So it looks like UT is very CPU bound on this K6-2 rig. Thanks for doing that, it's good to see the UT can run will on the old Voodoo card, even if that means medium detail. 😀

I wonder if one of the 6MB Voodoo cards would run well on high?

I personally had no problems with UT on a k6-2 w/ Voodoo2 on high back in the day. 😉

It's the poor MMX implementation on K6. One game unplayable on K6-2 but well playable on a Pentium II is Serious Sam

Yeah, it seems like K6+Voodoo is the best combination for UT. The Glide driver seems to be way better than the D3D or OpenGL drivers, not just visually, but in performance too.

Found this old article, which seems to say exactly that: http://www.anandtech.com/show/647/12

Of the stock game choices, Glide is the best choice. S3 Metal is a close second though from what I've seen.

But today UTGLR is without a doubt the best option (either its D3D9 or OpenGL renderer). It will work with cards as old as GeForce3 and Radeon 8500.

UTGLR 3.4 gave me about 4-5fps extra using the FX5200. 3.5 and 3.7 don't work on Windows 98SE. Not bad at all. 😀

Edit: It think my Voodoo card might have a fault. If I install the Voodoo driver 3.01.01, any Glide, D3D or OpenGL title causes the screen to go black. If I swap the video cable back to the 2D card, I find that Windows has bluescreen'd.

When I downgrade to 3.00.01, things work, but Q3A has totally messed up colour. 3DMark99, UT, and GLQuake all work fine, but Q3A exhibits the issue.

Weird huh?

Theres also a Direct3D 10 renderer for unreal.

leileilol wrote:

I personally had no problems with UT on a k6-2 w/ Voodoo2 on high back in the day. 😉

It's the poor MMX implementation on K6. One game unplayable on K6-2 but well playable on a Pentium II is Serious Sam

Didn't the K6-2 dramatically improve MMX performance over the K6 classic (not sure if the III or + models did any better for MMX).

Depending on the benchmarks, even the K6 classic seems to fare well against a PII of similar clock speed . . . at least if this is any indication:
133 MHz Challenge - 5th/6th gen CPU per clock performance
or The "Not a 486 anymore, sadfase" 6x86 Build

tgod wrote:

Theres also a Direct3D 10 renderer for unreal.

I don't like how that renderer forces the FOV to 100. I play at 90. Also its lighting is too dark.

kool kitty89 wrote:

Didn't the K6-2 dramatically improve MMX performance over the K6 classic (not sure if the III or + models did any better for MMX).

Depending on the benchmarks, even the K6 classic seems to fare well against a PII of similar clock speed . .

K6-2 scales very badly with clock speed because of the memory subsystem. Both the main RAM and the L2 cache are really slow on Super 7 boards compared to a typical Intel-based Slot 1 setup. But even with on die L2 there's still a significant deficit to P2/P3 in these games which is probably partly caused by FPU and main memory performance.

Quake 3 displays similar disparity between Intel and AMD K6-x. This game really benefits from high performance RAM and caches. Watch P4 Willamette stomp Athlon.

Unreal does use MMX a lot, specifically for audio and the software renderer. I too thought AMD was similar to Intel here but maybe leileilol knows more.

Trident 3DImage 9880 "Blade3D" 8MB AGP
http://www.youtube.com/playlist?list=PL995DE461767BFB71

It's surprisingly functional. Failure on the fog table support though.

More PowerVR OA stuff, but this time running a timedemo on the included demo. It forces lightmap (no vertex) and low picmip and such when you play back the demo so it gets a more intended performance measurement. It doesn't handle it very well without vertex lighting.

I recorded it earlier, but I uploaded it as an April Fool's joke since the performance was so pathetic. The benchmark took 18 whole minutes.

OA 088's kind of sizey (near 400mb) so it's not exactly a comfortably small thing to benchmark (I hate the size issue myself). It does makes use of detail textures and specular tho so it's quite more intensive than Q3, almost close to Unreal texturage

swaaye wrote:

K6-2 scales very badly with clock speed because of the memory subsystem. Both the main RAM and the L2 cache are really slow on S […]
Show full quote

kool kitty89 wrote:

Didn't the K6-2 dramatically improve MMX performance over the K6 classic (not sure if the III or + models did any better for MMX).

Depending on the benchmarks, even the K6 classic seems to fare well against a PII of similar clock speed . .

K6-2 scales very badly with clock speed because of the memory subsystem. Both the main RAM and the L2 cache are really slow on Super 7 boards compared to a typical Intel-based Slot 1 setup. But even with on die L2 there's still a significant deficit to P2/P3 in these games which is probably partly caused by FPU and main memory performance.

Quake 3 displays similar disparity between Intel and AMD K6-x. This game really benefits from high performance RAM and caches. Watch P4 Willamette stomp Athlon.

Unreal does use MMX a lot, specifically for audio and the software renderer. I too thought AMD was similar to Intel here but maybe leileilol knows more.

Do all the SS7 boards/chipsets have such inferior main RAM performance compared to Slot1/S370 boards (be it Intel or VIA, or other) when comparing equal FSB speeds (ie 100 vs 100)?

Or is it just more an issue with specific games and drivers that heavily catered to Intel chips/chipsets?

And when you say Athlon, you mean pre-DDR, right?

kool kitty89 wrote:

Do all the SS7 boards/chipsets have such inferior main RAM performance compared to Slot1/S370 boards (be it Intel or VIA, or other) when comparing equal FSB speeds (ie 100 vs 100)?

Or is it just more an issue with specific games and drivers that heavily catered to Intel chips/chipsets?

And when you say Athlon, you mean pre-DDR, right?

Athlon XP vs. Northwood is similar.
http://www.tomshardware.com/reviews/pentium-4,407-12.html

Regarding K6, I don't know if some chipsets were better than others from a memory performance standpoint.

The most aggressive K6-2 optimization effort I know of is AMD's Quake2 3DNow patch that also includes a Voodoo2 3DNow ICD. I think I remember that 3dfx and NVIDIA were the most aggressive with 3DNow driver optimizations.

For the fans I made another run with PowerVR:
https://www.youtube.com/watch?v=SEfwxINAX7o

This time I disabled UglyRefresh which was kind of VSync disable for PowerVR, so the previous recording had rather strong tearing.

Just look at the framerate if it comes to close combat, especially using the rocket launcher. It goes close to 10 fps. I guess this is one reason why PowerVR SGL got unsupported in later patches of UT.

i've supsected the reason for the sgldrv regression are the UI bugs and the skeletal mesh additions

http://www.youtube.com/playlist?list=PL995DE461767BFB71

Parhelia (original AGP 4x model)
-irritating fan that's much louder than a G400 Max fan.
-Doom3 is bugged and quite slow.
-3DMark shows Radeon 8500 and 5200 Ultra to be faster.
-massive multitexturing fillrate result in 3DMark (so what?)
-no fog table support for DX3-5 games
-Didn't bother with FAA because it's slow without AA.

Matrox specified a 10% tolerance in their clock speeds. As a result, core clocks from 198-220 MHz are possible. This one is 207 MHz.

Permedia 2 (Diamond FireGL 1000 Pro 8MB AGP)
-I'd rather use a Riva 128. 😉
-missing features as shown by 3DMark99, Quake2 and Quake3.
-slow 3D
-very nice VGA signal from the Diamond card

http://www.youtube.com/playlist?list=PL995DE461767BFB71

Do you think a Permedia 3 would do any better in benchmarks and games?

sliderider wrote:

Do you think a Permedia 3 would do any better in benchmarks and games?

I looked up Permedia 3 results the other day and decided to not buy a card
http://www.sharkyextreme.com/hardware/reviews … _oxygen/e.shtml

It's surely an improvement over Permedia 2, but TNT2 Ultra blows it away.

elianda wrote:

Powerstrip says: My Rage IIC AGP 62 MHz (no info if core or memory), does 147 3DMark99 Rage Pro Turbo PCI Core 75 MHz, Memory 10 […]
Show full quote

Powerstrip says:
My Rage IIC AGP 62 MHz (no info if core or memory), does 147 3DMark99
Rage Pro Turbo PCI Core 75 MHz, Memory 100 MHz.

The Rage Pro feels at least 10 times faster.

Somehow I missed the post of your Rage II and Pro recordings back in March, but you're not kidding when you say the Rage Pro is faster. 😉

The Rage II you tested seems to be fairly comparable in speed to your S3 Trio 3D with a few differences in bugs/driver issues in those benchmarks. The Virge seems to have an advantage in visual quality in some areas (including filtered translucent textures), but seems to be similar in general. Albeit, I'd imagine a considerably greater difference for drivers/applications properly supporting 32-bit rendering on the Virge -as with Incoming. Both also have the issue of lacking OpenGL support. (or does the Rage II have OpenGL drivers?)

On another note, (when working properly) both the Rage II and Trio 3D seem noticeably faster than the Rendition Verte V1000. (though still significantly slower than the Vooodoo)

The Rage Pro recordings seem to show similar performance to the Riva 128, but with different artifacts (seaming and Z-fighting seem to be much less prominent though). The dithering is much less noticeable than the Rage II and maybe the Riva too. (it's hard to notice any dithering in some of the videos)

The helicopter sequence in 3DMark 2000 also seems to have different ground textures (all green rather than brown/tan areas) unlike the other captures. (including the Rage II) This seems rather odd.

One has to wonder if S3 had put effort into porting games to S3D if Virge would have been more successful in the eyes of gamers. Verite V1000 is almost as useless for D3D and OGL as Virge-arch chips after all, but Rendition put a lot of effort into getting popular games supported.