feipoa wrote:

I have seen similar results with a piece of junk PC Chips M918 motherboard. If I slow everything down, I see a cache step in Speedsys, however if I run it at full speed, the cache does nothing. Other motherboards of the same chipset do not exhibit this problem, so I believe there is a cache implementation issue with the motherboard. I wonder if the issue you are having is similar, that is, the cache isn't being fully utilised.

Maybe the UT6164C32Q-6 cache chips are too slow for faster CPUs. Maybe your PC Chips M918 motherboard have also UT6164C32Q-6 cache chips.
And maybe UT6164C32Q-6 cache chip based boards can boot Tillamook with L2 cache because they gives a very small plus performance.
But the fact that there are exist so fast CPUs which they do not exist motherboard cache can fast enough to not slow down the overall performance.
For example the board that I have used for K6-III+ was Tyan (which one is not a junk MFG). When the L2 was enabled, the results was slower.

amijim wrote:

Hello Gona,can you benchmark the bigass demo ?Also i looked at your speedsys mem benchmarks and i saw about 128mb per sec for l2 and 94mb per sec for the sdram.I am now playing with award edit software to give lower values cas and mem parameters down to 1 so as to overclock a little the mem access.I believe we can reach the l2 speeds so as to make the l2 mem unecessary.Of course i am using pc133 memory modules.

Hello. The Voodoo1 was not a good idea from me to compare chipsets/motherboards because Voodoo1 is too "CPU independent" there are very small differences.
Do you have Voodoo2 or Riva128 PCI? Or maybe we should use WinQuake and some PCI graphics card.

Goodmornig Gona,
yes i have a voodoo2 and a riva128 pci.The old trusty diamond viper v330.It is interesting approach.Also can you upload speedsys tests running at both 266mhz on 66mhz bus and 75mhz?It seems that i hit the wall with the 64mb cache bottleneck in my 430tx.Also try award bios edit to play with undocumented mem settings to lower the cas ras values.It seems that we are getting somewhere.Speaking about award bios edit , i found out that i can disable the jumperless mode in qdi titanium iiib and thus allow a k6-2 and k6-3 to post in my qdi , though i want to keep it in my original spec with an intel pentium mmx ,preferably the tillamook.Maybe a usefull tip for non posting mobos with k6-2 cores.

I have found my Diamond Viper V330 PCI too. The bios on the PC Phip M537-DMA33 board is already modbinned and set to the fastest config.
I will put my latest async board (that has Intel FX chipset) to this test also. Now I working on this board. Later I will made some speedsys graph capture too.

Gona wrote:

feipoa wrote:

What is the Quake score with and without L2 enabled? If the results are about the same, perhaps the L2 cache isn't being used properly. Do you have the Speedsys graph with L2 enabled?

Both with or without L2 gives 27.0 FPS (40.3 sometimes 40.4 seconds), nothing difference. In the earlier test, the SysId show a little advantage of the L2 cache enabled: memory MMX 409 vs. 407 (I have repeated this test with the same results: 409 vs. 407). I think the L2 cache is works but too slow to be able to give advantage for the 266MHz Tillamook (as my VIA MVP3 board has not enough fast cache and K6-III+ 550MHz is faster without L2 cache).
I have downgraded the Tillamook as slow as I can (2x50MHz) and I have made Speedsys graphs with enabled and disabled level 2 cache. It seems to me L2 is works.

Here it is Speedsys 4.78 - Tillamook 266MHz in PC Chips M537-DMA - L2 enabled

I have done GLQuake 0.97 timedemo bigass1 on Riva128 with nVidia reference driver "RIVA128 Driver for Windows 9x PCI Version: 3.37" (4.10.01.3.37) on 640x480x16
6222 frames 400.3 seconds 15.5 fps

Hmm.. Either your system's L2 cache is too slow, or the RAM is too fast to show much improvement. If you run the system with a P233MMX with and without L2 cache enabled, do you get the same percent of performance improvement with the P233 (L2 on vs. L2 off) as you do with the P266 (L2 on vs. L2 off)?

Nice remark there feipoa.Stay tuned for my benchmarks.I think we are safe to make conclusions here.The best cpu for the socket 7 era mobos is the k6-3+ and for those mobos that the bios does not post with a k6-2 core,is the pentium 266mhz tillamook that can run safe on 2.8 vcore with a good socket A heatsink and overclocked to 4*75mhz..Stay tuned for more details.

feipoa wrote:

Hmm.. Either your system's L2 cache is too slow, or the RAM is too fast to show much improvement. If you run the system with a P233MMX with and without L2 cache enabled, do you get the same percent of performance improvement with the P233 (L2 on vs. L2 off) as you do with the P266 (L2 on vs. L2 off)?

OK, here they are:

It was noted that the Quake score was the same with and without L2 cache when using the P266MMX. Is the score the same with and without cache when using the P233MMX?

feipoa wrote:

It was noted that the Quake score was the same with and without L2 cache when using the P266MMX. Is the score the same with and without cache when using the P233MMX?

Note that those benchmarks has maden on Voodoo1 which card is really "CPU independent" and this caused lot similar results. I have made new benchmarks with Riva128 and Voodoo2:

So with the Riva128, you get a 2% performance increase with the P266MMX when enabling L2 cache. What do you get with the P233MMX? We need to know the Quake result for the P233 when L2 is disabled. If it is the same 2%, then I suspect your Tillamook works fine with L2 cache and that the L2 cache just doesn't do much.

feipoa wrote:

So with the Riva128, you get a 2% performance increase with the P266MMX when enabling L2 cache. What do you get with the P233MMX? We need to know the Quake result for the P233 when L2 is disabled. If it is the same 2%, then I suspect your Tillamook works fine with L2 cache and that the L2 cache just doesn't do much.

Good idea. I have to reinstall the OS because I qot better MMX 233 results than before. So I retest all. Tillamook results are the same (except one 0.1 difference) but all P55C results are better. The MMX 233 with and without speed really similar to Tillamook results. This L2 can provide this, level2 cache of the PC Phips M537-DMA works with Tillamook CPUs too.

It looks like with the Riva128, the use of L2 increased Quake results by 1.37% for P233, while it increased by 2.65% for the P266.
When using a Voodoo2, the P233 results increased by 1.34%, whereas with P266, results increased by 0.84%

The increase for the P233 seemed more consistant between Riva128 and Voodoo2, but I'm not sure what else we can deduce except that L2 doesn't help all that much. Kinda makes me wonder if the increases would be greater if using Quake in software mode.

Hello Gona and feiopa.I think we can now conclude that the tillamook at 4*66 or better at 4*75 or 4*83 are better Than the pentium 233 mmx even running without l2.You see i have a deskpro 2000 ,an ibm 300gl and a qdi titanium iiib that do not post with a k6-2 or k6-3 core and were stuck to the 233mmx core.At last a small upgrade for our socket 7 retro. Win98 pcs.

Are there some chipsets which do not work at all with the Tillamook, even without L2 cache? I sorta recall testing the Tillamook on my 430TX-based board and it didn't work at all, however, you'd have to dig up some old posts to be certain of this recollection. I would postulate that each individual board would need to be tested with the Tillamook vs. P233MMX to determine the best fit. Keep in mind that some people run the P233MMX at 262 MHz (3.5x75) or 300 MHz (3x100) and can keep their L2 cache.

i wanted to buy a few Tillamooks but I'm not sure how high one can clock these? I got the impression that they are easier to overlock than the standard fare MMX233s. The second level cache did not have too much of an impact on performance when I tested the few boards that I have.

Clockwise for most PMMX chips the limit seems to be either 250 or a flakey 292(83x3.5). Although at latter speed these processors pack quite the punch if the board is good.

I had no problem running my P233MMX stable at 262 MHz on a 430TX board.

I vaguely recall talk of the Tillamook running at 350-400 MHz. Re: Tillamook

If I disable "SDRAM Speculative Read" in bios, I can boot Tillamook on my 430TX-based board: ASUS TX97-XE (TM tech cache chips). If I not disable this, the board will freezes before drives detection. I wanted to compare the cache performance of this board to the cache performance of PC Chips board.

We can see that the ASUS TX97-XE is faster than the PC Phips M537-DMA but the performance increase by L2 cache with P233MMX also small (even smaller). I think all of the boards that times gives small performance increase to P233MMX and Tillamook. And we can see also that the ASUS TX97-XE without L2 cache is faster than the PC Phips M537-DMA with L2 cache.

Gona wrote:

If I disable "SDRAM Speculative Read" in bios, I can boot Tillamook on my 430TX-based board: ASUS TX97-XE (TM tech cache chips). If I not disable this, the board will freezes before drives detection.

Hmm... I'm not sure if I had this feature enabled or disabled when I tested on my TX boards. Hopefully, I'll remember this the next time I test the Tillamook.

Gona wrote:

the ASUS TX97-XE without L2 cache is faster than the PC Phips M537-DMA with L2 cache.

That's the first thing I noticed! The difference is not substantial though. If you're determined to use the Pentium MMX series with maximum performance and 64 MB of RAM is sufficient, then the 430TX may be your best bet. However, the 430TX board's best attempt is 292 MHz, while the MVP3 may do 400 MHz reliably. I personally run my 430TX board with an 83 MHz FSB and haven't had any issue.

Do you have any interest in running the tests with Quake in software mode to see if the percent difference is any greater?

feipoa wrote:

Do you have any interest in running the tests with Quake in software mode to see if the percent difference is any greater?

OK. Do you want DOS Quake (with 64 MB of RAM) or (WinQuake with 256 MB of RAM) tests?