G400 & KX-133 Wierdness...

[IceStorm]

It's VIA's AGP implementation. The forum's down over at www.ars-technica.com, but others have seen dropoffs in their 3D Mark scores when on VIA/AMD boards, not using Matrox cards, and the VIA boards fair worse than the AMD ones.

I've got a VIA Apollo Pro 133A (same AGP implemetation as the KX133 and all their other AGP-capable chipsets). It blows. I removed the board and put in a 440BX-based board, and my 3D Mark scores went from 2600 to 3000, and this with my PIII-550E at 733, not the 750 that the VIA board was at.

Guess we have to wait for AMD's 760, or deal with VIA's poor AGP implemetation. Or stick with Intel...

[PhilG]

Win 98 SE, Athlon 600, G400, 5.52 drivers, 800*600,32 bit.
Abit KA7, 128 Mo SDRAM PC133

Paradyse, I get AGP*4 ( pcilist) if i set the agpfallback to 1 in the registry ( section class-display-detting-cpu).
Always run perfectly, but, parfois when i reboot the system, the key "agpfallback "changes to 2 and i only get agp*1 ! ( i don't understand why)
So, i edit the registry to rechange the agpfallback key to 1.
sorry for my english

[CannyOne]

IceStorm I would have agreed with your conclusions completely before I read this thread. My impression was clearly that VIA's implementation of AGP, and specifically the AGP Driver, was the cause of the drop in performance I experienced when moving from the AMD 750 to KX133 chipset board.

However, I talked the first person, who didn't have the VIA AGP Driver installed, into installing it. His scores are now:

Now my bus xfers are:
2d = 145 [was 31]
3d = 194 [was 45]

This is with a GeForce video card, and the same motherboard as I have (an EP-7KXA). His scores are lower than mine at 2D, and much higher than mine at 3D. There's just no getting around it! Part of this difference has to be with the Matrox drivers. I'm somewhat committed to using a VIA chipset board. It's the best alternative for the Athlon at this time. I now want Matrox to alter their drivers so that I can get the best performance possible, that's all.

BTW, I also have a couple of BX board systems. I'm definately interested in the new Celeron IIs. This combination would work fine with my G400Max. Since it doesn't really support AGP 4x. But I really prefer the AMD processors. And I "darn" sure am not going to buy any "i8xx" board of any shape, size, or description. The options are limited here. None of them seem to have "everything" I want! So I'm just picking what I think is the most "viable" solution.

[anid]

I'm starting to agree more and more with Canny... If it is a chipset issue (besides the normal ones for Via, I mean - BX does tend to perform better, no surprise that the 750 does as well), it's in the framebuffer, which I don't have the skill to dig through, and shouldn't have that large of an impact anyway. Those default AGP settings are pretty conservative though... Paradizer, if you are feeling brave, try plugging F3 into register 40 and FF into register 41 (using wpcredit) and seeing what effect that has on the transfer rates. Just curiousity on my part, it should boost them, but I wouldn't imagine by much. And using Final Reality for this does seem a bit ridiculous... I think even the fillrate score in 3DMark would be more accurate.

------------------
"I wrestled with reality for 27 years and I'm happy to say I finally won out over it."

[Paradizer]

Ok guys...

Just for information: I figured out that my G400MAX was one of those lame ass edition thingies!!! The serial# was G4+MMDHA32GR and NOT G4+MMDH4A32GR wich indicates the 4x feature... So f..k Matrox for fooling their customers!

I returned it to the dealer and plugged in my old 2mb Mystique cause I just sold my FireGL 1000 to a friend :-(

Now I really gotta think bout all that again! I guess I gotta wait for the new GeForce II or those Voodoo5500 or that new ATI chip - or the G450/800...

Unfortunately I need that cool DVDMax feature or something compareable. Damn isn't there another company building as good tv-outs as Matrox???

Cheers, Para

[Greebe]

Para, first off there is no difference between the performance with a 2x or a 4x G400. Matrox did not try to fool anyone, as there were no 4x compatable MB's out when the G400 was originally released. The G400 chipset was stated to be 4x compatable not that the card could actually do it. If they said it did, then people like you would have whined that it couldn't do it, even though their MB's didn't support it. FYI the only thing that uses AGP2x/4x are benchmarks, that's it.

[Paradizer]

LO Greebe,
you got me completely wrong! I'm definitely not one of those 18-year-old framerate junkies... So please be careful using terms like "people like you" as you won't know me already
I know - as well as you do - that 4x is currently only used by benchmarks. But as I'm not going to buy me a new card every 6 month (like many others do), I just want to get the most out of my system cause I'm sure that 4x will be used in the near future.
IMHO if representetives at the CeBIT, Websites-contents and product-packages pretend that the card (not only the chip) supports 4x then it just has to support it.
In case there are two different models they should tell their users in a certain way...

But hey, although I feel a bit disapointed
bout Matrox I went in the shack a few hours later and bought me a new G400MAX (model 4A) cause those lovely thingies like DVDMax and image quality are (for me) more important then Quake III performance!

Now everything (incl. 4x) works fine under Win98SE & Win2k :-)))

For all who are interested:

Abit KA7 (RK)
AMD Athlon (model 2) 700@800 (34-37°C @ 1,7V)
AGP aperature = 128MB
AGP 1 Waitstate (r/w)
256MB Mushkin PC133 DRAM @ 114MHz (CAS/RAS 2)
VIA 4-in-1 4.2 (standalone AGP used)
"Patched" machine.inf under W2k
G400Max (4A with 1.5.022) with 5.52 and 5.03
SB Live! Player
IBM DPTA 20,5 GB (7200 upm / UDMA66)
Maxtor Diamond MAX 5,1
Pioneer 104S DVD
AVM ISDN PnP (ISA)

Thanks for all your nice support guys! We'll meet again...

Any further questions are welcome!

[ID]

I've noticed some more things that (apart from demonstrating my lack of AGP knowledge) are a bit strange:

Once the VIA chipset "patch" (Epox's VIA_INF.ZIP) is applied to Windows 98, it makes no difference whether the VIAGART.VXD (Via AGP Miniport) is installed or not. All the programs that I have that can recognise AGP still recognise it.

The VIAGART.VXD contains the following text:
"SOFTWARE\Matrox\PowerDesk\Current Settings" followed by "Bus Mastering".

The Matrox driver installation installs it's own version of VGARTD.VXD (the Intel equivalent of VIAGART.VXD).

The actual "GART" functions are contained within a driver called PCIMP.PCI (the Windows 9x PCI miniport driver), which is used as the driver for the VIA CPU-AGP controller.

AMD provide a version of PCIMP.PCI for their chipset called AMDMP.PCI which contains the same "GART" functions.

The only thing that all of this shows really is that the VIA AGP miniport driver is "aware" of Matrox cards, and that the Matrox drivers are expecting VGARTD.VXD to be the driver for the AGP port.

Renaming the VIAGART.VXD to VGARTD.VXD and hacking the file so that all references to VIAGART were replaced with VGARTD was not the magic fix that I hoped it would be...

As far as the benchmarks go... Final Reality is the only benchmark I know of that measures bus transfer speeds. Measuring Fill-Rate on a more recent benchmark will not indicate bus transfer speeds. As CannyOne mentioned earlier, it does not matter whether the figures that FR produce are accurate or not. If I run EXACTLY the same hardware (apart from the Motherboard) and the same set of drivers (apart from the chipset) through FR then the comparison between the results is valid. If suddenly the bus transfer speeds are 1/4 of their previous values then this indicates a problem with the interaction between the drivers, the MB & the chipset, not a problem with FR. Especially as running FR on both Motherboards with a Voodoo2 gives similar results (the VIA results are slightly higher, probably due to the 133Mhz memory clock). More importantly, the lower performance of the G400 is across all 3D applications (including G400 Tech. demo). Using FR has helped in as much as it has indicated a possible cause for the slowdown.

I've posted this topic on the Matrox technical support forum twice. The first time nobody answered it at all. The second time, I got one vague "chipset identifier" response from one of their tech. support people.

[Rant begin]
I really hope that VIA, Matrox (or even Microsoft?) resolve this problem soon. I've spent far too much time on it already...

We the customers should not have to solve these problems.
[Rant end]


[This message has been edited by ID (edited 09 May 2000).]

[CannyOne]

Yeah, I agree Ian. And it definately is a Matrox driver issue. I've got a number of other people to report scores to me. All their scores are at least 2x to 3x higher than with a Matrox card.

At the same time there is most likely an issue with the VIA AGP driver as well. Anyone who has been keeping up with the Win2k mess is aware that VIA have not been on the level with their customers either. (For those that don't know - they said Win2k had built in support for the KX133 - in truth it only really has support for the much older MVP3 Super Socket 7 chipset.)

So the bottom line is both companies share responsibility for our situation! If they don't work it out soon. I may not buy products from either ever again.

[Targon]

For the past 2+ years, people have said that VIA has a slower AGP implementation than Intel. Time after time, people have said that VIA can make a chipset that is OK, but has a POOR AGP implementation. This is true of ALL VIA chipsets. The fact that AMD, who is not even a real player in the chipset business can do a better job. Now, the real question is if ALi will be able to produce a chipset for the Athlon(Thunderbird/Duron/Mustang), that will have all the features of the KX133 chipset, but with a good AGP implementation. The fact that it is supposed to support PC266 DDR SDRAM should also prove to be interesting, but I just want the best motherboard to go with my G400 MAX, and when it comes out, the G800.