DN3D looks vsynced on 8400 and both ATIs.

Is that with or without fastvid? (Pretty important for Quake and Duke.)

I did not use fastvid.

clueless1 wrote:

I did not use fastvid.

I would use fastvid, it improves performance quite a bit.

MrEWhite wrote:

clueless1 wrote:

I did not use fastvid.

I would use fastvid, it improves performance quite a bit.

I recall Phil (or maybe someone else) explicitly not using fastfid in their benchmarks for some reason. Does it not support all graphics cards, thus giving some cards an unfair advantage over others? Maybe it's a tool that not everyone is aware of, so it's better to benchmarks without so that the playing field is more even...

I would think fewer variables and "helpers" would make for cleaner results. But now I'm curious enough I may rerun the numbers.

As it was CPU-limited perhaps you should repeat the experiment with a faster CPU.

PCBONEZ wrote:

As it was CPU-limited perhaps you should repeat the experiment with a faster CPU.

That's as fast as I've got. 😀

I've got a K6-2 550, but no AGP slot.
I've got the Celeron 333 system (that's the fastest cpu the board takes) with both AGP and PCI.
I've got a P133 with no AGP.

I've thought of maybe picking up a Deschute PII-333 in the hopes that it would run on this board. Not sure how much the extra L1 would help over the Celeron, though, or whether for sure that cpu would run in this board (it's an OEM HP-ified Asus P2L98-XV).

Then I suggest looking for maybe a BX chipset board and a PIII 500-600 or so.
The Celeron 333 should also work in the BX board for a couple of chipset comparisons to your LX board.

An AGP board for the K6-2 550 would also be interesting.
.

PCBONEZ wrote:

Then I suggest looking for maybe a BX chipset board and a PIII 500-600 or so. The Celeron 333 should also work in the BX board f […]
Show full quote

Then I suggest looking for maybe a BX chipset board and a PIII 500-600 or so.
The Celeron 333 should also work in the BX board for a couple of chipset comparisons to your LX board.

An AGP board for the K6-2 550 would also be interesting.
.

I was just trying to share results with the community here with what I have. It's not something I'm going to spend money on to remove bottlenecks. At the very least, the Doom results are pretty relevant with a decent range of graphics cards represented. When I get a chance, I'll update the numbers with fastvid, and if a faster board happens to fall in my lap down the road, I'll be happy to redo the tests as well.

Thanks!

I tested fastvid with one video card (Intel i740) and it did make a significant difference in a few (not all) benchmarks, so I'll make some time to get the rest of the cards retested. Is it pretty standard procedure to use fastvid when reporting benchmark results? What if the system in question does not have a cpu that is compatible with fastvid?

Here are the before and after results on the i740:
3dbench--189.5/239.7
pcpbench--no difference
doom--80.0/90.1
quake 320x200--no difference
quake 640x480--15.6/30.2
duke3d--no difference in any resolution

FASTVID basically does the same thing most Windows drivers do and sets a MTRR to boost video memory access speed for SVGA modes. If you can't run it on a CPU later than a Pentium Pro (older CPUs do not support that feature anyway), you can try to run stuff in Windows 9x and see if they run faster.

FASTVID works on the Celeron 333, but it doesn't seem to work on my Pentium I 133. Does that make sense?

Yes, Anything below Pentium Pro does not support MTRRs. Those are used to control how to cache certain ranges of memory and, in Intel processors, is only supported on P6 and later families.

I've updated the results in the first post to include results with FASTVID. Thanks for making me aware of it! 😀

clueless1 wrote:

I tested fastvid with one video card (Intel i740) and it did make a significant difference in a few (not all) benchmarks, so I'l […]
Show full quote

I tested fastvid with one video card (Intel i740) and it did make a significant difference in a few (not all) benchmarks, so I'll make some time to get the rest of the cards retested. Is it pretty standard procedure to use fastvid when reporting benchmark results? What if the system in question does not have a cpu that is compatible with fastvid?

Here are the before and after results on the i740:
3dbench--189.5/239.7
pcpbench--no difference
doom--80.0/90.1
quake 320x200--no difference
quake 640x480--15.6/30.2
duke3d--no difference in any resolution

I'm surprised this does anything for DOOM and nothing for Duke3D. I would have expected it to be the other way around.
MTRR's also exist in some K6-2 models and in K6-3. You might need another Tool for that. Athlon also have MTRR's, fastvid may work there.

idspispopd wrote:

I'm surprised this does anything for DOOM and nothing for Duke3D. I would have expected it to be the other way around.
MTRR's also exist in some K6-2 models and in K6-3. You might need another Tool for that. Athlon also have MTRR's, fastvid may work there.

I had no expectations, because I didn't know anything about fastvid til it was mentioned in this thread. My understanding now is that it's supposed to improve SVGA performance more than VGA (I guess because there is more headroom for performance increases?)

Did you look at the rest of the scores in the original post? The i740 you quoted was one of the cards that showed the least improvement with fastvid.

I wish I had an explanation for why it does nothing in Duke3D. Anyone else care to try on their system and confirm? Does anyone have anything similar to a Celeron 333?

But a quick summary:
-the ATI Rage 128 Pro is easily the best VGA card, but is also one of the worst in SVGA.
-the Intel i740 is tied with a few nvidias for fastest SVGA (though the i740 is one of the slowest in VGA)
-the GF2MX and TNT2 M64 appear to be the fasted overall and best combination of VGA and SVGA performance.
-something is clearly not right with the 8400GS. Perhaps because it is so new compared to the others. Maybe nvidia's VESA implementation by the time this came out cut some corners.

clueless1 wrote:

I had no expectations, because I didn't know anything about fastvid til it was mentioned in this thread. My understanding now is that it's supposed to improve SVGA performance more than VGA (I guess because there is more headroom for performance increases?)

The higher the resolution, the more bytes have to be moved per frame, so there is more potential for performance gains. Also VESA 2.0 modes have a higher bandwidth ceiling than vanilla 320x200 VGA. (VESA modes can reach the theoretical maximum of 133MB/s while vanilla VGA modes will never be close to that. Not that it's necessary: One frame of 320x200x256 need <64k, so for 100fps you'd just need <6.4MB/s plus more so the CPU has cycles left to actually run the game and render the image , not exactly demanding.)

Did you look at the rest of the scores in the original post? The i740 you quoted was one of the cards that showed the least improvement with fastvid.

According to your results, DOOMs runs faster on all cards with fastvid, but the difference is quite small for all of these.
And you did write that fastvid doesn't make a difference for Duke3D.

...
-something is clearly not right with the 8400GS. Perhaps because it is so new compared to the others. Maybe nvidia's VESA implementation by the time this came out cut some corners.

Exactly. After a while extreme DOS performance wasn't considered necessary anymore. In fact I remember a post which said that the DOS performance of Geforce cards went down after GF3.

idspispopd wrote:

And you did write that fastvid doesn't make a difference for Duke3D.

Yep. I'd guess that it's because Duke is CPU-bound in all the VESA resolutions, but what about the ATIs? Based on the rest of their SVGA scores, maybe poor VESA implementation.

I had trouble using FastVid with VGA range memory, Commander Keen started to mess up.
Currently use MTRRLFBE by rayer for this. Less fancy name then FastVid, but better functionality. (Requires 'DPMI server' file also on that page.)
The command line would be 'mtrrlfbe lfb wc', This way it won't affect VGA range, just VESA.

Setting LFB Write combining for VGA range is kinda useless, because all systems fast enough to support LFB Write combining can run the low resolution modes at 60 FPS anyways.

How do you guys bench Duke 3D? AFAIK it does not do a timedemo, just a nervous -DNRATE display at the bottom right.