As you are running a K6+ chip with integrated L2 cache, from what I understand the motherboard L2 cache becomes L3 cache and the chipset caching limitations only apply to the L3 cache.

Based on your description it sounds likely that the motherboard cache is not taking effect, due to the cacheability limit. From what I remember reading, the OS fills memory from the opposite end of where the cache starts working, so if you only use 64MB, you'll be using the uncached part of the RAM.
However, I don't think this is causing you a big problem.
The integrated L2 cache on the K6-2+ makes the motherboard cache largely unnecessary. It was really only there because it was needed for older socket-7 CPUs. With a properly working K6-3, K6-2+ or K6-3+ CPU, I don't think benchmarks ever showed the L3 to be very effective. Maybe some other people have better insight into this, I only have a vague notion that the L3 isn't a big performance issue.

The integrated L2 on your processor supports larger memory sizes. I don't remember the limit, but it should be caching all of your RAM, it's only the L3 that's going out of range.
I don't know what scores you should be getting but if your system is underperforming, these are my guesses:

1) Make sure your BIOS version is truly intended to support the K6-2+. It might not be initializing the CPU into a properly optimized state. There might be software utilities that can address this, but maybe only for DOS.

2) Have you gone over all the various bus and divider jumpers? I don't know the SiS chipset but I know on VIA MVP3's, it was possible for the RAM to be clocked at 2/3 of the FSB instead of full speed. This was for compatibility with older/slower RAM, but it also hurt performance. On my board the way this was labeled was that you could run the RAM either on the FSB clock, or the AGP clock.
There might also be BIOS settings to choose between SDRAM and EDO, and presumably you should be using PC100 CL2 SDRAM.

3) Maybe there's some obscure BIOS option that's having a significant effect on performance, or worse, maybe it's something completely hidden. If you haven't already, try clearing the CMOS and then load optimized defaults. Sometimes weird things can happen with hidden settings, especially after doing a BIOS update which might not interpret all the stored data the same way the old version did.

According to http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm the official F3 BIOS 05/09/2000 should properly support a K6-2+.
IIRC you should be able to limit the RAM usage to less than the installed RAM, for example in msconfig. I never had the need to try this, though. Still with a K6-2+ this should not be much of an issue since the integrated L2 cache will cover all RAM.

Ah yes, the old RAM tag problem.

So I actually have a pentium 1 system that has the smaller RAM tag. What happens when I incrementally go past 64 MB is that the performance on some games gets slower each 128MB increment. So 128MB of RAM is slightly faster than 256 MB-- and 384 MB being the slowest.

If you're not noticing a change in performance when incrementing the memory back and forth, then the slowness isn't related to the RAM tag. As already mentioned, the K6-2+ has the ability to bypass the RAM tag.

Do you have a regular Pentium 1 to put in the system? This would be useful for two reasons:
#1 you'd be able to analyze the memory problem better, and see if changing from 128 MB to 384 MB has a huge hit on performance
#2 the problem may be a defective CPU or some weird compatibility-- or even the K6-2's inferior FPU

Is there a turbo function on the mainboard? If yes, make sure its enabled.

Hi Digitaldrifer!

You can compare your results with my V2 CPU scaling project: Phil's Voodoo 2 (and SLI) processor Scaling project - 65 pages, 875 benchmark results

On slower machines I would actually recommend going with W95 instead of W98. It is less demanding and will give you a more responsive system.

Now I see you use the mobile chip with on-die L2 cache. This means that the entire memory is cached with the on-die L2 cache. You might as well turn off L2 (which is now L3 cache) in the BIOS and see little difference. These mobile chips have no limit as to how much memory they cache.

But might be time to source some more SDRAM. Good news is that they are very cheap.

Some boards run slower with a K6-2+ K6-3 than with a K6-2 plain one.
Because of some bug or chipset / bios incompatibility, so you might face this problem.
Did you run the last bios for the board?
The K6-2+ and 3 have the TLB bug also if i am right, but i do not know what it does to performance as i dont own one.

So some type of RAMDISK wouldn't help?

Mau, that document is amazing, and it confirms my suspicions. I'm only at about half of the frame rates I should be seeing, and I have no idea what the issue is. There are plenty of helpful suggestions here, though, and once I'm done with this stupid training I have to do at work, I'll get to messing around.

I do have another system lying around with a classic Pentium 166. However, it is in an old Sony Vaio desktop, and may not be easily removed. Now that I think about it, when I bought the motherboard it came with some sort of standard K6-2. I'll try it in that.

I probably wasn't the clearest in my initial post (executive functioning disability here), so here's where I stand currently (changes inclusive):
Quake 2 Baseline: K6-2+ @550, 384mb ram, Voodoo 2 sli, Win98 === 34.2fps
Sli removed (single card) === 32.5fps
128mb ram === ~32 fps
Onboard cache disabled === ~32 fps

All bios speed options are enabled, and I'm certain I have the latest bios (although it detects my chip as a K6-III). I am running at 5.5x100mhz because anything higher and the onboard video gets fuzzy, so it's not a bus speed issue, and my ram is PC133 anyways. Baffling as hell. I'll follow the suggestions you guys gave me and I'll let you know how it turns out.

Joel

JaNoZ wrote:

Some boards run slower with a K6-2+ K6-3 than with a K6-2 plain one.
Because of some bug or chipset / bios incompatibility, so you might face this problem.

Is this phenomena not specific to the ALi Aladdin V chipset-based Asus P5A-B motherboard? Do you know which other 100 MHz FSB-capable motherboards have improper support for K6-2/3+ CPUs?

digital drifter, what frame rates are you expecting from a Voodoo2 with a K6-2+ in Quake II? Using the Ultimate 686 Benchmark Comparison as a starting ground, I noticed 33.4 FPS in Quake II with an AMD K6-2+ 550, Windows 98SE, and a Matrox G200. The 3DNow! patch was not implemented. The motherboard utilised was based on the VIA MVP3 chipset. A board tested based on the 430TX chipset scored about 1 FPS less.

With a single V2 I got just under 60 fps at 800 x 600. A bit of SLI scaling, but not much: 61 fps at 1024 x 768.

Digitaldrifter, I wonder do you have another board to test?

And yes, do test with a Pentium. MMX 233 should get 40 fps pretty much at all resolutions.

Mau1wurf1977, what fps do you get at 640x480? What motherboard/chipset did you use for testing?

While I realise the Voodoo2 is certainly the preferred card for gaming compared to the Matrox G200, I did not realise the performance obtained would be nearly double. Mau1wurf1977, if you have a G200, what FPS at 640x480 do you get using the same motherboard?

It's all detailed here in my 60+ page document 😊

https://dl.dropboxusercontent.com/u/94628483/ … g%20project.pdf

Motherboards, driver information, BIOS settings are towards the back.

If you meant the results with the G200 are detailed in that document, I could not locate them. I do see, on the other hand, that you used a DFI K6XV3+/66 motherboard based on the MVP3 chipset and obtained roughly 66 fps at 640x480 with a single Voodoo2.

I got some Matrox cards. But never bothered testing them. Will any regular G200 AGP do?

For this experiment, I would say any G200 will do. But if you have, both, the AGP and PCI versions, testing each would be most beneficial in determining if low frame rate is due to the PCI bus. My benchmarks were on a PCI bus.

feipoa wrote:

For this experiment, I would say any G200 will do. But if you have, both, the AGP and PCI versions, testing each would be most beneficial in determining if low frame rate is due to the PCI bus. My benchmarks were on a PCI bus.

Sorry AGP only...

I got one that has MGA-G200A-D2 written on it. Is this the one?

I believe I also have a 400.

PS: I don't have the S7 platform around any more.

Ok did a quick run on my Pentium III-S 1.4 GHz reference platform: http://www.philscomputerlab.com/pentium-iii-s … nce-system.html

G200:

1024 x 768: 20.3
800 x 600 32.1
640 x 480: 46.3

G450:

1024 x 768: 55.2
800 x 600: 84.8
640 x 480: 120.1

G450 is pretty decent. Is there anything interesting about it?

Thanks for the test results. It seems like the G200 is just that much slower than the Voodoo2 in Quake II.

To me the interesting aspect of the G450 was the PCI version because it contained a DVI port. It seems like not a lot of consumer-class graphics cards at the time of production had a DVI port.