386SX wrote on 2020-09-21, 16:24:

Hi, […]
Show full quote

Hi,

I'm trying a Geforce 210 512mb ddr3 into an old but modern dual core 1.9Ghz/ddr3 industrial mini-itx board. The board has only one single PCI old style connector and I thought to try pushing its limits with this card being theorically faster than the on-board one or the FX5200 PCI I already had.
Win 8.1 and the latest 32bit drivers. Some old games like Doom3 clearly fly compared to the FX 5200 PCI I tried but also to the on board GMA SGX545 gpu. Other games and bench seems to be just as slow or only a bit better. I've checked the cpu usage and mostly in games is @ 50% so I suppose it's a bus problem. Are there any way to optimize the PCI connection like some config to force to speed up the bandwidth, tweaks or whatever? In the bios it says it's using the 133MB/s mode. Generally the GPU is surely faster in the web browser, just as ok hw decoding videos compared to the SGX one but I find it to be overally disappointing.. I know it wasn't a fast gpu but still ..maybe is the bridged PCI-PCIEX connection that make it even slower who knows..
Thanks

IMHO, the issue is one of expectations . You will not get decent performance from an extremely bandwidth limited legacy PCI slot .

Legacy PCI is limited to 133MB/second . Even PCI Express 1.0 x1 is 250 MB/s and the x16 slots are 4GB/second . If you have a PCI Express x16 slot, get a card that works in that . If you only have PCI Express x4 or x1 slots, you would likely still be able to get it to work with a normal PCI Express x16 video card by using a PCI Express extension/riser . Something like this https://www.wish.com/product/5d35640a011f4723 … =true&share=web or this https://www.newegg.ca/startech-com-model-pex1 … N82E16815158223 EDIT : Another way is to file down the PCI Express connector on the PCI Express x16 card to fit . This is obviously destructively irreversible, though .

Modern PCI Devices run 66Mhz, 266Mbytes/s if you don`t have a old Device with 33Mhz inside at same Time.

You really, really don't want to be using a PCI GPU for any OS newer than XP. The display composition routines that newer OSes employ was never intended for such a slow bus.

When I tested my GT520 PCI in a C2D E8600 machine, I couldn't even get YouTube playing smoothly at 360p. Even the lame onboard GMA X4500HD was fully capable of smooth 1080p/60 and 1440p/30. Adding a PCIe GTX 1650 even allowed for 2160p/60 playback with very low CPU usage.

Video playback wasn't the only drawback of the PCI card. Chrome and Firefox both took noticeably longer to draw up web pages with the PCI card, and once the sites were loaded, scrolling was extremely choppy. The IGP and PCIe GPUs were tied in terms of page render times, but the PCI-E GTX 1650 was definitely the best for scrolling smoothness. All tested on Win10 build 1909 by the way.

matze79 wrote on 2020-09-21, 17:50:

Modern PCI Devices run 66Mhz, 266Mbytes/s if you don`t have a old Device with 33Mhz inside at same Time.

"Modern" parallel (legacy) PCI devices do not really exist anymore. If you are referring to the last native parallel PCI cards that were produced, some of them, like disk controllers, were indeed 66Mhz capable . A natively PCI Express GPU using a PCI Express to PCI bridge could support 66MHz operation, if the manufacturer chose a 66MHz capable bridge chip, which is not a given (EDIT: I have no idea about the GT210 PCI reference design, if there is one) .

Also, AFAIK, most legacy PCI bus implementations on motherboards are of the 33MHz only variety, especially those implemented through a bridge chip. Some (EDIT: at some point, many) workstations, servers and industrial boards had 66MHz capable slots when these were implemented natively by the motherboard southbridge .

Standard Def Steve wrote on 2020-09-21, 18:21:

You really, really don't want to be using a PCI GPU for any OS newer than XP. The display composition routines that newer OSes employ was never intended for such a slow bus.

When I tested my GT520 PCI in a C2D E8600 machine, I couldn't even get YouTube playing smoothly at 360p. Even the lame onboard GMA X4500HD was fully capable of smooth 1080p/60 and 1440p/30. Adding a PCIe GTX 1650 even allowed for 2160p/60 playback with very low CPU usage.

Video playback wasn't the only drawback of the PCI card. Chrome and Firefox both took noticeably longer to draw up web pages with the PCI card, and once the sites were loaded, scrolling was extremely choppy. The IGP and PCIe GPUs were tied in terms of page render times, but the PCI-E GTX 1650 was definitely the best for scrolling smoothness. All tested on Win10 build 1909 by the way.

I quote this message to write some UPDATES:
You might be surprised that the video decoding was the "best" thing that worked in Win with the Geforce 210 PCI tested with 1080p 60fps h264 videos but also youtube vc1 codec 720p-60fps both around 40% of an old dual core Atom cpu..... so I suppose the PCI isn't really a problem for video decoding. 😀
I'd say it runs even faster than the AGP (or) PCI-E internal SGX545 decoding engine that makes a great job with decoding quality but imho with a bit higher cpu usage (even if definetely fully hw accelerated).

The expectation problem sure is true but still something seems "strange" how the card performed into Win 8.1 32bit, so I tried installing on another disk a Linux/LXDE with a 5.4 x64 kernel and latest suggested 340.108 proprietary nvidia drivers. To my surprise even using Wine as Directx to Opengl wrapper for old games and bench I was shocked to see the same games and bench that seems to not run well, even using what I suppose has an API wrapper and not natively, runs like two/three times as fast in Linux. 3DMark03 in Win 8.1 gave 3600 points and in Linux with Wine gives 5600 points and the frame rate is real, 3DMark05 I didn't finish the test but as soon as the first test start was clear as the sun the frame rate was like using a whole different gpu generation when instead is the very same system without any differences but the o.s.
I mean, in linux with a sw like Wine I'd expect game running slower than a native Win oriented config and not faster with same perfect rendering quality. No antialiasing or vsync problems.. in win some bench/scenes gives me a single digit fps result..

I suspect there may be a PCI bandwidth problem but might not be the most important limit here; I suppose the whole subsystem (not the vga that has Win8.1 drivers, probably more the chipset cpu<->pci drivers who knows) wasn't just designed to work in Win 8.1 with PCI video cards. Also along with how the GUI of these modern o.s. works, might explain the bad results. Cause why an Atom cpu with the same PCI 133MB/s (bios report that speed) bus and even with a not supported Linux installation, not native Wine software with Directx calls sent to Opengl, I get twice the speed but imho even higher in some specific bench/games points compared to the same system without any conflicts and all proprietary native drivers into a Win installation? The only thing I might think of, could it be that latest Win o.s. might simply not be optimized for old PCI bus?
The PCI bus speed itself might be a problem if compared to PCI-E but here the problem imho is the combination of the system with that modern o.s. This might explain why any PCI cards I tried couldn't get as fast even as supposed to even when they could. The system worked I mean.. no problems or conflicts. In Win I even got a perfect (!) DXVA2 acceleration with both the SGX545 and the Geforce 210 even works into the firefox web browser video acceleration but the 3D part seems like just can't work as supposed to no matter what.

Now I might try to see if using the x64 installation with even newer Nvidia drivers might solve something but I suppose it will not.

Interesting .

Out of curiosity, what motherboard are you using and do you know what model PCI bridge chips your video card and motherboard (assuming it is using one) are using ?

I am wondering if perhaps your PCIEX, PCIEX-PCI, PCI-PCIEX, PCIEX "sandwich" might be PCI 66Mhz capable, but your BIOS and Windows are initializing at PCI 33MHz, but Linux is actually running it at 66MHz .

darry wrote on 2020-09-22, 09:30:

Interesting .

Out of curiosity, what motherboard are you using and do you know what model PCI bridge chips your video card and motherboard (assuming it is using one) are using ?

I am wondering if perhaps your PCIEX, PCIEX-PCI, PCI-PCIEX, PCIEX "sandwich" might be PCI 66Mhz capable, but your BIOS and Windows are initializing at PCI 33MHz, but Linux is actually running it at 66MHz .

The mainboard is a generic layout Intel NM10/ICH7/Atom D2x00 design that was basically an old netbook oriented config into an industrial mini-itx board. These netbook hw had a sort of complicated support story in the gaming community but time after time I begin to sort of understand why only Win7 Starter 32bit o.s. was the only target designed official o.s at first for the internal iGPU that was known in the community for quite some win/linux problems but I sort of pushed these limits using unfortunately the only modern o.s. I got Win 8.1 was in facts unsupported even if it works in the 32bit version.

The vga bridge chip should be a PLX PEX8112 x1 Lane PCI-EX to PCI chip and the mainboard also report reading another PCI bridge the 'Intel 82801 Mobile PCI Bridge (rev e2) (Subtractive decode)' device.
This in linux but I didn't check them into win system devices (that might be interesting) even if also in Linux I don't see any specific kernel modules loaded for it nor I see drivers for this type of bridges around. I suppose they just works behind the o.s. with no necessary drivers or something similar beside the o.s. installed ones.

Anyway after long and stressing tests I have these thoughts about this system...IMHO
Win 8.1 (32bit) unsupported works with this specific system with the FX 5200 (bridged) PCI vga or Geforce 210 (bridged) PCI card (also the SGX545 -agp to pciex1 bridged?- internal gpu) BUT I suppose modern win o.s. with this design combination and drivers might not be fully optimized for PCI bridged fast speed devices like a 3D vga and not even for the multiple bridge chips logics. I'd not be surprised that also the whole WDDM 1.x driver model and/or how these modern o.s. works might run the system compatible in a slower mode for many 3D games probably suffering who knows from the GUI working in a different ways (?), maybe 3D games can't isolate themself from a background aereo GUI, just a user opinion obviously and mostly for the FX5200 and SGX545 theorically unsupported cards but not from the Geforce 210 PCI having its own modern drivers.
But after that, positive was that the GUI still work smooth and 3D games "runs" and is interesting that both internal and PCI DXVA2 Win native acceleration impressively off-loads such old low end cpu from that task. For the DXVA2 I might prefer the SGX545 that feels a bit better on the quality side but the Geforce 210 PCI seems to get lower cpu usage and it even decode a 720p/60fps web youtube stream @40/50% cpu that before had difficulty @ 90% even in 480p/30fps before without it.
But the linux test is quite an interesting one seeing that it's not the PCI bus here the limit. Maybe more the PCI to PCI-EX bridge to PCI Intel bridge to PCI-EX1 mainboard native connection plus these modern heavy o.s. at the top of these wonderful combination.... maybe ... the problem.. 😁 Still is unclear why in linux this combination of old/new techs is not a limit and games run faster so the o.s. design might have something to do with it and your point on the 66Mhz PCI initialization might be interesting to check! The specifications of the PLX bridge chip says PCI 3.0 up to 66Mhz.

Update:

3dmark05 in Win 8.1: around 1000 points
3dmark05 in Linux with Wine: around 2200 points

I checked and in the system win devices there's a MS default drivers for the Intel 82801 PCI bridge but I don't see any drivers or sign of the PLX vga bridge. Could it be maybe I'd need some official drivers for the Intel one? I'm looking for it before trying installing from zero my own x64 installation disc but I suppose it'd not be better I don't know if this chipset had x64 drivers..

Also I found in the win system devices the PLX connection with a generic "Adapter PCI standard from PCI to PCI" but the VEN_10B5&DEV_8112 line suggest me is the PLX 8112 bridge chip. I can't find Win 8.1 drivers for it. Might it be that the generic official nvidia drivers can work with it but doesn't install the specific driver and that might be the problem?

I would be curious about seeing results in Windows XP and the officially supported Windows 7 starter edition .

The ICH7 82801 variant is limited to 33MHz legacy PCI operation according to the datasheet I've seen.

darry wrote on 2020-09-22, 14:00:

I would be curious about seeing results in Windows XP and the officially supported Windows 7 starter edition .

The ICH7 82801 variant is limited to 33MHz legacy PCI operation according to the datasheet I've seen.

Interesting but how linux would start it to 66Mhz then? In the bios the only thing that says me is the speed is set (and that value is fixed) @ 133MB/s and the only other PCI related option is the latency timer that is set to 32 (but could go up to 264 I think).
Unfortunately I don't have Windows XP to try nor Starter but I hope to buy a retail boxed version soon. Even if XP was probably not even intended for this system cause too old.. the iGPU for example only had Win7 drivers but I suppose the subsystem might have XP drivers.

386SX wrote on 2020-09-22, 17:25:

darry wrote on 2020-09-22, 14:00:

I would be curious about seeing results in Windows XP and the officially supported Windows 7 starter edition .

The ICH7 82801 variant is limited to 33MHz legacy PCI operation according to the datasheet I've seen.

Interesting but how linux would start it to 66Mhz then? In the bios the only thing that says me is the speed is set (and that value is fixed) @ 133MB/s and the only other PCI related option is the latency timer that is set to 32 (but could go up to 264 I think).
Unfortunately I don't have Windows XP to try nor Starter but I hope to buy a retail boxed version soon. Even if XP was probably not even intended for this system cause too old.. the iGPU for example only had Win7 drivers but I suppose the subsystem might have XP drivers.

Sorry, I realize I was unclear . What I meant to say is that, having seen the datasheet, I doubt that you are getting 66MHz PCI under Linux, as the motherboard tops off at 33MHz PCI . The performance difference between Windows 8.1 and Linux is likely due to something else .

darry wrote on 2020-09-22, 18:11:

386SX wrote on 2020-09-22, 17:25:

darry wrote on 2020-09-22, 14:00:

I would be curious about seeing results in Windows XP and the officially supported Windows 7 starter edition .

The ICH7 82801 variant is limited to 33MHz legacy PCI operation according to the datasheet I've seen.

Interesting but how linux would start it to 66Mhz then? In the bios the only thing that says me is the speed is set (and that value is fixed) @ 133MB/s and the only other PCI related option is the latency timer that is set to 32 (but could go up to 264 I think).
Unfortunately I don't have Windows XP to try nor Starter but I hope to buy a retail boxed version soon. Even if XP was probably not even intended for this system cause too old.. the iGPU for example only had Win7 drivers but I suppose the subsystem might have XP drivers.

Sorry, I realize I was unclear . What I meant to say is that, having seen the datasheet, I doubt that you are getting 66MHz PCI under Linux, as the motherboard tops off at 33MHz PCI . The performance difference between Windows 8.1 and Linux is likely due to something else .

No problems, thank you I understood the point and I was almost hoping that was the problem because this situation seems so complex.. 😀
For sure I know the video card CAN give faster frame rates and it's not a PCI or GPU fault the speed I'm getting when with the same cpu and rams and gpu even with a linux wrapper so not the ideal native driver/api situation the frame rates are double the Win ones. I suspect still the problem might be in these subsystem components drivers or config. I tried a program to see the listed PCI devices and as said both the Intel 82801 bridge and the PLX chip are showed. Maybe are there any utility to force PCI speed? I remember back in the past some video card utility that helped to change PCI register options and check their speed.. Anyway it sure has something to do with this specific o.s. and I suppose W10 might be even more difficult.
I tried Rivatuner but seems to not be compatible with this GPU.

I just had a thought about a really hacky workaround . If your board has no PCI Express slots but does have a free mini PCI Express slot for a WIFI card, you can get an adapter that should allow a PCI Express video card to work in that at x1 speed .

I know this not answer any questions about why performance sucks in Windows 8.1 , but it may give you a workaround for better performance .

darry wrote on 2020-09-22, 19:38:

I just had a thought about a really hacky workaround . If your board has no PCI Express slots but does have a free mini PCI Express slot for a WIFI card, you can get an adapter that should allow a PCI Express video card to work in that at x1 speed .

I know this not answer any questions about why performance sucks in Windows 8.1 , but it may give you a workaround for better performance .

Eh eh I wish I could. Unfortunately the mini pci-ex bus is factory unsoldered in this version while still there and the system definitely see PCI-EX line in the system devices. I think that somehow even the SGX545 is working with some PCiEX to PCI bridged connection cause I don't think it was a PCI-EX native iGPU but more an AGP one.
Anyway i tried something interesting, increasing the PCI Latency timer into the bios from 32 to 128. I read around that if the PCI bus is a single one it might be better and in fact it seems 3dmark values are definitely increased almost double.. also I forced an old Win7 Intel Atom D2xxx PCI host driver into the device drivers instead of the MS one from the chipset package.
Problem solved? Not really.. it still clearly working not as fast as it should... I would expect higher results of the Linux test from native Directx API calls. I might try increasing the PCI latency timer to 263.I wish to find also some PLX native drivers instead of the MS default ones.. but I only found some SDK sw kit. This whole PCI-EX to PCI bridges are simply awful when a PCI-EX would have been possible. The choice of the PCI bus instead of a PCI EX1 connector was simply absurd.. I understand it might have been for industrial components reasons, but in the 2013 PCI-EX was already quite common and in this mobo is like not even used even if fully there. If I didn't installed my own retail Win8.1 boxed license on this mobo I'd have already switched it to another one but the other most modern I have is a G41/775 mobo.. not that modern too but at least it has a PCI-EX 16x connector.