Blue and White G3 66MHz PCI Slot experimentation

[mectojic]

The Blue and White G3 has a unique PCI slot which no other Power Mac has – a 66MHz (i.e. dual-speed) PCI slot. It was designed for the purpose of graphics acceleration, and could be considered to be a sort of AGP 1x slot; however, AGP is much more powerful a slot, since it runs independent to the PCI lanes, while the 66 slot on the G3 still shares the PCI bus.
The ATI Rage Pro 16mb card that came with the G3 utilised the 66MHz acceleration, as did some other 3rd party cards – but many cards did not support it, making it effectively a 33MHz slot in most cases. Besides, the speed increases gained in the graphics department were never going to be that spectacular, when still stuck on PCI anyway. So it was a cool idea, but not particularly utilised.

However, there is one advantage that the 66MHz slot has over the G4's AGP slots, and that is compatibility with the PCI interface. There were (and still are) other cards that support the 66MHz PCI spec, and therefore could theoretically be used at the faster speed in the B&W G3. Although the G3's 66 slot was intended for graphics, you can use a graphics card in any slot, leaving the special slot free.

One example I've been trying out, inspired by a thread on OS9Lives, is to put a supported, flashed SATA PCI card into the 66MHz slot. For reference, the card I used was this one: https://www.local338shop.com/produc...card-support-os-9?_pos=2&_sid=0a35b1f15&_ss=r

I was hoping it would "just work". But that's not quite what happened. Here are the results.

First I confirmed that the Sata card worked in a 33MHz slot. It was able to boot OS 9.2.2 and 10.4.11 just fine. Despite the seller claiming it works with older OS X, I couldn't get it to boot 10.2.8 or properly install from 10.2 / 10.3 install CDs, which was a shame, but not essential to these tests.

Then, the switch: putting the Sata card into the 66, and my Radeon 9200LE into the 33. Sadly, nothing would boot – neither CDs nor any pre-installed OS. I wonder if there are certain limitations on the 66MHz slot, or in some part of hardware in the G3 that prevents this.

Finally, I did some troubleshooting, booting from ATA with an SSD to Sata card in the 66 slot. At first, the Sata drive was not detected, although the PCI card was detected in System Profiler. Then I unplugged and replugged the Sata cable, and suddenly, voila – the drive showed up. I tested this a few times, and it was same each time – under OS X, the G3 would only detect a drive in the 66 slot if I replugged the cable. While testing some of this, the Mac did occasionally crash, though I couldn't establish why.

However, I had my proof at least – the 66Mhz slot does work with non-graphics cards – but does it work at a faster speed?
I did some speed tests. I'm not a benchmark guy, so this is just a basic reference and proof of concept:

Test: Duplicate 250mb of files using the PCI Sata SSD –
In the 33MHz slot: 11.5 seconds
In the 66MHz slot: 8.5 seconds

Test: Duplicate 1GB of files using the PCI Sata SSD –
In the 33MHz slot: 47.5 seconds
In the 66MHz slot: 30.5 seconds.

Test: Copy 1GB of files from the PCI Sata SSD to the ATA SSD (IDE-SATA adapter):
In the 33MHz slot: 40 seconds
In the 66MHz slot: 34.5 seconds

As can be seen, there is about a 30% faster write speed, which is awesome.

Now, if it weren't for the need to reconnect the sata port each time, and the occasional crashes, I think this would be a really neat way to utilise the B&W – in fact, this ability to do faster read/write than G4 Macs actually means that the B&W G3 does have one advantage over them. Indeed, this 66MHz slot method must be the fastest possible r/w speeds possible on a G3, and would work very well in a server setup. Perhaps you could go all out, using a 4-slot card attached to eSata and run 8 or 16TB in raid – I don't know. Sounds fun though

I've heard there are other PCI Sata cards (Seritek red cards; Sonnet Tempo?) – I would ask if other B&W owners could please try some of these experiments too, and bring back results.

In summary with my card:
√ 33MHz slot: can boot OS 9.2.2 and 10.4, no drivers. Can't boot 10.2 / 10.3 at all.
√ 33MHz slot: is stable and can be used for secondary drives just fine (as expected).
X 66MHz slot: can't boot any OS.
X 66MHz slot: causes 10.2 / 10.3 to crash, even when booting the OS from ATA.
√ 66MHz slot: can be used for secondary drives in OS 9.2.2 and 10.4. Drives detected in 10.4 only after replugging the sata cables. Possibly unstable in OS X.

 

[joevt]

The 33 MHz slots are 64 bit, so they should have similar performance as the 32 bit 66MHz slot (for memory transactions - configuration and I/O space accesses are limited to 32 bit). Does anyone make a 64 bit 66 MHz bridge chip that can allow communicating between 33MHz 64 bit host and 66MHz 32 bit peripheral?

 

[ervus]

Interesting, thanks for sharing your results. There are variety of results in this thread also:

forums.macrumors.com/threads/post-your-ssd-cf-sata-pata-powerpc-benchmark-results.2063361/

I wonder if one of the 64-bit slots could be run at 66MHz? Is there a separate clock chip for them? The sg500 chip in the Cube has a 33MHz and 66MHz output pin.

 

[joevt]

The 33 MHz slots are 64 bit, so they should have similar performance as the 32 bit 66MHz slot (for memory transactions - configuration and I/O space accesses are limited to 32 bit). Does anyone make a 64 bit 66 MHz bridge chip that can allow communicating between 33MHz 64 bit host and 66MHz 32 bit peripheral?

Or what about a 64 bit PCI to PCIe bridge chip? Preferably something that can do PCIe gen 2 or gen 1 x2 since 64bit 33MHz is 266 MB/s and PCIe gen 1 is limited to 250 MB/s.
Then you need a PCIe to PCI bridge chip that can do 66MHz on top of that...

Test: Duplicate 250mb of files using the PCI Sata SSD –
In the 33MHz slot: 11.5 seconds
In the 66MHz slot: 8.5 seconds

Test: Duplicate 1GB of files using the PCI Sata SSD –
In the 33MHz slot: 47.5 seconds
In the 66MHz slot: 30.5 seconds.

Test: Copy 1GB of files from the PCI Sata SSD to the ATA SSD (IDE-SATA adapter):
In the 33MHz slot: 40 seconds
In the 66MHz slot: 34.5 seconds

Assuming MiB and GiB, those numbers seem kind of slow if PCI 32bit 33MHz is supposed to do 133MB/s.

22.795 MB/s
30.840 MB/s

22.605 MB/s
35.204 MB/s

26.843 MB/s
31.122 MB/s

but duplicating files means reading and writing from the same device? In that case I suppose those numbers should be doubled since PCI is half duplex (can't read and write at the same time like PCIe can). SATA is also half duplex, so you would need two SATA drives to get full duplex in PCIe.


Possible single direction speeds (from doubling the above combined read/write speeds):

45.590 MB/s
61.680 MB/s

45.210 MB/s
70.408 MB/s

53.686 MB/s
62.244 MB/s

Those still seem low considering you are using a 32 bit 66 MHz slot which should allow 266 MB/s.
Are the SATA SSDs SATA I speed = 150 MB/s?

How many files were you testing? I would just test one large file.
To test max single direction speed, I would use a benchmark that writes the same 512 bytes from RAM to multiple blocks.

 

[Amethyst1]

I'm not a benchmark guy […]

I’d still be interested in some QuickBench results, if you were willing to give that a shot.

 

[Intell]

The Yikes! PowerMac G4 (PowerMac1,2) also has this 66Mhz PCI slot. It's basically a blue and white G3 (PowerMac1,1), but in a different case and a G4.

 

[LightBulbFun]

The Yikes! PowerMac G4 (PowerMac1,2) also has this 66Mhz PCI slot. It's basically a blue and white G3 (PowerMac1,1), but in a different case and a G4.

the Xserve G4 also has 66Mhz PCI slots unique in that its both 66Mhz and 64 bit

although technically not a PowerMac LOL

 

[eyoungren]

the Xserve G4 also has 66Mhz PCI slots unique in that its both 66Mhz and 64 bit

although technically not a PowerMac LOL

PowerServe.

 

[Amethyst1]

PowerServe.

RackMac FTW!

 

[DearthnVader]

The Blue and White G3 has a unique PCI slot which no other Power Mac has – a 66MHz (i.e. dual-speed) PCI slot. It was designed for the purpose of graphics acceleration, and could be considered to be a sort of AGP 1x slot; however, AGP is much more powerful a slot, since it runs independent to the PCI lanes, while the 66 slot on the G3 still shares the PCI bus.
The ATI Rage Pro 16mb card that came with the G3 utilised the 66MHz acceleration, as did some other 3rd party cards – but many cards did not support it, making it effectively a 33MHz slot in most cases. Besides, the speed increases gained in the graphics department were never going to be that spectacular, when still stuck on PCI anyway. So it was a cool idea, but not particularly utilised.

Most PCI graphics cards set a bit in the PCI registers to tell if they are 66mHz compatible. It can be checked with ReggieSE from Apple's C.H.U.D. tools.

There is an OS 9 app that can be used to check the throughput to the graphics card, to see if throughput exceeds 133MB/s.

Macintosh - ThroughPut

 

[mectojic]

I’d still be interested in some QuickBench results, if you were willing to give that a shot.

The post that inspired mine includes some results. It's French so my link is a Google Translation:

SATA et vieux Power Mac : comment obtenir les meilleurs performances - Le journal du lapin

Le journal du lapin, mon espace personnel ou je parle de trucs retro, d'expérimentations et de produits Apple. Et évidemment de prototypes.

www-journaldulapin-com.translate.goog

 

[joevt]

Or what about a 64 bit PCI to PCIe bridge chip? Preferably something that can do PCIe gen 2 or gen 1 x2 since 64bit 33MHz is 266 MB/s and PCIe gen 1 is limited to 250 MB/s.
Then you need a PCIe to PCI bridge chip that can do 66MHz on top of that...

#1 talks about PCIe/PCI/PCI-X adapters. There exist 64bit 133MHz PCI-X to/from 2.5 GT/s x4 PCIe bridge chips. The StarTech PCIX1PEX4 used one of those bridge chips but I think it was limited to 33MHz and 133MHz - not sure if it could do 66MHz. StarTech doesn't sell them anymore and I couldn't find an alternative.