AGP8X card stuck at AGP2X bandwidth on AGP4X motherboard

[lafoxxx]

Hi.
GA-6VTXE-A, latest BIOS from their website
ASUS V9280 (AGP 8X)

BIOS: AGP Mode set to 4x

Win98/XP: In Adapter properties, it's 2x.


Quick googling shows that the issue might be related to Chipset. Some NVIDIA drivers allegedly see VIA chipsets and force the card to use AGP2X mode. It cannot be changed in RivaTuner.

Before I try with different board (TUSL2-C), anyone had similar issue? Should I try other Drivers? Which ones?
Or I won't notice speed difference at all, so I shouldn't worry about it?

[STX]

See https://www.vogons.org/viewtopic.php?f=46&t=69289.

[liqmat]

...and viewtopic.php?f=46&t=68860

a few people have had good luck with the VIA 4-in-1 v4.43 chipset driver as you will see in that thread.

[Bige4u]

+1 for v4.43 via chipset driver.

[lafoxxx]

Thanks.
The funniest thing (sorry for not mentioning before!) is -- VIA 4in1 driver version 4.43 (only version I could find) is what I installed initially, after installing Win98, among with other device drivers.
Maybe I installed them in wrong order? But everything seems to work fine.


Looks like I found the required checkbox -- it was located in very unusual (for me) location of the RivaTuner utility. What I initially thought to be the required setting was wrong parameter.
Now AGP shows 4x, as it should.

Thanks again!

[PARKE]

Would you be so kind and run a couple of benchmarks (like for example 3Dmark2001) to see the difference between 2x and 4x ? Thanks in advance.

[lafoxxx]

Not sure if the difference is statistically significant, but there's slight increase when using AGP4X mode.
Can perform 4 more runs if you want.

[PARKE]

Thank you.
Repeating is not necessary, your outcome supports other observations.

I have just run a somewhat similar configuration:
XFX ti 4200 // ASUS P3V4X // 1GHz 133 PIII SECC // 3DMark2001 // 1024x768x32
and the outcome for AGP 4x (set in Bios) = 6698 versus AGP 2x = 6696

Your outcomes may be consistently a tiny bit higher in 4x mode but the difference is nothing to write home about even it is not based on statistical error.
Anand ran an article on this subject in 2000:
https://www.anandtech.com/show/556/4
but it did not include any Gf4 ti cards so I was curious if you could produce a more substantial difference.

[lafoxxx]

This card might be severely bottlenecked by my CPU and Chipset... If I had at least TUSL2 and 1400 MHz Tualatin the results would have been much, much better.

I mean really, there is some guy on eBay selling GeForce2 Ti, and their result is almost 2 times better than mine when I tested my GF2 Ti (they have Athlon XP). And I highly doubt they specifically used Driver version 4270 -- I installed it because according to this page, this version shows best result in 3dMark2001 according to some comparison I found online:

Just look -- these results are almost 2 times better on faster CPU (Pentium 4 2.8 )!
Full article: https://overclockers.ru/lab/show/15515/ ... nogo_opyta

So even simply changing Driver version would (in theory) benefit more than moving to AGP4X.

[PARKE]

Here two 3DMark2001 results that are somewhat in line with what you want to know:
Gainward ti 4800 // ASUS TUSL2-C // 1400 MHz Tualatin S // 3DMark2001 // 1024x768x32 // 8516
Gainward ti 4800 // ASUS TUSL2-C // 1400 MHz Tualatin S // 3DMark2001 // 1024x768x32 // 8517

Athlon XP is in a different league compared to Pentium III, I think,

[PARKE]

So even simply changing Driver version would (in theory) benefit more than moving to AGP4X.

PS
The first outcome of the ASUS P3V4X combi that I posted was done in Windows XP with driver 41.09 and I do not think the differences between driver versions qualify as significant in many (most?) cases.
The difference between the fastest and the slowest in the graph that you posted is less than 10% in 3DMark2001

[The Serpent Rider]

Difference between modes is more profound when AGP texturing required (not enough video RAM).

[PARKE]

Difference between modes is more profound when AGP texturing required (not enough video RAM).

With what type/period of software/games does that become an issue ?

[swaaye]

Really I think that from a practical, gaming standpoint that there is little difference between AGP 2x and 4x. Synthetic tests can certainly tell you other things though.

If a game starts to use a lot of AGP texturing it might cause stuttering even with AGP 8x so you'd probably just want a video card with more local memory. 128MB is already quite a bit for games you'd run on a P3.

One thing that does use A LOT of AGP bandwidth is ATI's software-based Truform (anything not Radeon 8500/9100). While the 8500/9100's hardware Truform is a technology meant to reduce bus traffic with more complex geometry, the software based emulation on other cards does the opposite. This will be slower if you are at AGP 2x. But it is also extraordinarily CPU limited so it won't matter with a P3. It will be too slow regardless. You need Core 2 to beat the CPU limits.

It's also possible that you might run into a bottleneck in special cases with a Glide wrapper. Some situations use a lot of bus traffic. It's pretty rare though from what I understand and you might need PCIe to get around that entirely anyway.

[liqmat]

Here two 3DMark2001 results that are somewhat in line with what you want to know:
Gainward ti 4800 // ASUS TUSL2-C // 1400 MHz Tualatin S // 3DMark2001 // 1024x768x32 // 8516
Gainward ti 4800 // ASUS TUSL2-C // 1400 MHz Tualatin S // 3DMark2001 // 1024x768x32 // 8517

Athlon XP is in a different league compared to Pentium III, I think,

Interesting. I have that Chinese Gigabyte GA-6VTXE-A ripoff board that uses the infamous VIA 694x(NB), VIA686(SB) chipset family that I bought from Alibaba. The Gainward 5950 Ultra I have is only benching 8900(ish) in 3DMark2001 @ 2x which seems slow considering your 4800 is getting very close to that. Also using the same CPU as you. What video driver revision are you using?

Edit: Duh, I just got to thinking the 5950U really should be paired with a P4 class CPU to unlock its potential.

[The Serpent Rider]

Define "potential". For 3DMark2001 you need Core 2 Duo level or better to fully load FX 5950 or Radeon 9800.

[liqmat]

Define "potential". For 3DMark2001 you need Core 2 Duo level or better to fully load FX 5950 or Radeon 9800.

Neat.

[agent_x007]

^AGP 8x with OC'ed Pentium DC E5800 (unless GPU doesn't support 8x).
Here's my ~6,5k Radeon 7500 score :

4600 Ti and ~15k result :

[PARKE]

8><CUT
Interesting. I have that Chinese Gigabyte GA-6VTXE-A ripoff board that uses the infamous VIA 694x(NB), VIA686(SB) chipset family that I bought from Alibaba. The Gainward 5950 Ultra I have is only benching 8900(ish) in 3DMark2001 @ 2x which seems slow considering your 4800 is getting very close to that. Also using the same CPU as you. What video driver revision are you using?
8><CUT

The driver used for that benchmark was 45.32. At an earlier moment it ran on 77.72 but the 3DMark 2001 result was some 500 points lower.

[The Serpent Rider]

Later drivers removed some clever "optimisations" in the Nature test.