I have an ATi 9250 pci. so far it works in EVERYTHING. Even 5v pci slots. I've ran it overclocked and its always worked.

Be careful about pci 9250's though. Amiga crackpots have overpriced them to high heaven. at least the 5v ones. i'd read into it more, but I couldn't care less about amigas.

luckybob wrote:

I have an ATi 9250 pci. so far it works in EVERYTHING.

What computer did you put an ATI 9250 in for 40 MHz PCI operation? Are there non-486's which can run with a 40 MHz PCI bus?

When you overclock anything from a pentium2 to a pentium 3. The pci clock goes up with the fsb clock. I know for a fact, the Assu P3B-F (440bx) if you set the FSB to 150, you get a 38mhz pci bus. There is one setting where 120mhz fsb gives you a 40mhz pci. I'd used that in the past, but never for any length of time. I remember getting hard drive errors, video worked, but the hard drive i was using couldn't cope.

Voodoo 3 2k with sgram up to 41.5mhz and solid as a rock.

luckybob wrote:

When you overclock anything from a pentium2 to a pentium 3. The pci clock goes up with the fsb clock. I know for a fact, the Assu P3B-F (440bx) if you set the FSB to 150, you get a 38mhz pci bus. There is one setting where 120mhz fsb gives you a 40mhz pci.

Or super socket 7 at 75 or 83 MHz FSB. For 100 MHz FSB there is usually a 1/3 divider to get 33 MHz PCI (and 66 MHz AGP) but for lower FSB usually a 1/2 divider is used so PCI will be running at 37.5 or 41.5 MHz.

I would assume that any chips that are also available as AGP will support a higher PCI clock but if any other clocks on the board are derived from the PCI clock that might not work.

I never overclocked the PCI. I used to overclock the ISA bus in a 386, though (40/3 = 13.3 MHz). Worked fine.

I've been playing around with the 40 MHz setting on some PCI 486 boards. My S3 Virge DX 4MB card will flicker from time to time after 10 minutes of uptime. The GPU also feels pretty warm for this card.

Diamond Voodoo Banshee 16 MB doesn't seem to have any of the above mentioned issues, but I am just doing some basic DOS testing so far. The Banshee also has a fan on the heatsink.

I'll generate a list to the OP if there are sufficient respondents.

Matrox M3D....... is not one of them 🙁 Freezes.

You get improved benchmark scores running at 40mhz?

dirkmirk wrote:

You get improved benchmark scores running at 40mhz?

Are you asking me? For heavily graphic applications, yes there is improvement, however reduced cache and memory timings for all but 256K double-banked cache sorta negate this improvement.

I am doing some benchtop "research" in attempt to get to the bottom of how increasing cache size, cache banking, and the number of memory sticks adversely affects cache and DRAM timings in the BIOS for FSB frequencies of 40 MHz. I will be using 4 common PCI 486 motherboads with different cache configuraitons (256K double, 512K double, 512K single, 1024K double) and memory configurations. The results are a little shocking, but I'll write up on this when I'm finished; it is a different topic from the OP. During this effort, I did run accross some graphic cards which didn't like 40 MHz, which is how I got the idea for the current topic.

The main problems when overclocking PCI are the hard disks plugged to the usual PCI IDE controller. Typical issues are data errors.That may not be obvious on first glance since the error correction with higher UDMA modes does its job, but the transfer rate will drop considerably. For some HDD the electronics just dies after a while.

elianda wrote:

The main problems when overclocking PCI are the hard disks plugged to the usual PCI IDE controller. Typical issues are data errors.That may not be obvious on first glance since the error correction with higher UDMA modes does its job, but the transfer rate will drop considerably. For some HDD the electronics just dies after a while.

I believe the Promise Ultra ATA PCI cards from the mid-2000's are spec'd for 66 MHz. They work fine in the PCI 486's I've tested. I am limiting my study, though, to cache and memory. The odd combinations of cache and DRAM timings for different cache configurations is something which caught me by surprise. In short, don't think that because you can run 256K at the best cache and DRAM timgings means you can run 512K or 1024K at these same timings. It seems the Northbridge needs more time to decode the added information. Makes me wonder the true advantage of more cache (at 40 MHz). Memtest provides a lot more insight for stability than HIMEM alone. This will take some time to complete.

Any more PCI graphics cards known to work well or poorly at 40 Mhz?

I would say if it works in a 66mhz PCI-X slot, then it should also work in a motherboard that overclocks the PCI bus.

luckybob wrote:

I have an ATi 9250 pci. so far it works in EVERYTHING. Even 5v pci slots. I've ran it overclocked and its always worked.

Be careful about pci 9250's though. Amiga crackpots have overpriced them to high heaven. at least the 5v ones. i'd read into it more, but I couldn't care less about amigas.

And Mac people push up the prices of 9200/9250 cards, too, because they are some of the best cards that you can flash for use in a Powermac that has only PCI slots for expansion. The Mac versions are expensive but the PC versions are cheap enough that even if you had to replace the PC ROM chip with a higher capacity one to contain the larger Mac ROM file, it would still be worth doing.

elianda wrote:

The main problems when overclocking PCI are the hard disks plugged to the usual PCI IDE controller. Typical issues are data errors.That may not be obvious on first glance since the error correction with higher UDMA modes does its job, but the transfer rate will drop considerably. For some HDD the electronics just dies after a while.

Some boards handle this a lot better than others other than that you can always try to cherry pick what cards you are going to use. The 66mhz compatible controllers do away with this issue fairly well.

Ok, maybe I was not exact enough. I wanted to say that if you have a Standard PCI ATA-Controller that is often included in the southbridge of the chipset, then the transfer clock for UDMA modes is directly linked to the PCI clock. So if you overclock the PCI then in such direct clock linked case the HDDs drive electronic has to cope with the higher data transfer clock given by the integrated controller. Some drives do not work well when you push e.g. UDMA133 to UDMA160 (40 MHz).
If you have a seperate PCI card with a seperate clock then this is no problem.