brassicGamer wrote:

- How significantly (or not) certain BIOS settings can affect performance (a bit of demystifying this area too).
- At what point does a motherboard or CPU become a potential bottleneck for AGP cards?
- Which cards overclock best and is there a way to predict what other untested cards this might apply to?

-For cards of this era setup should be moot point. Just check if all relevant/possible AGP features are on. There might be strange things happening, but the encounter is really hard to predict. Common sense reflection upon results does wonders imho. Plenty programs can log configuration for you, so save it along with benchmarks results.
-At any point, speaking precisely. You may reach safe range of statistical errors, but the way rendering is done there will always be CPU-only time.
-Just no. Cards differ, even each chip.

I understand your motivations, but please try to be more creative with benchmarks. People did this before. Scientific is big word.

Putas wrote:

I understand your motivations, but please try to be more creative with benchmarks. People did this before. Scientific is big word.

Well, my main motivation for asking in the first place was that I was worried I might be wasting my time, so I'm glad I asked!

I want to see this. Sounds like a valuable effort.

One little bit of data I'd like to know about those cards is their stability with an overclocked AGP bus. Motherboards lacking proper dividers can run out of spec, in particular the 440BX chipset at 133MHz system bus runs AGP 2X at 89MHz.

I forgot to add the benchmarking process was actually just supposed to be a means to an end - once I had done that bit, I was then going to identify the bottom, top and middle performers and then use these to tweak lots of other stuff including cache, RAM, CPU, mobo and see if any of the cards behave significantly differently in other circumstances. I guess what I was hoping for was to discover something weird that couldn't be predicted.

firage wrote:

I want to see this. Sounds like a valuable effort.

One little bit of data I'd like to know about those cards is their stability with an overclocked AGP bus. Motherboards lacking proper dividers can run out of spec, in particular the 440BX chipset at 133MHz system bus runs AGP 2X at 89MHz.

Good one! This is something I could add to the second phase, thanks.

Edit: I have a number of PII boards but none of them do 133Mhz, unless I can find some undocumented jumper settings. I do, however, have an SS7 board that I can overclock to 120Mhz by using a K6-2. This should simulate the same effect.

I agree, this would be very useful. If you're going to the trouble of testing all this, it would be great to see some tests done in pure DOS as well. For example, do the cards have VESA 2.0 BIOS or is the BIOS only VESA 1.2? Do they support all the VGA registers or are there features missing? Is their EGA and CGA compatibility good? e.g. I've seen cards that don't support setting the VGA border colour, or are very jerky when scrolling in EGA modes, or output a fixed resolution like 800x600 and unevenly upscale lower video modes to that.

It'd be good to pick a few test games (maybe even some known difficult demos that use obscure video card features) running natively in DOS and test for these sorts of things as well, so your results are relevant for even more people.

My interest would be because I'd like to use a 440BX-era system for DOS gaming, but with a video card having DVI output so I don't get such a blurry signal if I run it a long way from the box into a projector or HDMI TV. This would limit me to certain later AGP cards, about the time the DOS compatibility started to take a back seat, so it'd be great to know which cards offer the best EGA/VGA compatibility.

Malvineous wrote:

I agree, this would be very useful. If you're going to the trouble of testing all this, it would be great to see some tests done in pure DOS as well.

I would consider this to be a parallel project, as the one I was considering was going to be pure-XP but I really like this idea. I saw a thread the other day of a project someone is doing to test LCD monitor compatibilityand the software set includes some really obscure stuff. I could easily repurpose it for your idea.

Just beware there are A LOT of factors that influence the performance of a vídeo card, and a card that performs better than others in one PC, might actually be slower on another. Example: one user here in the forums tested and claimed a Banshee yelded better results than a Voodoo3 in his K6-2 (or was it K6-2+?), probably because the driver is better optimized for his processor. There is also the issue of not-so-perfect AGP implementations on non-Intel chipsets. Also, when testing pre-DX8 cards with very fast processors another issue arises: even a card that has no acceleration at all can run most DX7 titles (unless it is a victim of buggy drivers/hardware) - albeit with visual glitches - because even if the card doesn't accelerate a certain feature, it can run in software. Hardware T&L is a good example of that: when running on a super fast processor that can do T&L in software at a reasonable speed (compared to older cards), cards with hw T&L acceleration will not show the same performance increase as it would on a slower processor.