66MHz PCI in PCI PowerMacs?

[TylerEss]

http://computor.insaneotaku.com/pci_overclock.html

Has anybody here ever tried this? It seems like hocus-pocus to just stick a 66MHz oscillator in there and have it magically do its thing, but I've heard of stranger. I've got a 7X00 (I think it's a 7600, but I forget now) that I'm thinking of testing this theory with.

Also, the author says that MESH barfs beyond a certain speed... seems like lifting up the +VCC pins on the MESH chip would give you a Mac with "failed" MESH, and would probably allow System 7 to boot... thoughts?

 

[Bolle]

if this works it would really kick ass... :O

I´d definetly do to my 9500. when i run across a free PCi Powermac i´ll try this.

 

[Anonymous Freak]

I don't know why you would care. Older Macs don't really have the speed to properly utilize anything faster than 32-bit, 33 MHz PCI anyway. (The main memory on the PCI PowerMacs is barely faster than 32-bit/33 MHz PCI.)

Besides, all 66 MHz PCI cards are backward compatible with 33 MHz PCI. (And, in fact, 133 MHz, 64-bit PCI-X cards are backward compatible with 33 MHz, 32-bit PCI, as well. So you could, in theory, slap Apple's PCI-X FibreChannel card into a PCI PowerMac, and it would work just fine.)

 

[TylerEss]

The reason I'd care is in that word "properly" in your first sentence: the system bus IS faster than PCI (and in ideal conditions, faster than 66MHz PCI, too) so if we can get a little more, we get a little more. The fact that it isn't as fast as proper 66MHz PCI would be doesn't diminish the fact that it (would be / is) faster than it was before.

For example, I've got a Voodoo2 SLI setup here; I note that the Voodoo2 chipset can utilize AGP1x as 66MHz PCI, so it's likely that my Voodoo2 cards will work on a 66MHz PCI bus. It might be fun to run them at that speed, if they'll take it, just to see what performance increase I can see, if any.

It's a fun thing, like nearly all vintage computing is. It's highly unlikely that the PCI overclock, if it worked, would somehow save an old 9500 from needing to be replaced, for example. It's just for the hack value, and perhaps the added speed would make the experience a little more enjoyable.

 

[alk]

I'm pretty sure that B&W benchmarks indicated that the real-world benefit of 66 MHz PCI slots over 33 MHz PCI slots is something like a 10% boost. Still, for hack value, this would be interesting. But it begs the questions: how would you verify that your PCI bus is operating at a full 66 MHz, and how would you test the throughput on the bus when your system bus can't saturate it?

Peace,

Drew

 

[Unknown_K]

Was the card design the limiting factor? I recently snagged a Rage 128 for my B&W since 10.3 supported it better then the A/V rage card that came with it, and it used that flipped 66Mhz slot leaving me the other 3 for cards.

 

[TylerEss]

I'm pretty sure that B&W benchmarks indicated that the real-world benefit of 66 MHz PCI slots over 33 MHz PCI slots is something like a 10% boost.

I remember seeing statistics like that, too; Latency acutually ends up having a lot to do with things like fill rate from RAM to framebuffer, not just raw bus speed.

Still, for hack value, this would be interesting. But it begs the questions: how would you verify that your PCI bus is operating at a full 66 MHz, and how would you test the throughput on the bus when your system bus can't saturate it?

An oscilloscope could verifty 66MHz operation. }

As for the throughput, you'd just have to benchmark bus-stressing things before and after the hack; even though the system bus is faster than either 33 or 66MHz PCI, you'll never saturate either one of them becuase the PCI controller sucks and the memory controller sucks. (Where by "sucks" I mean "not designed with absolute performance above all other considerations. ")

The hope is that if you can saturate 50% of the 33MHz bus, you might be able to saturate 30% of the 66MHz bus, which would be a total increase of 10% bandwidth, even though the utilization went down.

If that webpage is for real, it does improve game framerates and disk throughput a little bit, which is pretty cool for just adding a different oscillator.

 

[trag]

Well, it's true that the PCI bus on the PowerSurge machines is clocked in exactly the way he writes. I'm surprised that the PCI clock buffer, the PCI arbiter and especially Bandit will work at 66 MHz though. I'm especially surprised that Bandit will still synch up with the CPU bus if you're running the PCI side at 66 MHz, but I guess it isn't too surprising since Bandit must adjust to many possible CPU bus speeds as well. They would have had to use a synchronization method which is largely independent of specific clock rates.

Oh, and I guess Grand Central and probably CURIO would need to run at the increased speed as well.

 

[Bunsen]

Hmm. Wonder what it would be like with the PCI bus and the system bus running at the same frequency?

 

[Bolle]

taht would be cool but seems to be impossible... never got the system bus over 59.5 mhz before it wouldnt boot anymore on any mac i tried.

 

[TylerEss]

!! PowerComputing clone with 66MHz bus as a stock setting would do it. Anybody got one?

 

[trag]

!! PowerComputing clone with 66MHz bus as a stock setting would do it. Anybody got one?

I believe you are thinking of the PowerCenter Pro which shipped with a CPU card with a 60 MHz bus speed and 180, 210 or 240 MHz CPU speed (PPC604e).

Some of the other Catalyst based Power Computing models could also be clocked up to similar heights. I don't know if any of them would go all the way to 66 MHz.

On the one hand, all of the candidate machines are Catalyst based (7200 family) rather than being Power Surge machines (x500 family). On the other hand, the 7200 also uses the Bandit PCI bridge and the same arrangement of PCI oscillator and clock buffer, so it should work...

 

[Bunsen]

What about setting the PCI bus to 50 or 60MHz?

 

[coius]

btw, setting the PCI to 66Mhz only makes so much of an increase. Remember, it is tied to the south-side bus, which means the max speed of that chip, is the max data through-put of the PCI bus. Just because you move up the speed doesn't change the fact that you have a MAJOR bottle neck. You also are going to have problems with cards that are even designed for 66Mhz, because it's not going to truly give it the full data through-put it is requesting. Video cards are an example. The PCI bus is choking as it is, but increasing it is going to make it worse since it is going to involve errors because of timing issues. If you don't get the PCI synced up exactly like it should, it could choke the cards because it doesn't have Error correcting.

You may also notice components not working. or burning out quicker. Increasing the speed can also increase voltage, which like Overclocking a CPU, can damage it if it's forcing the chip to move faster than it can. there is a reason PCI was left at 33Mhz, and that was because the chips at that time couldn't handle any faster. once they upped the South-Side bus, they could put faster chips. but they had to use chips that COULD go faster in order to make it work.

That's my $0.02

 

[trag]

Ummm, the x500 (Power Surge) family of Macs do not have a bus which is called the "south side bus".

There is a memory/CPU bus. It is arbitrated by a chip called Hammerhead. It connects the CPU, the Hammerhead and the PCI bridges. Hammerhead provides the connection to the RAM.

 

[Quadraman]

It might be a good idea to pass up on 66mhz operation and just stick to 64mhz. It's still a substantial improvement without the incompatibilities and problems of 66mhz. For that extra 2mhz it isn't worth not being able to use most PCI cards and not being able to use the Radeon 9200 with another card on the bus is unacceptable.

 

[OtakuMegane]

Wow, I've been away wayyy to long. Forgive me bumping an older thread.

That site actually belongs to me and yes, the hack is quite real and does work. The particular system mentioned is still in use right now, though with a 1GHz Sonnet G4 card instead of the G3.

The practical applications for the 66MHz hack, however, are extremely limited, especially since there's maybe a dozen useful PCI cards that are both compatible with old PowerMacs and a 66MHz bus. Not to mention the fact Radeons seem to hate sharing at 66MHz. The old memory bus remains a serious bottleneck too, You get around 10%-20% improvement with simpler everyday stuff and a decent boost in GPU-heavy gaming but not a lot else. Turning on Quartz Extreme with the faster PCI does boost video playback considerably.

If your PowerMac is already getting too slow for your needs this won't fix very much in long-term. It was part of a personal challenge to see how far these old systems could be pushed. I considered the project officially "done" about half a year ago since there's not much left to do unless we get better CPU upgrades or PCI graphics cards (very unlikely in both cases).

 

[TylerEss]

Did you ever try the hack on a machine with dual PCI busses? Does it like having a 66MHz bus and a 33MHz bus?

 

[Quadraman]

And for all the effort you are going to go through hacking the bus up to 64-66mhz, you might as well spend a few bucks and just get a newer machine like a Sawtooth, which are cheap as heck right now.