VOGONS


Reply 20 of 33, by Jasin Natael

User metadata
Rank Member
Rank
Member
LubieCipy wrote on 2021-10-05, 20:49:
Jasin Natael wrote on 2021-10-04, 21:18:
Not calling you a liar but what games did you test? […]
Show full quote
LubieCipy wrote on 2021-10-03, 10:35:

I can't exactly answer your question but I noticed that even covingon celeron @ 266 MHz is sufficient for Voodoo3 for games from 1998-2000 at 1024x768 resolution

Not calling you a liar but what games did you test?

I find it hard to believe that Unreal/UT would run acceptably at that resolution on a cacheless Celeron.

I'm sure there are others as well.

Sorry for delay 😀 it took me a while to find the old tests. Games with build in timedemos (Quake 3, turok 2 etc.) scores 25-35 fps in 1024x768 and even 1280x1024. It is obvious that the CPU is a bottleneck so the results at 640x480 are very similar. I did not check UT 😀

Well now I am confused.

In you original post you say that the Covington @ 266mhz is "sufficient" for a Voodoo 3 at 640x480.
Then is this post you are telling us that it is "obvious the CPU is a bottleneck"

So which is it? Maybe I just misunderstood your first post and if so I apologize.

I'm quite certain that the Covington is a bottleneck to a Voodoo 3 at ANY resolution, and maybe even at any speed. They are quite slow in my experience unless the application is completely cache agnostic.

Some games are, but most are not.

Reply 22 of 33, by LubieCipy

User metadata
Rank Newbie
Rank
Newbie
Jasin Natael wrote on 2021-10-05, 20:57:
Well now I am confused. […]
Show full quote
LubieCipy wrote on 2021-10-05, 20:49:
Jasin Natael wrote on 2021-10-04, 21:18:

Not calling you a liar but what games did you test?

I find it hard to believe that Unreal/UT would run acceptably at that resolution on a cacheless Celeron.

I'm sure there are others as well.

Sorry for delay 😀 it took me a while to find the old tests. Games with build in timedemos (Quake 3, turok 2 etc.) scores 25-35 fps in 1024x768 and even 1280x1024. It is obvious that the CPU is a bottleneck so the results at 640x480 are very similar. I did not check UT 😀

Well now I am confused.

In you original post you say that the Covington @ 266mhz is "sufficient" for a Voodoo 3 at 640x480.
Then is this post you are telling us that it is "obvious the CPU is a bottleneck"

So which is it? Maybe I just misunderstood your first post and if so I apologize.

I'm quite certain that the Covington is a bottleneck to a Voodoo 3 at ANY resolution, and maybe even at any speed. They are quite slow in my experience unless the application is completely cache agnostic.

Some games are, but most are not.

Yes, covington is a bottleneck at any resolution when paired with Voodoo3 but also can deliver good speed for games from 1998-2000. These two things can happen simultaneously. I think that 25-35 frames per second is sufficient at standard settings (25 fps in Quake3 @1024x768 in high detail).

Reply 23 of 33, by Jasin Natael

User metadata
Rank Member
Rank
Member
LubieCipy wrote on 2021-10-05, 22:45:
Jasin Natael wrote on 2021-10-05, 20:57:
Well now I am confused. […]
Show full quote
LubieCipy wrote on 2021-10-05, 20:49:

Sorry for delay 😀 it took me a while to find the old tests. Games with build in timedemos (Quake 3, turok 2 etc.) scores 25-35 fps in 1024x768 and even 1280x1024. It is obvious that the CPU is a bottleneck so the results at 640x480 are very similar. I did not check UT 😀

Well now I am confused.

In you original post you say that the Covington @ 266mhz is "sufficient" for a Voodoo 3 at 640x480.
Then is this post you are telling us that it is "obvious the CPU is a bottleneck"

So which is it? Maybe I just misunderstood your first post and if so I apologize.

I'm quite certain that the Covington is a bottleneck to a Voodoo 3 at ANY resolution, and maybe even at any speed. They are quite slow in my experience unless the application is completely cache agnostic.

Some games are, but most are not.

Yes, covington is a bottleneck at any resolution when paired with Voodoo3 but also can deliver good speed for games from 1998-2000. These two things can happen simultaneously. I think that 25-35 frames per second is sufficient at standard settings (25 fps in Quake3 @1024x768 in high detail).

Well acceptable frame rate is objective. But in some cases yes 25fps can be acceptable. In the modern day I can't see any reason for limiting a build with a Covington unless it's just as a curiosity. Back in the day if that's all you had then by all means.

But I still find it hard to believe very many people would have paired a Voodoo 3 with a Covington Celeron.Doesn't make much sense spec wise or money wise.
My initial argument wasn't whether or not the Covington could play some games or not.....it was that it was sufficient to completely saturate a Voodoo 3. And it just isn't, at any resolution.

Reply 24 of 33, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I decided to run some of my own benchmarks in Windows 98SE w/Via C3 Nehemiah at 1.4 GHz. I have DirextX 8.1 installed. I tested the following graphics cards. The specs. are according to Everest. Vsync disabled with PowerStrip.

XFX Nvidia GeForce4 MX440 w/AGP 8x
NV18
DDR - 128 MB - 128 bit - 143 MHz (287 MHz effective) - 4592 MB/s bandwidth
CPU: 275 MHz, RAMDAC: 350 MHz
Driver version 53.04
Running at 4x

PNY Nvidia GeForce4 MX440
NV17
DDR - 64 MB - 128 bit - 166 MHz (333 MHz effective) - 5328 MB/s bandwidth
CPU: 250 MHz, RAMDAC: 350 MHz
Driver version 53.04
Running at 4x

3dfx Voodoo3 3000 AGP
SDR - 16 MB - 128 bit - 166 MHz - 2656 MB/s bandwidth
CPU: 166 MHz, RAMDAC: 350 MHz
Driver version 1.07b
Running at 2x
MiniGL v1.49 used with GLQuake and Quake2

The purpose of both MX440 cards is because the PNY is slightly faster than the XFX card and if the graphics is starved for a faster CPU, the score shouldn't increase.

GLQuake - 1280x1024x32 - nosound
XFX MX440: 86.9 fps
PNY MX440: 87.0 fps
Voodoo3 3K: 72.2 fps

Quake2 - 1280x960x32 - nosound
XFX MX440: 134.7 fps
PNY MX440: 145.2 fps
Voodoo3 3K: 55.0 fps (55.0 fps using 3dfx OpenGL, or 53.7 fps default OpenGL)

Quake 3 - 1280x1024x32 - nosound
XFX MX440: 71.6 fps
PNY MX440: 88.9 fps
Voodoo3 3K: 31.7 fps (35.6 fps if OC GPU/RAM to 185 MHz)

From these results, does it look like the Voodoo3 is a bit underpowered for the Via C3 Nehemiah at 1.4 GHz? v1.07b of the 3dfx drivers are beta, could they be slowing it down compared to the v1.07 standard drivers? The PNY MX440 is almost 3x faster with Quake 2/3. I also noticed that the Quake 2/3 scores increased with the faster MX440 compared to the slower MX440. Does this mean that the CPU is able to keep pace with the MX440 and perhaps may even benefit further from a faster graphics card?

I had planned on swapping out my MX440 for the Voodoo3, but after the Quake 3 results I am having second thoughts. Unfortunately, I don't have another AGP system that would be suitable for my spare Voodoo 3 since they all are faster than the Nehemiah. I wonder if the performance gap between the MX440 and the Voodoo3 is similar with other games on the Nehemiah? Or are their some 1999-2001 era games that would actually benefit from the Voodoo3 on this system?

I tried to overclock the Voodoo3 3000 to 190 MHz, but Quake 3 hung up half-way. It seemed to do OK at 185 MHz though.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 27 of 33, by feipoa

User metadata
Rank l33t++
Rank
l33t++
kolderman wrote on 2021-10-11, 22:41:

Try 800x600. 1280x1024 was like 4k in 1999, and the voodoo3 was not a 3090Ti.

Voodoo3 3000 SDR with Quake 3 at:

800x600 = 73.5 fps
1024x768 = 50.7 fps
1280x1024 = 31.7 fps

Keep in mind that at 3D, the Voodoo3 can only do 16-bit colour. Seems to me like a Voodoo3 is under powered for a Nehemiah 1.0+ GHz. I have too many GeForce cards in my builds and was hoping to use something else in the Nehemiah. Maybe I'll play with some of the later Matrox cards and see what happens. However if my memory is correct, I think they dumped D3D support at some point.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 28 of 33, by Bancho

User metadata
Rank Oldbie
Rank
Oldbie

feipoa, do you have a particular reason to run the C3? A particular target for the build?

The reason i put the V3 in mine, is more because i picked the Nahemiah for its slowdown tricks than its speed. With the V3 having pretty good dos compatibility it seemed like a fitting card and paired with the Power VR card, the CPU has enough umphhh to drive both cards but the flexibility to slow down for older stuff (If need be).

If I was going for outright Performace on the the platform (I run my CPU on a Slocket / Slot 1 Board) I'd just used the fastest Intel CPU Possible instead.

I guess its what you want out of the build will determine which GFX card to use.

Reply 29 of 33, by mothergoose729

User metadata
Rank Oldbie
Rank
Oldbie

I have my nehemiah build for dos/win9x and an XP build and that is what I recommend instead of building a DOS machine and a fast win9x machine. A PCI vooodoo card is nice to have, but for almost all games I prefer the geforce. If you can't have both a voodoo 3 is a very reasonable choice IMO. If it is too slow, you can almost always play it on a faster machine.

Reply 30 of 33, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I was just hoping to see what others were doing with their Nehemiah's, then I was going to formulate a plan. Loosely, I was thinking to first make the system as fast as possible with 1999+ games, and later test out some slow-down conditions in DOS. Once I had those worked out, I'd test an Ezra-T to see if such chip really provides better slow down experience and see how much of the high-end it cuts away. From what I've read on Vogons, the Ezra-T is pretty versatile with a more linear speed selection range at the low-end, but suffers on the high-end. So most likely, I'd stick with the Nehemiah with a preference for the high-end.

While the Voodoo3 is touted for its DOS compatibility, are other cards, like the Matrox G-4, -5, and Parhelia P-6, -7 series, GF4, and Radeon 9000 series that far off? Are people playing 286 era games on their Nehemiah's and Ezra-T's?

At any rate, considering how fast the Nehemiah is performing on the high-end, I was thinking this system might be a good candidate for an all-out Matrox comparison, that is, Mystique 220, G200, G200A, G200 MMS, G400, G450, G550, P650, P750, Parhelia AGP, and Parhelia PCI. Unfortunately, I am missing the original Millennium, Millennium 2, and the P690. And probably a few other less popular offerings. On all my other systems, a Matrox card only serves as a 2D host for a Voodoo2. If the Nehemiah is fast enough, it might be fun to have a Matrox-only 3D system.

The purpose of the C3 is novelty more than anything else. I remember when they were designing the Nehemiah; it was originally supposed to support SMP, but they cut that. This is what the VIA sales rep. told me at the time anyway. I had planned on upgrading my dual 440BX board with dual Nehemiahs. Since that didn't pan out, I never switched to Nehemiahs but have always been curious about them. Then with setmul and the "time machine" revelations about the CPU, I thought it would make a nice build.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 32 of 33, by mothergoose729

User metadata
Rank Oldbie
Rank
Oldbie

I do play 286 games on my setup.

For DOS compatibility, nvidia and voodoo cards are the best while still being good for 3D. I wouldn't use matrox or ATI for a DOS focused machine.

I have only found 2 dos games I couldn't play on my nehemiah at all, and there are a few games that only run in CGA composite that "work" but aren't very good. Everything else works for me using an Nvidia GPU - with the help of some TSRs here and there for compatibility.

The nehemiah can go much slower than the Ezra - T. I can get down to a faster 8088 without using mo slo, and it can do slow 386SX@20hz speeds without clock skipping.

The Ezra-T can emulate 486 speeds, while the Nehemiah has a jump from a fast 386DX@40mhz speeds to a 486DX4100 speeds, without being able to hit 486DX66 or DX33 speeds without using clock skipping or mo slo. With that said, I have only found one game that needs a 486DX66 exactly to run properly, and it doesn't appear to mind running with throttle to approximate that speed anyway.

It's magic carpet, and I personally run it at DX4100 speeds which is a bit fast but not unplayable.

People build different PCs for different purposes. Personally, I try to get the most coverage of games with the fewest computers, and my nehemiah PC + my windows XP can play almost anything in a reasonable way.

Reply 33 of 33, by Jasin Natael

User metadata
Rank Member
Rank
Member
feipoa wrote on 2021-10-11, 10:24:
I decided to run some of my own benchmarks in Windows 98SE w/Via C3 Nehemiah at 1.4 GHz. I have DirextX 8.1 installed. I tested […]
Show full quote

I decided to run some of my own benchmarks in Windows 98SE w/Via C3 Nehemiah at 1.4 GHz. I have DirextX 8.1 installed. I tested the following graphics cards. The specs. are according to Everest. Vsync disabled with PowerStrip.

XFX Nvidia GeForce4 MX440 w/AGP 8x
NV18
DDR - 128 MB - 128 bit - 143 MHz (287 MHz effective) - 4592 MB/s bandwidth
CPU: 275 MHz, RAMDAC: 350 MHz
Driver version 53.04
Running at 4x

PNY Nvidia GeForce4 MX440
NV17
DDR - 64 MB - 128 bit - 166 MHz (333 MHz effective) - 5328 MB/s bandwidth
CPU: 250 MHz, RAMDAC: 350 MHz
Driver version 53.04
Running at 4x

3dfx Voodoo3 3000 AGP
SDR - 16 MB - 128 bit - 166 MHz - 2656 MB/s bandwidth
CPU: 166 MHz, RAMDAC: 350 MHz
Driver version 1.07b
Running at 2x
MiniGL v1.49 used with GLQuake and Quake2

The purpose of both MX440 cards is because the PNY is slightly faster than the XFX card and if the graphics is starved for a faster CPU, the score shouldn't increase.

GLQuake - 1280x1024x32 - nosound
XFX MX440: 86.9 fps
PNY MX440: 87.0 fps
Voodoo3 3K: 72.2 fps

Quake2 - 1280x960x32 - nosound
XFX MX440: 134.7 fps
PNY MX440: 145.2 fps
Voodoo3 3K: 55.0 fps (55.0 fps using 3dfx OpenGL, or 53.7 fps default OpenGL)

Quake 3 - 1280x1024x32 - nosound
XFX MX440: 71.6 fps
PNY MX440: 88.9 fps
Voodoo3 3K: 31.7 fps (35.6 fps if OC GPU/RAM to 185 MHz)

From these results, does it look like the Voodoo3 is a bit underpowered for the Via C3 Nehemiah at 1.4 GHz? v1.07b of the 3dfx drivers are beta, could they be slowing it down compared to the v1.07 standard drivers? The PNY MX440 is almost 3x faster with Quake 2/3. I also noticed that the Quake 2/3 scores increased with the faster MX440 compared to the slower MX440. Does this mean that the CPU is able to keep pace with the MX440 and perhaps may even benefit further from a faster graphics card?

I had planned on swapping out my MX440 for the Voodoo3, but after the Quake 3 results I am having second thoughts. Unfortunately, I don't have another AGP system that would be suitable for my spare Voodoo 3 since they all are faster than the Nehemiah. I wonder if the performance gap between the MX440 and the Voodoo3 is similar with other games on the Nehemiah? Or are their some 1999-2001 era games that would actually benefit from the Voodoo3 on this system?

I tried to overclock the Voodoo3 3000 to 190 MHz, but Quake 3 hung up half-way. It seemed to do OK at 185 MHz though.

This is interesting. I would have really thought that a Voodoo 3 would have scaled slightly better. I always thought there was benefit to had up to about 1GHZ Coppermine/Athlon give or take a couple hundred mhz depending on the title.