Bandit PCI Overclock Experiment

It seems the Tsunami motherboard is capable of running a 66MHz PCI bus. The other Powersurge-based systems might be able to pull this off as well but I haven't ever tried it.

The first step is checking the Bandit chips in the system. I believe all Macs based on the Powersurge architecture (7300-9600 and related clones) use the Bandit chip as a PCI controller and is usually clocked by a separate oscillator near the PCI slots. I have found at least 2 revisions of the Bandit that were used:
- A model/serial number of 343S1126C with a copyright date of 1994.
- A model/serial number of 343S0020-02 with a copyright date of 1995. The 343S0020 chip is the one that can do 66MHz. The older 343S1126C is not stable at 66MHz, but can function at up to 64MHz; getting it there can be tricky though.

This also requires all PCI cards to be compatible with 66MHz bus speeds as every PCI slot in the system will be 66MHz. 5V-only cards (single notch near the back of the gold connector) will almost never run properly and there's a chance they might end up fried, so keep them out. I know Adaptec's 29160 and 39160 cards are good for 66MHz as are ATI's Radeon cards and the SyncRAID series.

The hack is simple enough: replace the 33.333MHz oscillator near the PCI slots with a faster one. Some systems (mostly clones I believe) use a half-size metal can CMOS oscillator with Tri-state output, based on interpreting the original oscillator part number on the motherboard. Digi-key part number 'CTX258-ND' is what I used as a replacement. I have not tried replacing the surface-mount oscillator that Apple's own systems use. I may try to figure this out eventually.

For systems that use the metal can oscillators, I highly recommend also buying part number 'A463-ND' from Digikey. It is an IC socket that can be soldered in place of the original oscillator and then you can plug the new oscillator into the socket. Far easier to change between different speeds or revert to the original in case of problems. I'm not aware of such a socket for surface-mount oscillators, but soldering wires in place and attaching the oscillator to the end of those would avoid further soldering to the motherboard itself.

The systems with the 343S1126C Bandit chip become unstable at 66MHz. They are capable of up to about 64MHz which gives nearly the same boost in speed. The instability doesn't seem to be due to the overclock itself. Rather I believe this is because the PCI 2.1 spec was released in 1995 which included some tweaks to things like bus timing. The 66MHz speed was a superset of the PCI 2.1 specifications so any card running at 66MHz would rightfully expect to be interacting with a PCI 2.1-compliant system, not an earlier one. For anything below 66MHz it would seem to assume this could be a PCI 2.0 system and will interact correctly. I could be wrong however.

I have seen no negative long-term effects from running at the higher speed. There are however a few downsides and oddities from this hack I have observed:
-- Predictably, the Bandit chips get much hotter especially under heavy use. No instabilities related to heat yet. I added an old Pentium chipset heatsink to each Bandit with thermal tape, just to be safe.
-- OS 9 does not fully boot, but OS X is perfectly fine. OS 9 seems to crash part-way through. Using a 45MHz or lower speed for the PCI seems to work but 50MHz and up does not; MESH SCSI also works at 45MHz but not at 50MHz or above so the two could be related.
-- The MESH SCSI (internal 10MB/s connector) is broken. It puts up a few fussy error messages if you watch the verbose boot process in OS X and no devices can be accessed on that bus. It does not seem to affect overall stability. Grand Central, the external SCSI and all other subsystems seem unaffected.
-- The selection of PCI cards you can use is reduced. 66MHz-compatible or PCI-X only. Anything else will not work properly, and is likely to end up fried.
-- The Radeon 9200 refuses to exist on the same bus with any other PCI cards when the PCI is at 66MHz. The system turns on fine but never begins the boot process or gets video. I did get it to begin booting with a flashed Radeon 7000 on the same bus, but got a kernel panic early in the process and gave up. At 64MHz or lower, it has no problems. I do not know if this would be exlusive to that model or to other Radeons or ATI cards as well. I have not observed this stubborness in any other type of card.
-- Some PCI 2.2-compliant cards (PCI 2.2 does include 66MHz as part of the standard) will not function at 66MHz or the higher PCI speeds. I encountered this with a PC 802.11g PCI card (had the Broadcom chipset so it worked with Airport software); it worked fine down at 50MHz though. -- 66MHz is the absolute upper limit of the overclock. I've tried higher but it rapidly becomes unstable even at 67MHz or 68MHz.

I've compiled some benchmarks to get an idea of how much this hack improves things. I'm keeping them on a separate page here: PCI 33MHz and 66MHz Benchmarks

Quick summary: Disk reads showed about a 40% increase, writes got a 50% increase with the 66MHz clock. Graphics showed and average 25%-30% increase.
Personal experience with Quicktime performance while Quartz Extreme is turned on gives a slight boost over having it off. Combined with the 66MHz PCI, performance subjectively is around 70%-80% better in playback, allowing even 720x480 DivX and XviD files to play at respectable framerates.

Conclusion: This overclock kicks ass.

Original content here is copyright OtakuMegane. Names, content and other stuff NOT my original creation are copyrighted to the respective owners.