Experimenting the I/O speed
The area which has given me the most problems is the increased PCI speed. At 66 MHz the PCI bus runs at half the system bus frequency. At 100 MHz it runs at one third and at 133 MHz one quarter of this frequency. Thus if we increase the system bus, it also affects the PCI bus:
|System bus speed||Bus factor||Resulting PCI speed|
|66 MHz||The half||33 MHz|
|75 MHz||The half||37.5 MHz|
|83.3 MHz||The half||41.6 MHz|
|100 MHz||One third||33.3 MHz|
|112 MHz||One third||37.3 MHz|
|133 MHz||One quarter||33.3 MHz|
|153 MHz||One quarter||38.25 MHz|
First I made it run at 3½ X 75 MHz. It worked fine with CPU, RAM (10 ns SD) and hard disk (IBM DHEA). But the net card (a cheap 10/100 Ethernet card) refused. When I copied large volumes of files on the net, it froze up - stopped. It was quite obvious that the problem was in the net card.
I had to accept the traditional 66 MHz. But to soothe the pain, it turned out to run excellently with a clock factor of 4 - thus at 266 MHz.
Within a couple of weeks I was in the mood to experiment again. I now found an adjustment in the setup program. It is called PCI latency. It is not explained anywhere, but it has a default value of 32. I increased it to 36 and increased the bus frequency to 75 MHz – it works. Now the net card runs without problems.
Then I hoped to speed the system bus up to 83 MHz, which should give a significant performance improvement for all RAM transport. My 10 ns SD RAM can certainly handle 83 MHz. But no, it did not work. Regardless of the PCI latency, the PC would not start. This indicates that the PCI latency setting does not work like I expected. Maybe it has nothing to with this - I do not know.
My explanation is, that the video card could not tolerate the 41.5 MHz PCI frequency. Nothing appeared on the screen.
Now the PC runs fine at 4 X 75 = 300 MHz. There can be an occasional unexplained break-down in Windows 95 (that happens under other circumstances also), which I blame on the drastic over clocking. However, the advantages of the significant performance improvement far exceed the annoyance of these small interruptions, which happen far from daily.
Windows NT 4.0 does not install with over clocked CPU. The program tests for "genuine Intel", and seems to register the change in clock frequency. And then it will not work. But if you install NT first, then you can over clock afterwards and NT will work. Actually NT is quite sensitive. One of my friends experienced some peculiar errors. The solution turned out to be moving the RAM module from one socket to another!
Fake Pentium IIs
Since some Pentium II-233 perform very well at 300 MHz, they have been sold as such ones. To test your own Pentium II, you can download this test program from C't, which can check your Pentium II. Here is the interface of the Windows 95 version, which correctly detected my CPU to be over clocked:
To set the clock doubling, some small switches (called jumpers) have to be reset. They are located on the motherboard, as you see here:
You can read in the motherboard manual how to set them. Or you can look at the motherboard! In the picture below you can see some of the printed information on the motherboard (this is an ASUS TX97 with a Socket 7).
Here you can read which jumpers to set to select clock doubling 1, 1½, 2, 2½, 3, 3½ and 4 for 6 types of processors:
On modern motherboards you may find a software solution to the settings, and that is a lot better.
Also see: Module 3e - about the latest CPUs.
Read more about the boot process and system bus in Module 2b
Read more about I/O buses in module 2c
Read more about the motherboard chip set in module 2d
Read more about RAM in module 2e
Read about EIDE in module 5b
Copyright (c) 1996-2005 by Michael B. Karbo. www.karbosguide.com.