Parhelia is 2002, but it wasn't a particularly fast card even then. PCI-X died off pretty immediately after PCI-e came out, at least on the graphics card front, so that might be pretty much up there with the fastest full-64-bit-width 66MHz PCI-X cards.

Have you tested with the 3450 PCI? It's pretty up there as far as PCI video cards go, and as long as it runs what you want it to run it might not matter whether it's bottlenecked or not. I also think I remember something about some PCI graphics cards using bridge chips that support 66MHz PCI, so if it uses one of those it might not be as bottlenecked on a PCI-X slot as it is on a normal 33MHz PCI slot, even if it still is only 32 bits wide. It'd also be worth testing since a while back I had an Opteron server that had a BIOS that locked down dedicated graphics because the manufacturer (Dell) presumably didn't want to step on the toes of their own workstation hardware. It'd be wise to check whether dedicated cards work at all before going for a faster one.

I doubt you'll be able to find anything like a high-end GF7 without using one of those fancy PCI-X to PCI-E bridges you're talking about.

Ill have to check out what my PCI-X GPUs are, but I have a feeling they are just Sun XVR-600s (3D Labs GPU) with a proprietary vBIOS. In any case that should give you an idea that the Radeon is better.

Re: M.2 NVMe working on old school regular pci

You may find this interesting the bridge appears to be even slower than the pci bus

It's very likely that the system PCI bus itself is the bandwidth bottleneck on that adapter and the amount less than full PCI bandwidth that it got was a result of overhead or other devices on the PCI bus.

The bridge chip on that adapter supports 66MHz PCI. So, it could plausibly get double the speed in a suitably equipped PCI-X system. Unfortunately it looks like it doesn't support 64 bit PCI-x, only 32bit.

Edit: the bridge chip I see in use on that adapter (PEX8111) is actually very similar to the chip that appears to be used on the HD3450 PCI (PEX8112), and they both appear to support up to 66MHz. So technically I guess a PCI 3450 is pretty much exactly a PCI to PCI-e adapter + 3450 card.

Error 0x7CF wrote on 2021-02-02, 02:33:

It's very likely that the system PCI bus itself is the bandwidth bottleneck on that adapter and the amount less than full PCI bandwidth that it got was a result of overhead or other devices on the PCI bus.

The bridge chip on that adapter supports 66MHz PCI. So, it could plausibly get double the speed in a suitably equipped PCI-X system. Unfortunately it looks like it doesn't support 64 bit PCI-x, only 32bit.

Edit: the bridge chip I see in use on that adapter (PEX8111) is actually very similar to the chip that appears to be used on the HD3450 PCI (PEX8112), and they both appear to support up to 66MHz. So technically I guess a PCI 3450 is pretty much exactly a PCI to PCI-e adapter + 3450 card.

that was the only pci card installed at the time. this was on a core 2 quad

I also have a pci 610gt it can run furmark at 5 fps which ain't actually too bad.. haven't run any more benchmarks with it but be warned it's picky about which motherboards it works in It seems to like middle era pci boards the best so a opteron board may work

Error 0x7CF wrote on 2021-02-02, 00:18:

Parhelia is 2002, but it wasn't a particularly fast card even then. PCI-X died off pretty immediately after PCI-e came out, at least on the graphics card front, so that might be pretty much up there with the fastest full-64-bit-width 66MHz PCI-X cards.

Have you tested with the 3450 PCI? It's pretty up there as far as PCI video cards go, and as long as it runs what you want it to run it might not matter whether it's bottlenecked or not. I also think I remember something about some PCI graphics cards using bridge chips that support 66MHz PCI, so if it uses one of those it might not be as bottlenecked on a PCI-X slot as it is on a normal 33MHz PCI slot, even if it still is only 32 bits wide. It'd also be worth testing since a while back I had an Opteron server that had a BIOS that locked down dedicated graphics because the manufacturer (Dell) presumably didn't want to step on the toes of their own workstation hardware. It'd be wise to check whether dedicated cards work at all before going for a faster one.

I doubt you'll be able to find anything like a high-end GF7 without using one of those fancy PCI-X to PCI-E bridges you're talking about.

I haven't tested the card yet, as I'm still waiting on the Tyan board to arrive, but here's the video from Pixel Pipes where he benchmarks it on a C2D versus some other PCI cards. https://www.youtube.com/watch?v=Al_T6bSoclM
And watching it again, it seems I was mistaken that PCI was holding the cards back, and it's moreso the downgraded specs that they had as a result of being such budget/niche offerings as PCI cards.

border collie21 wrote on 2021-02-02, 05:15:

Error 0x7CF wrote on 2021-02-02, 00:18:

Parhelia is 2002, but it wasn't a particularly fast card even then. PCI-X died off pretty immediately after PCI-e came out, at least on the graphics card front, so that might be pretty much up there with the fastest full-64-bit-width 66MHz PCI-X cards.

Have you tested with the 3450 PCI? It's pretty up there as far as PCI video cards go, and as long as it runs what you want it to run it might not matter whether it's bottlenecked or not. I also think I remember something about some PCI graphics cards using bridge chips that support 66MHz PCI, so if it uses one of those it might not be as bottlenecked on a PCI-X slot as it is on a normal 33MHz PCI slot, even if it still is only 32 bits wide. It'd also be worth testing since a while back I had an Opteron server that had a BIOS that locked down dedicated graphics because the manufacturer (Dell) presumably didn't want to step on the toes of their own workstation hardware. It'd be wise to check whether dedicated cards work at all before going for a faster one.

I doubt you'll be able to find anything like a high-end GF7 without using one of those fancy PCI-X to PCI-E bridges you're talking about.

I haven't tested the card yet, as I'm still waiting on the Tyan board to arrive, but here's the video from Pixel Pipes where he benchmarks it on a C2D versus some other PCI cards. https://www.youtube.com/watch?v=Al_T6bSoclM
And watching it again, it seems I was mistaken that PCI was holding the cards back, and it's moreso the downgraded specs that they had as a result of being such budget/niche offerings as PCI cards.

yeah it's surprising that the pci actually isn't holding them back as much as you would think the better chips still seem to do better performance nearly a decade after pci was left by the wayside for graphics cards if memory serves the best one is a tossup between the 610 and the 9500 but you may want to look at native agp cards they seem to be more compatible with older motherboards though I think a opteron board may be likely to work with many of them

btw. in his vid the 610 is the 520 he says they are basically the same card

Native AGP adapted to PCI?

border collie21 wrote on 2021-02-02, 06:27:

Native AGP adapted to PCI?

yeah it seems they used agp gpus without bridge chips on pci which makes sense agp is just pci with faster clocks and more lanes? or something like that where as pci express is a whole other animal and requires a bridge

of the cards I have the ones that aren't bridged seem to work in the most motherboards but it's hit and miss. sometimes a card won't boot but it will work fine in windows. up to you to decide if that is acceptable but you'll just have to buy one and try it to know for sure the one that I have that seems to work in everything is the 6200 geforce

So I guess I'd be looking for any pre-PCIE cards that have a PCI variant then.

Thanks for the help!

border collie21 wrote on 2021-02-02, 06:54:

So I guess I'd be looking for any pre-PCIE cards that have a PCI variant then.

Thanks for the help!

your chances will improve I'd say but I do expect that there are pcie cards that may work. it's annoyingly hit or miss.

Sphere478 wrote on 2021-02-02, 06:46:

agp is just pci with faster clocks and more lanes? or something like that where as pci express is a whole other animal and requires a bridge

PCI 2.1 was the foundation on which AGP was originally designed. AGP 1.0 in 1x mode is in principle a 32bit 66MHz PCI bus with some extra features that is dedicated to a single device. It uses the same voltages and SDR transfers. There's no lanes like in PCIe. This is why it's relatively easy to adapt an AGP 1.0 card to the PCI bus. In theory you only need signal voltage conversion for 33MHz 5V PCI plus ideally termination for some AGP specific signals and extra supply voltage since PCI does not necessarily deliver the same currents AGP needs.
Starting with AGP 1.0 in 2x mode you get DDR transfers which are not possible on the PCI bus. That's the biggest bottleneck for most AGP chips running on PCI cards, they're limited to AGP 1x speeds. (Half of that on 33MHz slots)
PCIe is fully software compatible to PCI but uses serial transfers (among other electrical differences), so it requires a bridge in any case.

This is the extent to which I'm confident talking about similarities and differences between PCI, AGP and PCIe. 😅