Hey guys,

I've been searching the net, but can't seem to find a conclusive answer.

The 5500PCI, is capable of running in a 66Mhz PCI slot, but will it actually run that 66Mhz?
And will there be a performance hit when running at 33Mhz pci? as in, is the 33Mhz pci bus wide enough for all the data to get through comfortable?

the faster the cpu behind it, the more data it would produce I asume, so I will be doing some testing on my rig, as it has:

1: a powerfull enough cpu to test this
2: both 33Mhz and 66Mhz pci slots.

Will be testing on different resolutions, up to 1600x1200x32

Hi,

If you put a V5 5500 PCI in a PCI-X slot it will be more fast than in a PCI simple slot.

Better to use the PCI-X slot.

Trevor.

I assume that is correct, but since I can't find any conclusive data on the net, I am gonna test the difference.
You can find a lot of PCI vs AGP for the 5500, but nowhere it states what kinda pci slot is used.
Since those benchmark are made for the masses, I assume it is tested on a 32bit/33Mhz slot, mine are 64bit/66Mhz, so I wanna see what the difference is, within the same system.

ultima wrote on 25.11.13 at 19:40:48:

I assume that is correct, but since I can't find any conclusive data on the net, I am gonna test the difference. You can find a lot of PCI vs AGP for the 5500, but nowhere it states what kinda pci slot is used. Since those benchmark are made for the masses, I assume it is tested on a 32bit/33Mhz slot, mine are 64bit/66Mhz, so I wanna see what the difference is, within the same system.

Ok so we will know if it's right. Good Benchmarks. I will follow you here.

And I forgot to mention, all those benchmarks online don't show any noticable difference in performance between a pci and agp, so if it is true that the pci is tested on a 33Mhz slot, and it is almost on par with the agp, it means 2 things:

1: the agp part is just physical difference, it doesn't use agp features or it's added speed
2: if the pci really does run @66Mhz, it should be faster then it's agp counterpart, which performs the same as a pci @ 33Mhz

Only downside is I don't have an agp version anymore for comparison. :(

ultima wrote on 25.11.13 at 22:50:24:

And I forgot to mention, all those benchmarks online don't show any noticable difference in performance between a pci and agp, so if it is true that the pci is tested on a 33Mhz slot, and it is almost on par with the agp, it means 2 things: 1: the agp part is just physical difference, it doesn't use agp features or it's added speed 2: if the pci really does run @66Mhz, it should be faster then it's agp counterpart, which performs the same as a pci @ 33Mhz

1: Indeed, all 3dfx AGP adaptors are basically PCI adaptors simply "ported" on the AGP bus. They don't support typical features introduced with the AGP standard, like DIME, fast writes and sideband addressing, the latter except for Banshee and Voodoo3. AGP speeds are 2x max for Banshee/Voodoo3, 4x max for Voodoo4 and 1x for Voodoo5

2: Nope, all 3dfx AGP adaptors can do at least AGP 1x (66Mhz) which is double the PCI bus and that is that. The only improvements of the AGP over the PCI when it comes to 3dfx video cards are:
-Double bus bandwidth
-Dedicated bus, it does mean that no bandwidth is shared across other devices and it's all available to the VGA.

A Voodoo5 PCI should have exactly the same performances of the AGP one *if* you can run the card into a PCI-X slot, which is a PCI slot capable of 66MHz (PCI-X=AGP@1x).

So Voodoo5 PCI and AGP are basically the same thing. Point at issue is that as long as you can put the AGP one into the slot and get the card to work you'll run always the card at 66MHz bus clock. For the PCI version, in order to race against the AGP one, you have to choose the right slot, the one capable of 66MHz instead of classic 33MHz, which is the PCI-X slot as already stated.

In the end, if you don't notice any relevant difference between Voodoo5 AGP and PCI, even when the latter is using a PCI 33MHz slot only, is beacuse more likely the application doesn't require more bandwidth as much the 33MHz bus can offer.

So in the end, if I don't notice any difference in performance, even @ 1600x1200x32, and FSAA @ 4x,  that means the card just can't produce enough data to saturate the pci bus, meaning the agp part is just in existence so you don't have to sacrifice a pci slot, and a marketing gimmick.

We'll see soon enough I guess, will be doing the testing this weekend.

Note- you will find a lot of performance difference running to 66MHz instead of 33MHz when a SW an application needs bandwidth. For example run a V5-5500 PCI 33MHz and later 66MHz using 3DMark2001SE where T&L is emulated by CPU. Here you will note an important improvement. Though using Quake3 timedemo the difference will be minimum.

- Oscar.

thanks for the tip, I will check that.

I have tested 3DMark2001SE before @ 66Mhz pci and 2.38Ghz, I got just over 3000 points, I wonder what 33Mhz pci will give.

I did some looking up, and bandwidth is as follows:

33Mhz 32bit PCI = 33.33 x 32 / 8 = 133MB/s
66Mhz 32bit PCI = 66.66 x 32 / 8 = 266MB/s
32Mhz 64bit PCI = 33.33 x 64 / 8 = 266MB/s
66Mhz 64bit PCI = 66.66 x 64 / 8 = 533MB/s

So the question is, where lies the sweet spot for the 5500 :)

Voodoo5 PCI is a 32bit only device, you won't have any benefit using it into a 64bit capable slot.

goriath wrote on 26.11.13 at 13:45:22:

Voodoo5 PCI is a 32bit only device, you won't have any benefit using it into a 64bit capable slot.

nope, not from the 64bit part of it, just the 66Mhz part of it :) I know what my rig scores on the 64bit slot, will test what it does on the 33Mhz slot, see how much the performance drops.

What I meant to say was that your bandwidth is limited to 266MB/s MAX (66Mhz 32bit PCI = 66MHz x 32bit / 8), so is useless to consider 64bit in the math.

that is correct, I just added it there to make it complete.

But the 5500 is actually capable of running 66Mhz PCI, else the card wouldn't score the same as the AGP part.
Or the other way around, there is absolutely no need for an agp part, aside from having a different connector, it is not needed for the speed it can provide.

ultima wrote on 25.11.13 at 19:40:48:

I assume that is correct, but since I can't find any conclusive data on the net, I am gonna test the difference. You can find a lot of PCI vs AGP for the 5500, but nowhere it states what kinda pci slot is used. Since those benchmark are made for the masses, I assume it is tested on a 32bit/33Mhz slot, mine are 64bit/66Mhz, so I wanna see what the difference is, within the same system.

In my benchmarks with a fast CPU (Pentium III 1400-S) I used a Voodoo 5 in a 33MHz PCI-slot and using lower resolutions the PCI-model is slower than the AGP version. With higher resolutions the differences between PCI and AGP get smaller eventually the difference is really small (for instance 0.30 FPS with Quake 3 at 1024x768 16-bit). With a slower CPU the difference between PCI and AGP is almost zero.

Thandor wrote on 26.11.13 at 21:47:32:

ultima wrote on 25.11.13 at 19:40:48: I assume that is correct, but since I can't find any conclusive data on the net, I am gonna test the difference. You can find a lot of PCI vs AGP for the 5500, but nowhere it states what kinda pci slot is used. Since those benchmark are made for the masses, I assume it is tested on a 32bit/33Mhz slot, mine are 64bit/66Mhz, so I wanna see what the difference is, within the same system. In my benchmarks with a fast CPU (Pentium III 1400-S) I used a Voodoo 5 in a 33MHz PCI-slot and using lower resolutions the PCI-model is slower than the AGP version. With higher resolutions the differences between PCI and AGP get smaller eventually the difference is really small (for instance 0.30 FPS with Quake 3 at 1024x768 16-bit). With a slower CPU the difference between PCI and AGP is almost zero.

That seems kinda strange, I would expect the difference to be minimal in lower resolution, as the PCI bus wouldn't be hampering it yet then. I really need an AGP version to complete the testing, but for now 66Mhz vs 33Mhz will have to do.

@ultima,

Many years ago I did some benchmark using a C2D and ASUS WS PCI-x.

Perhaps it can help you.

V5-6000 + AGP2PCI ADAPTER
3DMark2001SE
V6k@166MHz (AGP2PCI) on PCI Bus - 33MHz - 4000Points
V6k@166MHz (AGP2PCI) on PCI Bus - 50MHz - 4800Points
V6k@166MHz (AGP2PCI - Modded) on PCI Bus - 66MHz - 5300Points


V5-5000 AGP/PCI
3DMark2001SE
V5 5500 PCI - PCI-X 66MHz - 3700Points
V5 5500 PCI - PCI 33Mhz - 3000Points
V5 5500 AGP (AGP2PCI modded) - PCI-X 66MHz - 3700Points.

- Oscar.

Hi Oscar,

thanks for that data, will be nice to see a good comparison.

I'll also test different cpu speeds and with that different PCI bus speeds.

ok, here we go, very late, but still:

tests are done with cpu at 2.26Ghz and 1GB ram, not quite stable enough yet at 2.38Ghz (40Mhz pci)

3DMark99MAX @ default settings @ PCI33 = 6997 3DMarks
3DMark99MAX @ default settings @ PCI66 = 7447 3DMarks difference = 450 3DMarks

3DMark2000 @ default settings @ PCI33 =4356 3DMarks
3DMark2000 @ default settings @ PCI66 =5052 3DMarks difference = 696 3DMarks

3DMark2001SE @ default settings @ PCI33 = 1976 3DMarks
3DMark2001SE @ default settings @ PCI66 = 3070 3DMarks difference = 1094 3DMarks
3DMark2001SE @ 1600x1200x32 @ PCI66 = 1512 3DMarks



Too bad my main rig doesn't have XP on it, it does have pci66 slots, along with 2 opteron 4280's @ 3.2Ghz.

So it seems that the pci bus is really hampering the 5500, when on a 66Mhz slot, the score on 1600x1200 is only 500 less than when running default benchmark in a 33Mhz slot.

I will add a the following cards to the list:

Geforce2 GTS 32MB
Geforce4 MX440 64MB
Matrox G550 32MB
Voodoo2 12MB SLI
Voodoo3 3500
Voodoo3 3000
Voodoo3 2000
Voodoo3 1000
Voodoo Banshee
Voodoo Rush

And probably some others I forgot.

Will make a nice spreadsheet and incase you're wondering......yes...I have too much free time. LOL

If possible, I will also test the cards under windowsXP for comparison.

It seems that once the CPU is not the bottleneck anymore, the PCI-bus will be te bottleneck. After all using a P2 450 the differences between AGP/PCI are almost zero and using the PIII-S 1400 the differences are getting bigger using lower resolutions. Because your Athlon XP is much faster the CPU speed is not an issue but the 33MHz PCI-bus just can't keep up.

I also wonder about the score of 3DMark2001 on 1600x1200 with the 33MHz PCI bus. Can you test that?

No problem, will do so today.

Also adding to the list, is the KyroII 64MB and a Geforce2 Ti 64MB.

Thandor wrote on 15.12.13 at 11:32:23:

It seems that once the CPU is not the bottleneck anymore, the PCI-bus will be te bottleneck. After all using a P2 450 the differences between AGP/PCI are almost zero and using the PIII-S 1400 the differences are getting bigger using lower resolutions. Because your Athlon XP is much faster the CPU speed is not an issue but the 33MHz PCI-bus just can't keep up. I also wonder about the score of 3DMark2001 on 1600x1200 with the 33MHz PCI bus. Can you test that?

The results are in, 1274 was the score

And I gotta say, this is more work then I expected; today I tested the 5500PCI @ 33Mhz and 66Mhz in 3DMark2000 and 2001SE at the following resolutions:

800x600x16
800x600x32
1024x768x16
1024x768x32
1280x1024x16
1280x1024x32
1600x1200x16
1600x1200x32

Only strange thing was, in 3DMark2000 it wouldn't run the test @ 1600x1200x32, the x16 it did, but on both pci speeds it crapped out with an error, maybe too little videoram?

And a friend of mine will loan me his 5500 AGP so I can test with that too :)

This is the list of cards currently up for test:

Voodoo5 5500 @ PCI33 = DONE
Voodoo5 5500 @ PCI66 = DONE
Voodoo5 5500 64MB AGP =
Voodoo3 3500 16MB AGP =
Voodoo3 3000 16MB AGP =
Voodoo3 2000 16MB AGP =
Matrox G550 32MB AGP =
Geforce2 MX400 64MB =
Geforce2 GTS 32MB AGP = DONE
Geforce2 Ti 64MB AGP = DONE
Geforce4 MX440 64MB AGP = DONE
Kyro II 64MB AGP =
TnT2 32MB AGP =

if anyone has a suggestion for a card to add, lemme know. I won't be doing tests with 3DMark99MAX for this, since it won't let me change resolution in the free version, 2000 and 2001SE I can change all I want.

EDIT:

I had DX9.0a on windows 98 SE, the Voodoo5 would run all benchmarks just fine, but when trying the GeForce series, they began nagging about dx. 8.1 is used with the rest of the cards

I just thought of a nice add-in for the test :)

I have only modded one of my XP-M's so far, so the other will default to 800Mhz when I insert it.

I remember seeing on Gary's site with his benchmarks that he got the remark that the benchmarks were not fair, since the Voodoo cards were being run on a system much faster then they would originally run on.

So I will test the cards also with the cpu @ 800Mhz, about the cpu speed they had at that time, and see how much really is gained.
Probably the voodoo's will be gaining the most, as they have software T&L and cpu speed is of big influence.

I won't be doing every test at 800Mhz as that is too much work, I'll just use 1280x1024x16 as resolution, just for fun :) Or if someone else has a better suggestion for resolution.....

Hello, my first post here.
I am not as skilled as Oschkar or other power users here.
I will be interested by your tests, I did some in the past.

All V5500 Pci, not only mac versions, have hardware capability to run @66mhz if receiving from the mobo a "run @66mhz" signal.

Briefly, this signal can be send to the V5500 pci  by 2 ways :
-Pci-X slot mobo direct plugging,
-Pci2PciE adapters (adapter must be compatible with Pci 66mhz cards) because the mobo will recognize the V5500 66mhz hardware capability through the PciE lane at the init and send the 66mhz signal to the V5500.

In my opinion, the difference between 33 and 66 mhz (everything else equal in your test) will be an average of 10% more in synthetics benchmarks like 3dmark.
In game, it depends on the CPU dependance, you will score more higher on games demanding a high CPU load.

V5500 will score higher on geometrics tests and the same on fillrate tests. So the fillrate will be the next bottleneck. Overclocking the 3dfx will help at this point, V5500 Pci takes often 185-190 mhz without modification.
At a certain point, with modern Cpu, the benefits are given by the cpu, not by the 3dfx. Some says it is artificial, that's true but if your game runs faster FPS, you will nevertheless enjoy it.

My tribute to the "66mhz way" on Pci-express.
Pictures of my V5500 (2900) dual-slot running @66 mhz.
http://img.canal-pc.fr/N0640488001394814763.jpg
http://img.canal-pc.fr/N0056111001394815050.jpg

Hei Jix-reggio :)
Welcome here mate. Your construction is fantastic.
keep up the good work !

Hello Mradu, thanks, I am interested by this thread, so I spoke of my experience with Pci-express adapters.

@Steven67,

Excellent work. Card looks very nice!! Very good job.

Would be nice if you can do some bench based on glide or opengl games using your V5 with the adapter and using it PCI 5V slot and see difference performance.  Possibly performance will be more or less same.

Regards,
Oscar.

Best pict for V5500 "Glide Killer"

http://img.canal-pc.fr/N0894240001395251896.jpg

Thanks Osckhar.
It is pretty boring to set and unset the Pci-E mod
and Ccmos is necessary each time.

So I had benched the mod on 3DM2001SE
with Pci-Express set to 90 / 95 / 100 / 105
to examine the impact of Pci or Pci-E increase.

It shows that 10% more on I/O leads to 2,2%
more on global score.
So the result is that going on Pci from 33mhz to 66mhz (100% more on I/O)
leads to 22% points more on global socre and confirm yours scores from 3000 to 3700 points.

IMHO, Glide perfs are very dependant on Glide drivers files.

This mod can runs UT Glide very fast with good drivers.
I posted videos on Youtube, but only one in Glide.

Look at this one : "How many FPS on Glide UT 1280*124"
In this video I try to generate a high GPU load by firing rockets as fast as possible  ;)
http://www.youtube.com/watch?v=_rEvfp437yI&list=UUoiMEZ7WL0u04lOnBMEDR2g

This is my mod video "Glide Killer emerge from past in 2014"
http://www.youtube.com/watch?v=aX8S7Wzs2IU&list=UUoiMEZ7WL0u04lOnBMEDR2g


I maybe could not perform tests next days but
please give me (safe) links where I can download 1 or 2
good Glide or Ogl benches and I will run as soon as possible.
Regards.

@Steven67,

Thanks for info.

Please, can you post here a 3DMark2001SE 1024x768 32Bits full test @166MHz stock speed I want to compare scores with my V5-5500 agp with my adapter AGP3.3.

Thanks,
Oscar.

Hi Osckhar,

do you need some benchmarks as well from aa 5500PCI on a pci-x 133Mhz slot with dual opteron?

just for comparison ofcourse :)

@ultima,

Yes, for sure. :)

Test for Oschkar V5500 @166 mhz
Driver XP Amigamerlin 3.1 R11


http://img.canal-pc.fr/N0503173001396555744.jpg

Excellent! This afternoon I will post my test.

Thanks,
Oscar.

ASUS RAMPAGE EXTREME SOCKET 775
AGP3.3 ADAPTER @66MHz
V5-5500 AGP @166MHz

https://dl.dropboxusercontent.com/u/4383540/test/V5_ASUS_WS/v5_64_default_1.JPG

Seems C2D @3.3GHz makes fly the V5-5500 AGP. Now we should test V5 in some GLIDE/OpenGL games where T&L is not emulated by CPU. Here we will have more realist FPS.

- Oscar.

Your Agp 3.3 adapter seems to be efficient, no problem.

The 3dm2001SE is a monothread bench very depending of the CPU frequency (and core power). I tested this when comparing A64 core from 2,2 to 2,8 ghz with my previous Opteron 150, and Trevormacro had tested on socket 775 ASUS WS mobos with a C2D 8400 if my memories are reliable.

Core power and frequency upgrade affects only " GPU / Cpu collaboration" on geometric computation and T&L as you can see on the bench results. The fillrate depending of Voodoo and the drivers was the same @315 / 571, but geometrics flyes with the C2D.

Notice your gain is very low on Lobby (very depending on texture/ fillrate performances) but good on drago (a mix of geometrics and texture) and very good on car chase (very depending on geometrics).

If you look on C2D (onecore) Passmark vs A64 "Manchester "(onecore) Passmark, you will see the difference for corepower.

In order to compare the efficience of your Agp 3.3 adapter vs my Pci-E 1X adapter, I suggest that you underclock your C2D to 2,0 or 2,2 ghz (for example) to simulate my A64 @2,4 ghz (or adjust your polygon and Vertex score to my values : 5,8 - 2,0 -22,7).

(Notice that I didn't use a 'LOD trick' to boost my geometrics, and run the whole bench in 1 pass with Pci-E locked @100 to give a reliable point of comparison.)

In my opinion, 3DM2001 scores would be the same in this case.
Pci-E-on-Pci adapters are interesting for µATX or mini-ITX configs.

I have minor setback....the ram I got with the mobo is faulty, I contacted the guy and he will send me new ram, till then, project on hold :(

@ultima : hope your mobo will run soon and you can get the best of it  !

In addition to the D3D score, I run GLIDE API with >100 FPS average 1280*1024 with my mod on A64X2 4200+.

http://www.youtube.com/watch?v=_rEvfp437yI&list=UUoiMEZ7WL0u04lOnBMEDR2g

will be ok, soon I will have the ram, 4 x 1GB DDR400, running in dual channel per cpu

will get my Aerocool Strik-X Air case in a few hours, will build it into the case and show pics :)

finally got the rig up and running.

score is 3132 all on default settings

Contribution to 3dfx

"Last rays from a dead star"

http://www.youtube.com/watch?v=07FqU0IpQS0&list=UUoiMEZ7WL0u04lOnBMEDR2g

great mod Steven !!!  :D

@ultima :
Could you show us some pictures of this rig ?
The 3dm2001se benchmark is a monothread one, I am not surprised you don't take advantage of your 4 Cpu cores @2,4 Ghz.
High cpu frequency is the key for high 3Dmark scores.
A game like IL2 Sturmovik can be played with Voodoo5
and can be optimized for dual-core.

@mradu :
I am thinking about a new model.
But first, I have to find 50 or 100 euros selling one or two things, to buy the furnitures.
On this occasion, would you like me to put online a 'workblog'
to show the steps of the construction?

IL2 Sturmovik D3D rendering on my rig.

http://youtu.be/f-qLaGi4cmg

Hei Steven ! that"s a great ideea !
I can"t wait to see the steps :)

Steven67 wrote on 01.05.14 at 14:11:27:

@ultima : Could you show us some pictures of this rig ? The 3dm2001se benchmark is a monothread one, I am not surprised you don't take advantage of your 4 Cpu cores @2,4 Ghz. High cpu frequency is the key for high 3Dmark scores. A game like IL2 Sturmovik can be played with Voodoo5 and can be optimized for dual-core.

I would rather wait with the pictures untill at least the cable management is done :)
24pins extender + 60cm IDE cables coming in tomorrow or friday at the latest, then after they are in place, I will make pictures and post them

see my other thread of what games I played so far on win2k and winXP

and here's some pictures:

1st up is the empty case, front view:

http://abload.de/thumb/emptycaseibk1v.jpg

Open case look from above:

http://abload.de/thumb/insidelookgrkyp.jpg

then I went and placed the mobo:

http://abload.de/thumb/moboinplaceyyk1r.jpg

And noticed my cables were a bit short:D

http://abload.de/thumb/morecablestooshort6vkq2.jpg

http://abload.de/thumb/whatamess8bjv0.jpg

So I got me some extension cables:

http://abload.de/thumb/wheresthecablesiqkk9.jpg

http://abload.de/thumb/alotbetter6gjyb.jpg

http://abload.de/thumb/img_20140508_211351vlkzi.jpg

the modding of the case can now get started :) and ofcourse place the 2nd cpu in there, since I got ram since a while now :)

whatta ya guys think?

That is a odd case. But it looks nice, what is that sound card like?

Yeah the case is more like an open benchmark table, used for frequently swapping hardware...ideal for testing....and it looks nice :) After some modding it will look awesome :)

Diamond MonsterSound MX300...Aureal Vortex 2 chipset....awesomeness in all it's glory :)

was better AND cheaper then anything Creative could offer at that time...very nice MIDI

and it has nostalgic value to me....got it for my birthday long long ago...then it broke and never saw one again...till some time ago......and just like the 5500 MAC card, I will NEVER EVER let it go :)

How much ram does it have ?
can it use sound fronts or midi instrument librarys? or are you stuck with the one biult into the card ?

As far as I can tell / remember, there is a daughterboard for this card , which adds extra feautures to the card, I will lookup what it does and let you know

EDIT:

there was the Roland add-in board, shown here:

http://en.wikipedia.org/wiki/File:Roland_SCD-15_on_Diamond_MX300.jpg

http://en.wikipedia.org/wiki/Creative_Wave_Blaster

The MX-25 daughtercard from Diamond itself is either never produced really or very very hard to find....but card can be made fully working under windows 7 even.