I was thinking about my S2720-533, which has PCI-X 133MHz ability IIRC. Though cards themselves may not throw data any faster, and the motherboard may adjust to run only at 32-bit/33MHz, might the board actually accomplish what it needs to, more quickly due to its ability to handle much quicker cards?
I thought this may also be important when eeking out V2 SLI setups.
Voodoo2 isn't PCI-X. PCI-X is physically different connector (you cannot "get" PCI-X via PCI connectors), and they can rival AGP and PCIe in terms of bandwidth (~4GB/s is max configuration). However graphics cards that support PCI-X are relatively rare - afaik the most powerful option is based on a Parhelia-family GPU, and they are not cheap. The slots should be backwards compatible with PCI cards, but the PCI cards will derive no benefit from that (since they're only PCI cards).
Remember that only 3.3v compatible PCI cards will fit in PCI-X slots. The bus itself didn't see much use outside of servers, so the cards that were released were mostly RAID and multiport Gigabit Ethernet cards.
I think some boards have 'primary' and 'secondary' PCI buses, and the PCI-X slots are usually on the primary. This can mean they have higher performance/reliability. I think it's called 'PCI-to-PCI bridging'.
Account retired. Now posting as Errius.
No go, by the look of it - the Voodoo 2's appear to be 5V PCI , and the motherboard has only a single legacy 5V keyed slot
The PCI-X slots are keyed for 3.3V
Voodoo 3 is dual keyed - and with the slots in 3 groups (2x up to 133MHz PCI-X, 2x up to 100MHz, 1x standard PCI), it's possible that they may not all be sharing throughput - which would help - A V3 or later PCI in one slot group, Sound card in a different slot group.
I suppose a system designed to support faster PCI might have less latency per individual transaction, but I don't think that would end up making any difference. I think the lower clock rate would still end up dictating the throughput over the long term. Also, keep in mind that Voodoo3s didn't perform any better in AGP slots than they did in 33MHz PCI slots (from what I've read), so it appears these cards don't need any more speed.
2 cards combined is a different story though, so there could be a benefit to having the cards on separate buses. PCI-X boards normally will have multiple buses available.
With newer stuff, PCI-X slots can be faster even with 32-bit cards if the card supports running at 66MHz, as many newer PCI cards do. But Voodoo2's are pretty old for this, and if they're 5V keyed then it's a non-starter.
There are some 440GX server boards (Intel that I'm aware of) that use the chipset's AGP feature to provide a second PCI bus running 5V 66MHz, but this violates the PCI spec so it isn't common. Normally 66MHz+ requires 3.3v signal voltage so the slots are keyed accordingly.
I've read comments about newer 32-bit PCI video cards being able to run at 66MHz, but I've never seen any official/advertised notation of this being supported. I'm not sure if the performance effect has been measured, or if it's just an assumption that they were running at the faster speed. The PCI bus is supposed to throttle to match the slowest device that is connected, but this just depends whether the card signals a desire for a particular speed.
With respect to newer video cards, it's a weird situation now that AGP has been unsupported for years, but newer PCI cards exist. I can imagine it being possible that a PCI-X server board with a PCI 9500GT, GT430, or GT610 might end up being more capable than many AGP based systems. Back when they were current, a server board without an AGP slot was useless for games.
It's a shame that higher speed PCI, even if only 32-bit, was never adopted on consumer motherboards. The PCI bottleneck is definitely something you have to watch for when building a system for I/O performance. Some later chipsets have disk controllers and ethernet controllers linked directly to the chipset without using PCI, and that's helpful if lots of data will be going through those interfaces. Replacing onboard features with a PCI card can actually slow those systems down in I/O intensive tasks.
I apologize for not being more clear. I know PCI and PCI-X are different. I thought that, the 5V keying aside, the controller itself may be more capable of getting things done, faster. However, I could be way off-base, and the board may use a more antiquated PCI controller, seperate from the big daddy PCI-X controller--i haven't checked.
I am however, curious if there are seperate PCI buses available on said boards, and if V2-SLI setups are compatible with such an arrangement.
Its been done years ago check it out.
On a far away planet reading your posts in the year 10,191.
wrote:Its been done years ago check it out.
Hang on, let me get my T221 out so I can view the images... 🙄 🤣
Also not quite the same as what Logistics is asking, and to answer that:
- No, there should not be any "big daddy" difference with a PCI-X equipped board versus a PCI board - having less devices saturating the PCI bus, however, would be important. And in a board with PCI-X and PCI you may have two PCI controllers which may make this easier to accomplish, as 2fort5r and shamino stated.
- It would depend on the board, however, what the controller arrangement looks like. What one board OEM does can't be assumed to be standard across any board with PCI-X, unfortunately.
wrote:Hang on, let me get my T221 out so I can view the images... :roll: :lol: […]
wrote:Its been done years ago check it out.
Hang on, let me get my T221 out so I can view the images... 🙄 🤣
Also not quite the same as what Logistics is asking, and to answer that:
- No, there should not be any "big daddy" difference with a PCI-X equipped board versus a PCI board - having less devices saturating the PCI bus, however, would be important. And in a board with PCI-X and PCI you may have two PCI controllers which may make this easier to accomplish, as 2fort5r and shamino stated.
- It would depend on the board, however, what the controller arrangement looks like. What one board OEM does can't be assumed to be standard across any board with PCI-X, unfortunately.
No kidding... even with my crazy T60p LCD I can barely see the full images.
The T221 is a bit annoying in a few ways... I would argue it's still 'too ahead' of its time to be useful yet.
(the IBM computer you see there takes it time counting all 32GB of memory... no way to bypass it. Suppose I could see about modding the BIOS...)
With that said, the T221 does a mean 640x480... no Gaussian smudgy blurs.
I remember reading somewhere (probably a server manual) that explained all this, and recommended that high performance cards (video/RAID/etc) should be mounted on the PCI-X slots, even if they were regular 32-bit PCI.
Account retired. Now posting as Errius.
