read burst settings and see what happens.
Here are the details from the Wintune98 test:
| CPU | (1) AMD K6-2 with MultiMedia Extensions@365 MHz |
| Video Board | Matrox Mystique PCI |
| Video Mode | 1024x768@16bits/pixel |
| RAM | 64 MB |
| OS | Windows 95 4.0.950 |
| Area Tested | Value |
|---|---|
| CPU Integer | 870.8403 MIPS |
| CPU Floating Point | 433.0963 MFLOPS |
| Video(2D) | 44.60299 MPixels/s |
| Direct3D | 28.96266 MPixels/s |
| OpenGL | 6.027558 MPixels/s |
| Memory | 537.5538 MB/s |
| Cached Disk | 57.94345 MB/s |
| Uncached Disk | 2.834094 MB/s |
Some screen shots (~55 KB) are here, showing Mercenaries making a slightly faster 20.8 fps (vs. 19.4 at 333 MHz), and FlightSim98 making 43.5 fps (vs. 40.0 at 333 MHz).
Since I got the RAM settings tuned for stability, I've gone (semi-)permanently to 366 MHz to get some of the speed back from the de-tuning.
UPDATE: After moving one of the RAM SIMMS from bank 1 to bank 2, (so that both banks were on even or odd banks) the DRAM burst read settings could be set to 222/333 again with pretty good stability. I had read something somewhere (:-) about the timings between the different banks being a little different. Evidently, keeping the SIMMS on the same odd/even banks kept the timings enough the same to allow for the faster K6-2 to operate cleanly.
Next, BIOS Considerations
Back to Results
Back to CPU Upgrade Table of Contents
Back to Steel Dust Editions' Home Page
Contact us at SteelDust21@mchsi.com.
This page last updated 2/14/99