cskamacska wrote:

dirkmirk wrote:

Until then some link spam: Suhian.com Ultimate PCI Round-up courtesy of the Internet Archive from almost 15 years ago Then avail […]
Show full quote

Until then some link spam:
Suhian.com Ultimate PCI Round-up courtesy of the Internet Archive from almost 15 years ago
Then available low and mid range cards PCI versions duke it out in era appropriate games.
I honestly expected more from the FX5600 compared to the FX5200, but looks like its optimisations are reserved for DX9 games that the PCI version cant run properly anyway. 😁 Its on par with the PCI Radeon 8500LE at least.

Thanks for that link never that round up before!

We have the advantage of optimized drivers since the testing was done in October 2003, I found the best drivers for the FX5600 were the 71.84 released March 2005 or 91.31 June 2006

I'll have to do some comparisons for the 9100 as well, those drivers were immature when first released(as the 8500).

My bias towards these cards were for windows 98SE and most games up til 2003 or 04(Not Doom3 or Farcry), if you want to play games up til 2006 I'd suggest looking at newer cards.

The nvidia 6200 is probably a better for later games as it had better compression technology and will play older games well, Far cry & Doom3 still run like garbage and was released in 2005, so its a question of how time correct you want to be with your build, to play those demanding games from 2004 on PCI bus you might looking at a 2006+ card...

A quick google bought up a Doom3 video on the 8400gs which is a 2007 card, this forum suggests their was a sparkle 8500GT & Albatron 8600GT on the pci bus, these would be stupidly powerful compared to anything else that came before.

https://www.techspot.com/community/topics/pny … d.108823/page-2

gdjacobs wrote:

Asaki wrote:

I didn't know there were DOS games that used nVidia, I've only seen a dozen or so that use Glide..

I think he's maybe overstating the case for TNT cards. They're pretty good for DOS compatibility and decently fast, but I'm not sure that they're better than VESA 2.0 S3 cards. For DOS 3d work, you'd almost certainly want a 3dfx based card (ideally a Voodoo 1) alongside or as the sole video card. There are also a handful of specialty titles that are best with a Verite or 3d blaster CGL cards (of which I'm definitely no expert).

Yeah thats what Voodoo 2 cards are for, if and when you want DOS 3D. 😜

But as a main PCI card? Banshee and Voodoo 3 cards are overpriced for what they do, not to mention V4/V5 or Rush. Yeah i know, 3dfx stuff is the go to solution to everything graphics related from 96 to 2001. Cue 300 euro offers for a Voodoo 5 PCI card, and can go higher than that. 😁
TNT2 derived PCI cards are prolly the cheapest(22USD currently for an M64 on teh bay, can go lower) and most plentiful of the era, have okay image quility, great 2D DOS game compatibility and performance(with VBE 3.0 so no need for TSR), Win9x/2000/XP drivers up to 2005, and even have beta Win3.1 driver support.

dirkmirk wrote:

A quick google bought up a Doom3 video on the 8400gs which is a 2007 card, this forum suggests their was a sparkle 8500GT & Albatron 8600GT on the pci bus, these would be stupidly powerful compared to anything else that came before.

https://www.techspot.com/community/topics/pny … d.108823/page-2

Sparkle made a lot of PCI cards, 7300GT, 8400GS, 8500GT, and even 9500GT.

Direwolf007 made some blog posts about his 8500GT: http://direwolf007.blogspot.com/2008/

Soo much hate that guy got on the techspot forums: So you only have PCI slots and want to game?

The people there did not realise even for one second that while PCI graphics card use became ever so situational after the Voodoo 2 era, and manufacturers dumped the idea of anything but the very lowest end after Inno3Ds mid range FX5600/FX5700 run, PCI graphics cards still had their place, and if need be they could be even used for gaming given proper over provisioning.

Like getting a GT430/520 or 9500GT for a game that worked great on a PCI-E or AGP 128bit Gef 6200 OC, because it will just suffer on a run of the mill PCI 6200 with a 64 bit underclocked memory bus. 😀

Seems 9500GTs are still for sale here and there: http://www.atdcomputers.com/sparkle-geforce-9 … video-card.html

SPARKLE GEFORCE 9500 GT DIRECTX 10 SP95GT1024D2LHP 1GB 128-BIT DDR2 PCI HDCP READY LOW PROFILE VIDEO CARD
Availability: IN STOCK - Only 1 left - $75.99

Oh man i wanted this(and its 8500GT based passively cooled forerunner) card for so long. 😐 But then the GT430 Zotac cards arrived(which i also didnt get personally 🤣 and regretted that dearly later), and the platonic love faded.

To give you guys a heads-up 😀

^Those are :

^Bottom left is Radeon 9250, also GPU-z gets confused with PCI-e to PCI bridges on newer cards...
When collection is completed, I plan to do full testing of these with OC'ed Core i7 980X and Rampage II Extreme 😀

cskamacska wrote:

gdjacobs wrote:

Asaki wrote:

I didn't know there were DOS games that used nVidia, I've only seen a dozen or so that use Glide..

I think he's maybe overstating the case for TNT cards. They're pretty good for DOS compatibility and decently fast, but I'm not sure that they're better than VESA 2.0 S3 cards. For DOS 3d work, you'd almost certainly want a 3dfx based card (ideally a Voodoo 1) alongside or as the sole video card. There are also a handful of specialty titles that are best with a Verite or 3d blaster CGL cards (of which I'm definitely no expert).

Yeah thats what Voodoo 2 cards are for, if and when you want DOS 3D. 😜

For DOS, the Voodoo 1 is better.
Voodoo 2 DOS Glide compatibility matrix

That 9500GT looks like an interesting card, I cant imgaine anything newer like the GT430s could be any faster for gaming.

What a surprise. Just earlier today I looked into a video about exactly that topic:
https://www.youtube.com/watch?v=yNEge5r6-mg

Basic result is, that even a GT 520 loses between 30 and 60% of it's performance compared to PCIe x1 and between 30 and 80% compared to PCIe x16.
The tests were run on a PCIe 1.1 board.
PCIe 1.1 x16 has circa double the bandwith of AGP 8x and PCIe 1.1 x1 is comperable to AGP 1x

So it would be pretty much pointless to go for the "fastest" PCI card available. (Which would be a pretty rare Zotac GT 430 with a GF108 chip, close behind are the GT 520/610 with the GF119 chip), but instead opt for something that doesn't lose too much of it's potential through that old bus. A card with proper memory would probably hold up better. Wider bus and bigger size would clearly help.

To the topic of what the fastest PCI card is:
https://www.youtube.com/watch?v=GCPsfWfdFmw
And yes, depending on the game they are either between a GF 4 Ti 4600 and a Radeon 9800 Pro, sometimes fater than those (But still slower than a 7800 GTX) but often even slower than both.

Ah, this is getting interesting! Thanks as always to everyone for the continued input.

What I'm taking away at this point are basically "duh" facts: performance from a PCI card is ultimately going to come down to its memory bandwidth and its architecture, in the context of which version of DirectX a game utilizes. Older PCI offerings with 128-bit memory bandwidth sometimes seem to do better with DirectX 8 games and benchmarks than newer cards, whereas newer cards do better with DirectX 9.0c offerings, and etc. for in-between. Of course this doesn't appear to be hard fact for every game, but there are advantages between memory bandwidth versus architecture depending – which I assume wouldn't be the case if all newer offerings, like the Zotacs, had the 128-bit memory bandwidth.

lordmogul wrote:

What a surprise. Just earlier today I looked into a video about exactly that topic: https://www.youtube.com/watch?v=yNEge5r6-mg […]
Show full quote

What a surprise. Just earlier today I looked into a video about exactly that topic:
https://www.youtube.com/watch?v=yNEge5r6-mg

Basic result is, that even a GT 520 loses between 30 and 60% of it's performance compared to PCIe x1 and between 30 and 80% compared to PCIe x16.
The tests were run on a PCIe 1.1 board.
PCIe 1.1 x16 has circa double the bandwith of AGP 8x and PCIe 1.1 x1 is comperable to AGP 1x

So it would be pretty much pointless to go for the "fastest" PCI card available. (Which would be a pretty rare Zotac GT 430 with a GF108 chip, close behind are the GT 520/610 with the GF119 chip), but instead opt for something that doesn't lose too much of it's potential through that old bus. A card with proper memory would probably hold up better. Wider bus and bigger size would clearly help.

To the topic of what the fastest PCI card is:
https://www.youtube.com/watch?v=GCPsfWfdFmw
And yes, depending on the game they are either between a GF 4 Ti 4600 and a Radeon 9800 Pro, sometimes fater than those (But still slower than a 7800 GTX) but often even slower than both.

I think after watching these videos and taking in all the other input, I may try three approaches instead of two: a 128-bit FX card, one of the better 64-bit Zotac offerings, as well as one of the in-between offerings, such as the 9500 GT with 128-bit bus, then benchmark them against games of different years, such as Pixel Pipes did. Definitely very interesting to see the huge spread of cards made over the years and widely varying benchmark results. Just makes me all the more excited. 😀

murrayman wrote:

I think after watching these videos and taking in all the other input, I may try three approaches instead of two: a 128-bit FX card, one of the better 64-bit Zotac offerings, as well as one of the in-between offerings, such as the 9500 GT with 128-bit bus, then benchmark them against games of different years, such as Pixel Pipes did. Definitely very interesting to see the huge spread of cards made over the years and widely varying benchmark results. Just makes me all the more excited. 😀

Here's a sneak peek at your results 😀
FX 5500 :

9500 GT :

GT 610 :

Biggest problem with GT 610 is it's SM configuration which kills Pixel Fillrate performance : LINK.
Because of this, regardless of how many ROPs it's got - it will never be able to fully take advantage of them.
Now, second issue is TMUs : In Fermi they got more lacal cache to play with vs. Tesla (ie. 9500 GT), BUT when cache isn't enough - Texel Fillrate performance tanks, as 1,3GHz 64-bit DDR3 < 800MHz 128-bit DDR3 in bandwidth.
FX 5500 is simply bad in newer stuff (DX9) and decent in older stuff (DX8 and older).
GF 6200 may be better in efficiency however you crash into another 128-bit vs. 64-bit debate 😀

agent_x007 wrote:

Here's a sneak peek at your results :) FX 5500 : ... 9500 GT : ... GT 610 : ... Biggest problem with GT 610 is it's SM configur […]
Show full quote

murrayman wrote:

I think after watching these videos and taking in all the other input, I may try three approaches instead of two: a 128-bit FX card, one of the better 64-bit Zotac offerings, as well as one of the in-between offerings, such as the 9500 GT with 128-bit bus, then benchmark them against games of different years, such as Pixel Pipes did. Definitely very interesting to see the huge spread of cards made over the years and widely varying benchmark results. Just makes me all the more excited. 😀

Here's a sneak peek at your results 😀
FX 5500 :
...
9500 GT :
...
GT 610 :
...
Biggest problem with GT 610 is it's SM configuration which kills Pixel Fillrate performance : LINK.
Because of this, regardless of how many ROPs it's got - it will never be able to fully take advantage of them.
Now, second issue is TMUs : In Fermi they got more lacal cache to play with vs. Tesla (ie. 9500 GT), BUT when cache isn't enough - Texel Fillrate performance tanks, as 1,3GHz 64-bit DDR3 < 800MHz 128-bit DDR3 in bandwidth.
FX 5500 is simply bad in newer stuff (DX9) and decent in older stuff (DX8 and older).
GF 6200 may be better in efficiency however you crash into another 128-bit vs. 64-bit debate 😀

Thanks for this! From what I had gathered before seeing these results, I had a feeling the 9500 GT would be what I end up being happy with, and this largely confirms that. If anything, I may start with that card, and then retroactively try the other two cards specifically with games versus 3DMark benchmarks. Again, I know the stats are out there for all of them (as they've already been listed in this thread), but I think it might be fun to play with them still. Thus far though, the 9500 GT seems to fulfill what I want from the stats I've seen, and it's close enough to period-appropriate for me to be okay with it. Not perfect -- I really am more of a purist to period-correctness than I probably ought to be with projects like this -- but it'll suffice.

If you plan to game in later titles (Shader based ones), both cards will do good.
9500 GT will get the edge in any game that is memory bandwidth or Fillrate bottlenecked.
I can confirm that both Quake III Arena and Doom 3 prefer 9500 GT, BUT GT 610 is quite close (ie. within 10%).
Bit of warning :
3DMark 99 MAX 9500 GT :

3DMark 99 MAX GT 610 :

Take a look at rasteriser score.
It seems Fermi loses some driver optimisations for really old stuff (I tested it again on same driver as 9500 GT - results were the same). Not sure why tho (since game tests are good).

Asaki wrote:

Wish I could use my Voodoo 3 in my current Win98 machine, but I can't get any PCI video cards that I've tried to work with this motherboard, it just defaults to AGP.

Hmm...though now that I think about it, I haven't tried updating the BIOS yet...

Well I don't know exactly if that's what fixed it, but after a BIOS update, I popped my Voodoo 3 into slot 1 and it's been working great! =D

gdjacobs wrote:

I think he's maybe overstating the case for TNT cards. They're pretty good for DOS compatibility and decently fast, but I'm not sure that they're better than VESA 2.0 S3 cards. For DOS 3d work, you'd almost certainly want a 3dfx based card...

Well he inspired me to pull out a TNT2 AGP card and give it a look. It was interesting playing through a few titles with it, but I definitely get better graphics and speed from the 3DFX.

Didn't test anything in DOS though, I don't know if I have any games that could even use it.

I can't say anything about value vs. cost, I pulled both of these out of some old junk boxes. I try not to think about what eBay is charging >_<

murrayman wrote:

...I had a feeling the 9500 GT would be what I end up being happy with, and this largely confirms that. If anything, I may start with that card, and then retroactively try the other two cards specifically with games versus 3DMark benchmarks.

I still think you should test run the integrated graphics before you get too worked up, but again, I'm no expert.

Asaki wrote:

...

murrayman wrote:

...I had a feeling the 9500 GT would be what I end up being happy with, and this largely confirms that. If anything, I may start with that card, and then retroactively try the other two cards specifically with games versus 3DMark benchmarks.

I still think you should test run the integrated graphics before you get too worked up, but again, I'm no expert.

No worries, that is what I intend to do first once I unbox it, then report back. 😀

Alright, here's what I'm working with, and 3DMark01 results:

Turned out to be an HP Compaq dc5100 SFF. Surprised at the performance of the integrated chipset FWIW, but there's a few years advancement between 3DMark01's release date and this rig. Will try a couple games this coming week when I get a chance and figure out where to go from there.

That's a more modern system than I thought. I didn't know there were LGA775 systems that still only had 32-bit PCI slots, but I guess it was bound to happen.

The performance does look quite decent. However, the problem I suspect you might run into is rendering glitches. Things don't always work properly with Intel graphics (of that era anyway). I don't know who is most to blame - maybe it's Intel, or maybe it's the game developers (who are inclined to assume everybody has NVidia or ATI). But either way it seems that gaming on Intel graphics can be problematic.

Looks like this system has a PCIe x1 slot after all. Don't have the time to look in it right now, but according to specs and the one video of this model I found on YT, looks like it has one -- which makes a lot of this null and void in terms of a solution for peak performance. But I'll tackle that once I confirm this for sure; I'll be playing around with the system and ideas pitched here either way.

murrayman wrote:

Alright, here's what I'm working with, and 3DMark01 results:

Hmm, I was going to try mine out on my P4 laptop, just for the fun of it, but 3DMark01 says I have to have DirectX 8.1 installed. I guess it doesn't realize that 9.0c is a bigger number.

It did decide to give me info in the error log, though:

1System info
2-----------
3
4System Info Version	2.2
5Installation ID	0x00000000
6OEM ID	
7
8CPU INFO
9CENTRAL PROCESSING UNITS
10Manufacturer	Intel
11Family	Pentium® 4
12Internal Clock	1.95 GHz
13Internal Clock Maximum	1.95 GHz
14External Clock	100 MHz
15Socket Designation	uFC-PGA Socket
16Type	Central
17Upgrade	ZIF Socket
18Capabilities	MMX, CMov, RDTSC, SSE, SSE2
19Version	Model 2, Stepping 7
20CPUID	0x00000f27
21
22CACHES
23Level	1 
24Capacity	8 KB
25Type	<unknown>
26Type Details	<unknown>
27Error Correction Type	<unknown>
28Associativity	<unknown>
29
30Level	2 
31Capacity	512 KB
32Type	<unknown>
33Type Details	<unknown>
34Error Correction Type	<unknown>
35Associativity	<unknown>
36
37DIRECTX INFO
38DirectX Version	8.0
39
40DIRECTDRAW INFO
41DirectDraw Version 	5.03.2600.5512
42
43DISPLAY DEVICES
44Description	Primary Display Driver
45Manufacturer	NVIDIA
46Name	NVIDIA GeForce4 420 Go (Toshiba)
47Total Local Video Memory	16 MB
48Total Local Texture Memory	16 MB
49Total AGP Memory	254 MB
50Display Driver	nv4_disp.dll
51Display Driver Version	6.14.10.4591
52Driver WHQL Certified	Yes
53Max Texture Width 	2048 
54Max Texture Height	2048 
55Max User Clipping Planes	0 
56Max Active Hardware Lights	8 
57Max Texture Blending Stages	8 
58Textures In Single Pass	2 
59Vertex Shader Version	1.1
60Pixel Shader Version	N/A

61Max Vertex Blend Matrices	0 
62Max Texture Coordinates	8 
63Vendor ID	0x10de
64Device ID	0x0175
65Sub-System ID	0x00011179
66Revision	0xa3

Asaki wrote:

Hmm, I was going to try mine out on my P4 laptop, just for the fun of it, but 3DMark01 says I have to have DirectX 8.1 installed […]
Show full quote

murrayman wrote:

Alright, here's what I'm working with, and 3DMark01 results:

Hmm, I was going to try mine out on my P4 laptop, just for the fun of it, but 3DMark01 says I have to have DirectX 8.1 installed. I guess it doesn't realize that 9.0c is a bigger number.

It did decide to give me info in the error log, though:

Copy code to clipboard

1System info
2-----------
3
4System Info Version	2.2
5Installation ID	0x00000000
6OEM ID	
7
8CPU INFO
9CENTRAL PROCESSING UNITS
10Manufacturer	Intel
11Family	Pentium® 4
12Internal Clock	1.95 GHz
13Internal Clock Maximum	1.95 GHz
14External Clock	100 MHz
15Socket Designation	uFC-PGA Socket
16Type	Central
17Upgrade	ZIF Socket
18Capabilities	MMX, CMov, RDTSC, SSE, SSE2
19Version	Model 2, Stepping 7
20CPUID	0x00000f27
21
22CACHES
23Level	1 
24Capacity	8 KB
25Type	<unknown>
26Type Details	<unknown>
27Error Correction Type	<unknown>
28Associativity	<unknown>
29
30Level	2 
31Capacity	512 KB
32Type	<unknown>
33Type Details	<unknown>
34Error Correction Type	<unknown>
35Associativity	<unknown>
36
37DIRECTX INFO
38DirectX Version	8.0
39
40DIRECTDRAW INFO
41DirectDraw Version 	5.03.2600.5512
42
43DISPLAY DEVICES
44Description	Primary Display Driver
45Manufacturer	NVIDIA
46Name	NVIDIA GeForce4 420 Go (Toshiba)
47Total Local Video Memory	16 MB
48Total Local Texture Memory	16 MB
49Total AGP Memory	254 MB
50Display Driver	nv4_disp.dll
51Display Driver Version	6.14.10.4591
52Driver WHQL Certified	Yes
53Max Texture Width 	2048 
54Max Texture Height	2048 
55Max User Clipping Planes	0 
56Max Active Hardware Lights	8 
57Max Texture Blending Stages	8 
58Textures In Single Pass	2 
59Vertex Shader Version	1.1
60Pixel Shader Version	N/A

…Show last 6 lines

61Max Vertex Blend Matrices	0 
62Max Texture Coordinates	8 
63Vendor ID	0x10de
64Device ID	0x0175
65Sub-System ID	0x00011179
66Revision	0xa3

Hmm, that is odd. I have 9.0c on the HP as well, and it didn't seem to give me any kickback. Running XP SP3 with all current updates.

I'm on FLP, and I know there's a couple hacks you have to do to get DX and .NET to install, so maybe that's where it got confused.

murrayman wrote:

Looks like this system has a PCIe x1 slot after all. Don't have the time to look in it right now, but according to specs and the one video of this model I found on YT, looks like it has one -- which makes a lot of this null and void in terms of a solution for peak performance. But I'll tackle that once I confirm this for sure; I'll be playing around with the system and ideas pitched here either way.

The Dc5100 should indeed come with a PCIe 1x slot.
You can download the manual here:
https://www.manualslib.com/manual/72038/Hp-Dc5100.html
I have a similar setup in a micro tower here, the sad thing is that (obviously) in order to save costs they did not solder in a PCIe 16x slot.
photo from the web:

PARKE wrote:

The Dc5100 should indeed come with a PCIe 1x slot. You can download the manual here: https://www.manualslib.com/manual/72038/Hp- […]
Show full quote

murrayman wrote:

Looks like this system has a PCIe x1 slot after all. Don't have the time to look in it right now, but according to specs and the one video of this model I found on YT, looks like it has one -- which makes a lot of this null and void in terms of a solution for peak performance. But I'll tackle that once I confirm this for sure; I'll be playing around with the system and ideas pitched here either way.

The Dc5100 should indeed come with a PCIe 1x slot.
You can download the manual here:
https://www.manualslib.com/manual/72038/Hp-Dc5100.html
I have a similar setup in a micro tower here, the sad thing is that (obviously) in order to save costs they did not solder in a PCIe 16x slot.
photo from the web:

The attachment 14950410708316_b.jpg is no longer available

Thanks for the link! Here’s a shot of the inside: