You said something in the beginning about this card working perfectly in a Core 2 Duo PC, why not use that for your old games then?

d1stortion wrote:

You said something in the beginning about this card working perfectly in a Core 2 Duo PC, why not use that for your old games then?

that's not mine. it belongs to my friend, used it to test the card and played NFS:MW.

i think i should stop at this point as i'm not progressing further. anyway i will be on travel this week and won't be able to devote time. if i get any breakthrough or buy some new part, i will share it definitely. if anyone has a via chipset (ple133 or apollo pro 133) based motherboard (listed below) and pci graphics card, please share your benchmarks.
Asus CUPLE-VM, Asus P3V4X
Gigabyte GA-6VMM
Aopen AX34
MSI-6309, MSI-6368
Soyo SY-7VCA/6VCA

Have you tried the VIA PCI Latency patch? It tweaks the PCI bus and on some boards with particularly poor BIOS programming can make a large improvement to bandwidth.
https://web.archive.org/web/20070331195454/ht … m/net/software/

swaaye wrote:

Have you tried the VIA PCI Latency patch? It tweaks the PCI bus and on some boards with particularly poor BIOS programming can make a large improvement to bandwidth.
https://web.archive.org/web/20070331195454/ht … m/net/software/

earlier tried both ver 0.19 and 0.20beta21, the performance actually went downhill. q3arena timedemo went from 113 to 99 fps with both patches. i also tried via 4 in 1 driver ver 4.43 and hyperion pro(though not advised) from via driver support website, but no change.

tomorrow i will install dos and share some benchmarks.
.

Oh well. 😀

ran phil's benchmark and entered the results into the google doc.

3dbench2=104.7
PCPBench=99.3
Doom= 2072 Realticks
Quake=77.2

Edit 1: changed 3dbench2 score from1047 to 104.7, missed the dot!!!

GeorgeMan wrote:

gandhig wrote:

Seeing the length of the posts in the first page itself, I think the forum's administrator should consider redesigning it with a 'recent posts - first' approach. just kidding.
the pain while scrolling down to the bottom of this post in my mobile phone everytime when there is an update!!!

You can always go straight to the first unread post by clicking that little yellow "page" button which appears on the beginning of the threads that have unread posts. 😉

@GeorgeMan, thanks for the tip. i didn't know that.

sliderider wrote:

PcBytes wrote:

I wish the Optiplex cases weren't proprietary.If they wouldn't have been proprietary,just imagine what you could hide in there.

Yeah, I know. I wish I could mount a standard ATX motherboard in them. As it is, though, I already have a 1.4ghz Tualeron, GeForce 9500GT 1gb PCI, and a standard ATX PSU with harness adapter to make it work with Dell's proprietary power connectors in one of mine. The only things I hate about it is A) No AGP slot and B) limited to 768mb of RAM. I also maxed out the RAM on the onboard Rage video because it had a slot for it and I got a box full of GX1 parts on ebay for cheap that included the memory module so I figured I'd put in there.

hi Sliderider, do you have this 9500gt pci system? if so can you confirm its really poor performance in games upto 2001 and share some benchmarks? i'm requesing because this card has some similarities with my gt520 pci.

after a hectic week of travel, i'm continuing from where i left last weekend. i found some benchmarks for the bandwidth testing:

D3D Bandwidth test:

Video Memory to System Memory (Download) = 65.32 MB/s
System Memory to Video Memory (Upload) = 61.57 MB/s

Pete Frame Buffer Read Test

Back Buffer:
Format (1,2,3,4) = {8.9, 11.6, 9.19, 11.52} MB/s

Front Buffer:
Format (1,2,3,4) = {9.16, 11.8, 9.03, 11.52} MB/s

DPCLatency:

Absolute Max = 1295 microsecs; Avg = 1029 microsecs;

CUDA-Z:

CPU--->GPU = (Pinned, Pageable) = {69.34, 58.24} MB/s
GPU--->CPU = (Pinned, Pageable) = {69.21, 59.15} MB/s
GPU--->GPU = 3856.42 MB/s

Sandra 2009:

Aggregate Memory Performance = 5.37 GB/s
Internal Memory Bandwidth = 7.52 GB/s
Data Transfer Bandwidth = 70 MB/s
System to Device = 70 MB/s
Device to System = 70 MB/s

yesterday i tried this pci card on my friend's core2duo system and did some benching to rule out some possible causes. the bold values are that of my system.

Quake3Arena demo1, normal settings with lightmap off:
Avg : 245 fps (113)

unreal 1: flyby, normal settings with volumetric lighting off & d3d renderer:
min : 78 fps (22)
max : 500+ fps (107)

3dMark01 SE (disabled threaded optimization from nvidia cpl):
Score : 10862 (5443)
Game1 (Lo/Hi) : (155.5,68.3) fps (59.9,17.6)
Game2 (Lo/Hi) : (107.2,68.4) fps (107.3,55.9)
Game3 (Lo/Hi) : (184.5,123.5) fps (65.0,27.9)
Dot3Bump Mapping: 131.4 fps (59.3)
Vertex Shader: 194.5 fps (69.3)
Pixel Shader: 100.2 fps (66.7)
Advanced Pixel Shader: 297.7 fps (280.6)

Final Reality:
Overall 3D, 9.158, Rmark (3.645)
Overall 2D, 66.225, Rmark (6.273)
Overall bus rate, 1.425, Rmark (1.227)
Video card bus transfer, 5A, 28.73, MBps (22.49)
Direct3D bus transfer, 5A, 30.62, MBps (28.31)
OVERALL SCORE, 25.118 Rmark (4.071)

D3D Bandwidth test:
Video Memory to System Memory (Download) = 113.47 MB/s (65.32)
System Memory to Video Memory (Upload) = 62.25 MB/s(61.57)

Pete Frame Buffer Read Test
Back Buffer:
Format (1,2,3,4) = {21.30, 28.40, 21.58, 29.18} MB/s {8.9, 11.6, 9.19, 11.52}
Front Buffer:
Format (1,2,3,4) = {21.50, 28.75, 21.69, 29.19} MB/s {9.16, 11.8, 9.03, 11.52}

CUDA-Z:
CPU--->GPU = (Pinned, Pageable) = {62.4, 62.26} MB/s {69.34, 58.24}
GPU--->CPU = (Pinned, Pageable) = {62.41, 61.44} MB/s {69.21, 59.15}

i guess, this rules out pci slot bandwidth limitation atleast for games upto 2001. however the implementation of pci in my chipset may not be cutting it. then there is the factor of the northbridge to southbridge link which uses the PCI bus in my case as there is no v-link or hub architecture. i can also rule out irq sharing as there was no significant improvement when i reinstalled windows with acpi disabled. i'm unable to rule out graphics driver optimization for sse2 due to lack of direction in which to proceed and starting to suspect more towards the present condition of my hardware and not the hardware itself.

Your system scored better in CUDA than a Core 2 Duo. How do you explain that?

d1stortion wrote:

Your system scored better in CUDA than a Core 2 Duo. How do you explain that?

yeah it was an aberation, albeit a small one. i think the core2duo system had more processes(around 47 iirc) running including antivirus compared to mine (20).
aside that, when i tweaked the pci latency timer register of the host bridge in my system, i was able to make it reach (pinnable,pageable) 89.9, 88.2 MB/s respectively, but the real world performance in games actually came down due to improper system balance. i observed that cuda-z results only indicate the maximum transfer rate in real-time and greatly varies depending on system load.

guys, i need help. i ran the final reality benchmark on the core2duo system with GT520 PCI few days back and yesterday i ran it in the same system with onboard intel graphics. the results are listed below, with the figures in brackets that of GT 520.

Final Reality:
2D Tests:
Radial blur, 5N, 586.92 (579.63), rips
Chaos zoomer, 5N, 1057.22 (1075.92), rips
3D Tests:
25 Pixel, 5N, 4349.66 (4226.32), kpps
Robots, 5N, 73.95 (222.72), rips
Fillrate, 5N, 17.50 (17.20), MPps
City scene, 5N, 96.99 (259.99), rips
Bus Transfer Tests:
Video card bus transfer, 5N, 2888.76 (28.73), MBps
Direct3D bus transfer, 5N, 1539.80 (30.62), MBps

Overall 3D, 6.651 (9.158), Rmark
Overall 2D, 66.275 (66.225), Rmark
Overall bus rate, 103.861 (1.425), Rmark
OVERALL SCORE, 39.120 (25.118) Rmark

the 2d results are basically the same along with the fillrate. i have limited knowledge about the intricacies of rendering (2d & 3d) and have a fundamental question. How important is the 2d performance of a card for a 3d game like Half life1, Deus Ex1, Quake3? am i correct in presuming the GT520 PCI has no effect whatsoever on the 2d performance(w.r.t above comparison) and it is completely emulated. now i'm not clear about the 'emulated' part - whether the cpu does all the work or i may sound stupid and ask whether the onboard graphics is used for it though it is disabled by bios/chipset? further i thought the fillrate is dependent on the graphics subsystem, whereas the above results indicate otherwise. however it might seem logical if the onboard graphics is involved with or without the discrete card.

This is not quite what you are asking for, but at least they are results from a PIII using a PCI graphics card.

Final Reality 1.01
2D Image Processing: 11.62 Rmarks
3D Image Processing: 5.00 Rmarks
Overall Bus Transfer Rate: 3.30 Rmarks

CPU: Tualatin PIII-S 1400 MHz
Motherboard: Intel SAI2
Chipset: Intel ServerWorks ServerSet III LE
Graphics: Nvidia Quadro FX600 PCI

This was many years ago, but I had similar problems with this motherboard while using a PCI graphics card which was also based on a PCI-E to PCI bridge. This motherboard has no AGP port, but it does have two PCI-X slots. I originally figured that I could get away with not needing an AGP port by using a PCI-X graphics card in the 64-bit, 66 MHz PCI-X slot. The only broad market graphics card I could find in PCI-X format was a Matrox Parhelia 256 PCI. They are fairly rare today. It took me half a year or more to find this card for around $50. I installed the card and the motherboard POSTed fine. WinXP also booted fine until I installed the card's drivers. For whatever reason, WinXP was unable to assign resources to the card's PCI-to-PCI bridge. I disabled or removed all other expansion cards and non-essential onboard hardware to no result. I ran across one other forum where someone tried a similar combination and also had problems like this. This other user got ahold of Intel, as this motherboard was not yet EOL, and was told that a BIOS updated was needed, however Intel was unwilling to develop one. I tried to contact Matrox before EOL and they were not willing to test the card on a Tualatin computer. They could not even provide a list of compatible motherboards. Matrox stopped replying to my inquiries very early on in the troubleshooting process. They wanted me to send the card back for a replacement and bill me $350. I knew that was a scam. Long story short, I eventually settled on a Quadro FX600 PCI, however I also ran a Geforce 6200 in the 66 MHz PCI-X slot for some time.

In short, I wonder if any PCI graphic cards with an onboard PCI-E to PCI bridge will work to properly in a PIII-class motherboard, and if so, which ones?

hi feipoa, i don't know whether to call this plain freaky or just coincidence, i was actually going through your thread "Dual PIII-S Tualatin 1.4 GHz Success Stories" today morning and even went to the local used/scrap market in the afternoon to explore the server chipset route for maximizing the potential of this discrete graphics card. specifically the pci-x slot available on these chipsets and you are replying to my post!!! i think the card supports 66 MHz in the pci to pci-e bridge's primary side and the secondary side is restricted to pci-e x 1, though i have to recheck it. it was just a preliminary check btw, now that i have already suffered the consequences of improper research before buying a card that too roughly equivalent to 60$.

i know i'm doing the reverse of what normally one should do i.e get a card to suit the system. but my card is new and other parts are old, so... Also it feels really good when you are able to squeeze more and more performance out of it with every tweak.

thanks for giving your time to reply (thought no one is going to), i didn't even ready it completely and was slightly freaked out about the timing of your reply.

i went through your benchmark results of speedsys and cachechk7 in yourl thread. i will do it on mine and report with the results. glad to see the view on the via chipset's peformance in your thread. i also have a sis630 chipset based ECS-P6STP-FL board, but i don't intend to touch it as of now due to still poor memory benchmarks. interestingly iirc it scored more 2d performance points with the same graphics card compared to the via chipset, but the 3d was worse.

After my experience, I'd be weary of trying any PCIE-to-PCI graphics card on a PIII-class motherboard, unless you can find the motherboard for next to free.

The only VIA chipset I have had stable success with was the VIA Apollo Pro266T. I recapped the board though. I've had trouble with Apollo Pro 133's and MVP3's, but I'm not at the point where I'd say every motherboard based on these chipsets are bad. I still have some higher-end candidate in que to test.

It is always fun to see just how far you can take a motherboard, but inevitably, the CPU or FSB will be the slowest link.

speedsys snapshot (detailed results in attachment):

low vesa memory throughput ~ 9.2 MB/s???

cachechk7 result (detailed results in attachment):
Read: 208.7 MB/s (cachechk -d -t6)
Write: 374.0 MB/s (cachechk -w -d -t6)

cachechk is reporting low read speeds than write speeds. is it normal? i saw the trend to be reverse in the dual tualatin success stories thread.

feipoa wrote:

This is not quite what you are asking for, but at least they are results from a PIII using a PCI graphics card. Final Reality 1. […]
Show full quote

This is not quite what you are asking for, but at least they are results from a PIII using a PCI graphics card.
Final Reality 1.01
Overall Bus Transfer Rate: 3.30 Rmarks
Chipset: Intel ServerWorks ServerSet III LE
Graphics: Nvidia Quadro FX600 PCI

do you remember how you got this great bus transfer rate result? i.e with a conventional pci or pci-x slot? can you also post the individual 2d & 3d bus transfer rates. i still don't understand the difference, but still reading.

feipoa wrote:

I had similar problems with this motherboard while using a PCI graphics card which was also based on a PCI-E to PCI bridge.

i guess you are not referring to the matrox parhelia, isn't it? can you mention which pci graphics card otherwise? do you still have it and whether the Intel ServerSet system is still working and online?

feipoa wrote:

The only broad market graphics card I could find in PCI-X format was a Matrox Parhelia 256 PCI. They are fairly rare today. It took me half a year or more to find this card for around $50. I installed the card and the motherboard POSTed fine. WinXP also booted fine until I installed the card's drivers. For whatever reason, WinXP was unable to assign resources to the card's PCI-to-PCI bridge.

the parhelia got sold in ebay for 335 US $ (http://www.ebay.com/itm/Matrox-Parhelia-256MB … F-/291069304663) I almost had the same problem when i installed it first time. But i didn't even get any output, just a blank screen as the card was not detected by the bios due to pci bridge spec compliance. the issue was overcome by doing what i have mentioned in the start of this thread.

feipoa wrote:

I tried to contact Matrox before EOL and they were not willing to test the card on a Tualatin computer. They could not even provide a list of compatible motherboards. Matrox stopped replying to my inquiries very early on in the troubleshooting process. They wanted me to send the card back for a replacement and bill me $350.

i think unless you have a have a friend in the 'inside', any chance of getting a chance to take your issue to the right people let alone get a solution or workaround is far fetching. atleast my vendor didn't offer me a replacement for 350$ or such. on the positive side, the forums are the best hope where you can find a solution.

feipoa wrote:

In short, I wonder if any PCI graphic cards with an onboard PCI-E to PCI bridge will work to properly in a PIII-class motherboard, and if so, which ones?

right now it is a million dollar question to me. in my case i don't know who will yield first, the system or the graphics card or me?

I do not recall seeing individual 2d and 3d bus transfer rates. The Tom2D benchmark comparison on this computer is very similar to another Tualatin board I have with an AGP slot, so I do not think anything is going wrong with 2D.

Those scores are with the Quadro FX600 plugged into a regular PCI slot. I had the GeForce 6200 working in a PCI-X slot for about a year, then I went to update the NVIDIA drivers, and the 6200 would no longer work in the PCI-X slot, even after reverting the drivers back and formatting the HDD. I also tried another 6200 card and doing a manual clean-out of all NVIDIA files and registry entries. But, alas, this motherboard no longer wants to work with a graphics card in the PCI-X slot. I never figured out why. I do have another Intel SAI2 motherboards that is NIB I have been meaning to try. The 6200 worked fine in a regular PCI slot, so I just left it like that. The reason for switching from the 6200 to the quite rare FX600 Quadro is because I wanted dual DVI, which is hard to come by on a PCI card.

I was refering to the Matrox Parhelia PCI 256 - according to Matrox, it uses a PCI-E to PCI bridge. Their P690, which is dual DVI, also uses a PCI-E to PCI bridge, so I didn't even bother trying to source that card. There are two newish ATI cards which have dual DVI that you might want to try out. I'm not sure if they have the PCI-E to PCI bridge or not. I have one that is in que for testing, like 100 other pieces of hardware. I'm constantly looking for the fastest dual DVI web browsing graphics card on my dual Tualatin rigs because I refuse to update my everyday use computers. I'm not sure what I am going to do come April 8, 2014. One dual Tualatin has 2 GB RAM, the other is 3 GB. I can probably squeeze by with 3GB of RAM in Win7, but the 2 GB AGP-based board needs more RAM, and unfortunately, it uses some low density 1 GB DDR sticks which I've had trouble finding, especially in CL2.

I am using a Matrox Parhelia 128 AGP on my AGP-based dual Tualatin. Even with a NVIDIA vs. Matrox DVI comparison, I found the Parhelia to have a little crisper display compared to the GeForce 6600GT I had in there previously.

Yes, the Intel ServerSet computer is still setup. Its my wife's everyday use computer. It is using a Quadro FX600 PCI. Good luck finding one at a decent price. If you don't need dual DVI, there are other PCI-based graphic cards to consider which are not based on the a PCIE-to-PCI bridge. You can probably grab some good ideas from this post, Best Fanless AGP/PCI Graphics Card

Yes, I still have the Matrox Parhelia PCI 256. I'm going to hold onto it just to see if it really works. I do have a far newer motherboard with PCI-X around somewhere I'd like to test it on one day. It is one of those newfangled AMD multi-core boards which don't really suit my fancy, at least not for another 15 years.

I cannot believe somebody paid $335 for a Parhelia PCI 256. It is a historic card in that there weren't really any other consumer-grade PCI-X graphic cards, but to dish out that kind of money is mystifying. I hope that whoever bought it is not trying to put it into a PIII-class computer, they will be surely disappointed.

As for the other newish ATI dual-DVI PCI graphics card, I went with an ATI FireMv 2260 PCI 256 MB. When selecting this card, I wrote AMD tech support to ensure that it did not use an PCI-E to PCI bridge. This is the response I received,

Thank you for the clarification. Both the FireMv 2260 and the X1300 do not have an internal PCI-E to PCI Bridge.

The PCI cards were not tested on the server boards you have mentioned so we aren't sure if its going to be compatible.

As this card comes in both PCI-E and PCI versions, I was concerned about the bridge. This card is based on the RV620 core, which was launched in 2008 according to Wiki. It is not really intended as a gaming card and this is not what I use it for, however it would still be interesting to bench it on the dual Tualatin and cross compare it to whatever fleet of 2000-2005 era cards I have laying around. Anyone care to guess how a Tualatin with a GeForce 6600GT would compare with a Parhelia 128 AGP, NVIDIA Quadro FX600, or ATI FireMV 2260 in Quake 3? Are they all fairly CPU-limited?

@feipoa, do you want me to have a go at your Intel Server Set system bios for getting the parhelia pci to work with it? i mean the passmark scores for parhelia & quadro fx 600 pci were 158 & 70 respectively, obviously and only if you don't need the Dual DVI in your present Intel Server Set system. i seriously doubt games upto 2000-01 will be able to frequently saturate the PCI bus so the usefulness of the PCI-X slot is open for discussion. since you had mentioned non-gaming requirements(quake 3 only for benching), then the difference maybe noticeable in addition to games after 2002.
I had no problem with the card in a 2004 P4 system, only with P3 as you said. I found this in the bridge vendor's faq section

Any special attention needed for the 7300D placed into a very old system?
The BIOS date is important. The PI7C7300D bridge counts 225 PCI clocks from the de-assertion of P_RESET# to the first allowable PCI config cycle, per PCI Spec 2.2**. This is 33 million clocks, about 1 full second at 33 MHz. For BIOS made according to Intel/Microsoft AC97 ("green PC") specification, this 1 second (at 33 MHz) delay is not a problem. But for older BIOS standard, this will block BIOS from enumerating the PCI devices on the secondary PCI bus and thus they have no resources assigned. Since DOS uses the BIOS for PCI enumeration (enumerate == "detect and assign resources"), a motherboard BIOS older than 1996 years might not detect and allocate resources to devices behind the bridge.

Isn't current web browsing a touchy topic for P3 systems? But i don't understand the usefulness of a graphics card for web surfing as it is severely crippled by the lack of SSE2. Won't 2 GB pass for Win7? It's a shame that Win8 can use full acceleration of the graphics card for total web browsing, but the requirement of SSE2 is a bummer for P3 systems.
BTW thanks for the post on agp/pci graphics cards.