AGP is 66Mhz, PCI 2.1 is 66Mhz, so there is no real Difference beetween Trio AGP and PCI.
The Trio 3D AGP is no AGP Chip. It does not even use a AGP Feature.

But considering PCI being shared, should it mean lower cpu usage? The Trio3D/2X version did not use it? Anyway even compared with other cards (Savage4, G200, Voodoo3..).

http://www.anandtech.com/show/37

matze79 wrote:

AGP is 66Mhz, PCI 2.1 is 66Mhz

66MHz support is optional, and I believe many consumer boards do not implement it. Also, it can only work if all devices on the PCI bus support it.

In the benchmark linked the difference is really thin. I think I could make an all PCI setup for my K6-2+.

Careful, those benchmarks are comparing an AGP ATi card agains a PCI Matrox. One could conjecture that the 2D core of the Matrox is just all around better - the Rage Pro really had no excuse to lose to the Millenium on ANY test based on bus speed alone. The intended resolution is also very important, as the bus only makes a big difference once it becomes the bottleneck, and lower reolutions might not get close to saturating the PCI bus when it comes to 2D workloads.

matze79 wrote:

AGP is 66Mhz, PCI 2.1 is 66Mhz, so there is no real Difference beetween Trio AGP and PCI.

AGP can do 2x signaling though, so if the chip supports it (which Trio doesn't afaik), even AGP 1.0 is twice as fast as PCI 2.1 (533 MB/s instead of 266 MB/s theoretical bandwidth).

Comparing the same card on the AGP bus vs. the PCI bus would provide more realistic insight to your question. For example, why not compare a Matrox G200 PCI vs. a Matrox G200 AGP? Which 2D benchmarks are you interested in?

Most early AGP/PCI chips do not use AGP features, in other words, they use AGP slot as if it was a PCI slot. I seriously doubt you'll ever see any significant difference here. Btw, the two cards to test should have equal chip and memory frequency, equal memory bandwith, same memory type, otherwise it would be pointless.

Voodoo3 comes to mind. 2000 and 3000 versions came in both PCI and AGP. It would be nice to do some testing while saturating the PCI bus.

RacoonRider wrote:

Most early AGP/PCI chips do not use AGP features, in other words, they use AGP slot as if it was a PCI slot. I seriously doubt you'll ever see any significant difference here.

Well, I wonder... the chipset could still make a difference, if it somehow prioritizes AGP over PCI. If AGP has a dedicated channel to the CPU, where PCI is shared with all cards. But I'm not sure how chipsets implement this. All I recall is that my Matrox PCI cards could get over 120 MB/s transfer in practice, so at least on my PCI board (Intel Triton 430FX chipset), there wasn't a lot of 'loss' to other devices on the PCI bus in practice.

feipoa wrote:

Comparing the same card on the AGP bus vs. the PCI bus would provide more realistic insight to your question. For example, why not compare a Matrox G200 PCI vs. a Matrox G200 AGP? Which 2D benchmarks are you interested in?

Yeah certainly with the same chipsets the tests would be nice. My point was to understand if for brosing/office/desktop stuff, a PCI card of that era (1997-2000) with a chipset/cpu of the same era, could be basically equal to an AGP one in terms of speed. Until when 2D speed advantage from the video cards has been better or worse in later cards?

The CPU is a bottleneck too. Especially for high resolution DOS games. But even with Windows GUI acceleration. Win9x GUI acceleration is also less robust than that of NT5. And GPUs of ca. 2000 have more GUI acceleration functions for NT5. Lots of factors to consider.

This is something that I have been meaning to fully characterise, however I will not get to is for some now to come. I'm lost in PGA-132 land at the moment (determined to get the Ti486SXL functional on non-supporting motherboards). However, as swaaye pointed out, it would be good to cross reference this over a variety of operating systems. It would also be interesting to see if some MB chipsets with PCI only prioritise graphics PCI differently to those chipsets which have a PCI and AGP bus. If I were testing this capability, I might test using a 440BX, 430TX, MVP3, and ALi Aladdin V. Perhaps test DOS, Win95c, Win98SE, NT4, and W2K. I'd need to fully research which 2D tests to run, perhaps Tom2D along with some of the Ziff-Davis 2D benchmarks. Any 2D games which have a benchmark feature? I'd probably test the Matrox G200 PCI/AGP, Voodoo3 3000 AGP/PCI, and I should have a GF2 PCI/AGP or GF4 PCI/AGP.

I will try to do some test on my WinME setup (500 K62+/VP3) with some Matrox cards (Millenium, Mystique PCI) and some AGP (G200, Savage4, V3).
On a "feeling" test, I would say that beetwen these cards and the Savage4/V3, Windows GUI seems to be more reactive on the AGP side but no numbers to confirm right now.
Strangely even 2D browsing smoothness seems to be better on WinME that on Win98 even if 3D numbers on benchs were lower last time I tested.

You could try transplanting USER32.DLL, GDI32.DLL, USER.EXE and GDI.EXE from Millenium to 98SE. I believe 98SE2ME does that (among hundreds of other things).

Interesting, I'll try that. It's a miracle I got to make my USB PCI card to run with the VP3 (but only with no DMA on second IDE..). I always had crash problems with same setups.

Windows ME also replaced a bunch of the built in VxD drivers with newer WDM ones, so depending on your exact hardware setup, ME can be more stable.

It really does depend a lot on the operating system. I never could see much of a difference in 2D speed under Windows 98. Even at relatively high resolutions like 1600x1200 modern PCI cards are nice and quick under Win98.

XP is a different story. At 1024x768, you can see a slight difference in 2D speed between PCI and AGP (I tested this with a Radeon 9250 PCI and 9800 Pro AGP on a 2.66GHz P4 system). Bump the resolution up to 1920x1200 and the difference between PCI and AGP is like night and day. For example, when you reload a web page, you can see the PCI card refresh the screen from top to bottom. With AGP, it all appears on-screen at the same time.

So at least under WinXP, high resolution 2D really does suffer. I found that reducing the color depth to 16-bit helped the PCI card immensely, but it was still visibly slower than AGP.

Now, if you really wanna see some slow 2D, try running running Win7 at 2560x1440 on a PCI card. I tried that with a Radeon X1550 PCI on a 3.33GHz Core 2 Quad system. Despite the quick CPU, performance was very much in the crapper--it couldn't even handle plain old DVD smoothly. Even simple stuff like window drags were choppy. The C2Q board was designed for PCIe video cards, so its plain old PCI performance may have been a bit lower than that of an older system.