Very well thought out already!

I have a suggestion: Add the 320x200x8bit PCPBench mode to get more information about VESA-framebuffer speeds. And since the higher resolution PCPBench tests already seem very fast, let's raise the color depth gradually (e.g.: 800x600x16bit, 1024x768x32bit).

Another nice thing would be SpeedSys screens.

Retro, if you pull this through this is going to be one of the most awesome testing I've seen so far! I dont think we need much more tests other then the ones you included, though you may want to include some Sandra scores (use the same version of OS and Sandra please), so we can have it as a reference.

And once more - good job!!!

Awesome lineup. I second the Sandra.

Thanks everyone! I will get all of these tests done today, including all suggesions. (I have edited the O.P. just to tidy it up a bit.)

Also, regarding Pcpbench - I will add pcpbench 130 (320x200x8) to my tests. When I run pcpbench 114 (800x600x16) and 118 (1024x768x32), these 2 tests do not display "LFB" on the pcpbench test screen. And consequently, the score obtained seems quite low. Do you know how I can tell Pcpbench to use "LFB" mode when I run it in modes 114 and 118? (I am running all of these Pcpbench tests inside Windows 98 SE. If I boot up the mobo, and then select either "safe command prompt" or "normal command prompt" DOS modes, and then run pcpbench, I seem to get poor scores, even though the test screen displays "LFB" on many of these tests.)

Also, regarding Speedsys - I think the best thing to do is gather up all these screenshots, and put them all in to a seperate post.

Speedsys 4.78 results. All run from Windows 98 SE's "safe command prompt only" DOS area.

Mobo A tests

• Test 1 - No FSB overclocking.

• Test 2 - FSB OC'd to 112.

Test 3 - FSB OC'd greater than 112 does not seem entirely stable, using the nVidia Ti4200. (Currently, there is no Speedsys screenshot.)

• Test 4 - FSB OC'd to 120.

Mobo B tests

• Test 1 - No FSB overclocking.

• Test 2 - FSB OC'd to 110.

Test 3 - FSB OC'd to 115. Windows 98 failed to boot up. No tests done, and no screenshot available.

Mobo C tests

Test 1 - No FSB overclocking. (I have reached the maximum number of allowed attachments for this thread. This screenshot cannot be posted to this thread. I will start a seperate thread...)

I know I am getting a bit spoiled, but incorporating an Abit BP6 + powerleap adapter instead of that Asus board would be an even more awesome intel testing-rig...

retro games 100 wrote:

Also, regarding Pcpbench - I will add pcpbench 130 (320x200x8) to my tests. When I run pcpbench 114 (800x600x16) and 118 (1024x768x32), these 2 tests do not display "LFB" on the pcpbench test screen. And consequently, the score obtained seems quite low. Do you know how I can tell Pcpbench to use "LFB" mode when I run it in modes 114 and 118? (I am running all of these Pcpbench tests inside Windows 98 SE. If I boot up the mobo, and then select either "safe command prompt" or "normal command prompt" DOS modes, and then run pcpbench, I seem to get poor scores, even though the test screen displays "LFB" on many of these tests.)

Oh, bummer. Seems that windows interferes with the LFB modes, but under plain DOS, most of the video memory is not cached.
On the Intel/BX mobo an utility called FastVid should help, but it can be a pain to set up, and I'm not sure whether it works correctly with the Athlon boards.
So, maybe it's better to stick your original plan (8bit modes only), it avoids a lot of hassle.

Speedsys 4.78 results, continued. (I couldn't add any more attachments to the original Speedsys post.) All run from Windows 98 SE's "safe command prompt only" DOS area.

Mobo C tests

• Test 1 - No FSB overclocking.

• Test 2 - FSB OC'd by 7 Mhz to 140, from 133.

• Test 3 - Upgrade CPU to a "t-bred" 2400+ (2000 Mhz). FSB not OC'd.

• Test 4 - Upgrade CPU to an "unlocked" (16x multiplier) mobile barton, running at 2176 Mhz. FSB also OC'd to 136 FSB, from 133 FSB. Please note: because the multiplier is "unlocked" inside Windows 98 (from 6x to 16x), I cannot run Speedsys in "pure" DOS. Instead, Speedsys must be run in Windows, and consequently, some of the memory timing tests cannot be done.

I have temporarily stopped testing, because something doesn't make sense. The 3Dmark scores on the non-OC'd Intel 440 BX system (100 FSB, 1.4 Ghz, AGP at 2x) are better than the non-OC'd VIA KT133A system (133 FSB, 1.467 Ghz, AGP at 4x). I don't understand why. Possible reasons -

1) VIA KT133A is not good.

2) The VIA 4-in-1 driver package could be "badly installed". Unfortunately, I can't remember what version I'm using. I think it's 4.35. For all of these tests, I am using the same hard drive. When I attach it to these 3 test mobos, I simply let Windows 98 "sort itself out". What I mean by that is - Windows 98 seems to now realise what mobo is attached to this Win98 hard drive. So, if I attach the Abit KT7A mobo to it, which has a VIA chipset, all of the VIA system drivers can be seen inside the System control panel area. If I then attach the Intel mobo to the hard drive, all of the Intel system drivers can be seen inside the System control panel area. Etc. What I don't do is reinstall anything. So, I don't reinstall the VIA 4-in-1 driver. Should I?

3) There could be something "wrong" with the t-bred CPU and the Abit KT7A mobo. What I mean is - the BIOS POST screen says "Unknown CPU Type". It seems to work OK, but perhaps it's not working to its "maximum potential"? If the mobo doesn't fully understand what the CPU is, then maybe the mobo is not using all of the CPU properly?

I could continue with the Abit KT7A mobo, or I could remove this mobo and replace it with another VIA KT133A mobo - the QDI Kinetiz. The advantage to using the QDI Kinetiz is that it definitely knows what a t-bred CPU is. I could also uninstall the nVidia/MSI Ti4200 graphics card driver, reinstall the VIA 4-in-1 driver, then reinstall the nVidia/MSI Ti4200 driver.

One last thing - oddly, the current Abit KT7A system will now not run 3Dmark 99 Max. It quits back to the desktop. I am not OC'ing it, and there's plenty of power because I'm using 2 PSUs. Also, I ran 3Dmark 2000, and it worked OK - although it was slower than the Intel 440 BX result!

Edit: I uninstalled the nVidia graphics driver, then reinstalled the VIA 4 in 1 package (all 4 options selected, including "turbo mode" for AGP), then I reinstalled the nVidia graphics driver. (I'm still using the Abit KT7A board, BTW.) 3Dmark 99 max still quits back to the desktop. So, inside the BIOS set up area, I increase the CPU's core voltage and IO voltage by 1 "notch". 3Dmark 99 max now works OK. But each time I run 3Dmark '99, I get about 8800. The Intel board scored about 9200.

Edit 2: I'm getting a bit suspicious about this t-bred 1700+ (1467 mhz) rated CPU. I have temporarily put the Abit KT7A mobo away, and set up the QDI Kinetiz mobo instead. The Kinetiz BIOS POST screen says that this t-bred CPU is an AMD Duron! I run SiSoft Sandra 2002 Pro, and here is the info about this t-bred 1700+ rated CPU -

SiSoftware Sandra

Processor(s)
Model : AMD Duron(tm) processor
Co-Processor (FPU) : Not installed
Speed : 1.46GHz
Model Number : 1705 (estimated)
Performance Rating : PR2137 (estimated)
Type : Standard
Multiplier : 11x
Generation : 7th (7x86)
Model Information : Duron M8/MP & Athlon 4/MP/XP (Thoroughbred) 1.5G+ 1.65V
Revision/Stepping : 8 / 0 (0)

Caches
Internal Data Cache : 64kB synchronous write-back (2-way, 64 byte line size)
Internal Code Cache : 64kB synchronous write-back (2-way, 64 byte line size)
L2 On-board Cache : 64kB ECC synchronous write-back (16-way, 64 byte line size)
L2 Cache Multiplier : 1/1x (equiv. 1461MHz)

I have solved the problem with the KT133A mobo testing! There is definitely something "wrong" with the t-bred 1700+ (1467 mhz) rated CPU. I just had a very lucky find, and found a t-bird 1400 mhz athlon CPU in one of my boxes of junk. That's perfect for this set of tests, because the Intel "Test A" board also uses a 1400 mhz CPU.

I reran the 3Dmark 99 max test again, this time with the t-bird 1400 mhz (133 fsb version), and at last I finally get a sensible score: 9689, 19801. OK, the 2nd CPU score is less than what was seen using the t-bred CPU, but the 1st 3D score is a big improvement.

Now that I have changed the mobo (removed Abit KT7A, replaced with QDI Kinetiz), and also changed the CPU (removed t-bred 1700+ rated, replaced with t-bird 1400 mhz), I will redo my "Mobo C" tests again. I will also edit the O.P. and update this hardware change.

retro games 100 wrote:

I'm getting a bit suspicious about this t-bred 1700+ (1467 mhz) rated CPU. [...] The Kinetiz BIOS POST screen says that this t-bred CPU is an AMD Duron!

🤣

I know I am getting a bit spoiled, but incorporating an Abit BP6 + powerleap adapter instead of that Asus board would be an even more awesome intel testing-rig...

While the BP6 is a good board, for sure, there wouldn't be much point to testing it in this comparison, since its main advantage is SMP capability, and the tests are being run under DOS/Win98. In single CPU use, the P2B is every bit as fast as the BP6, and is arguably a bit more stable.

Anyhow, it's sort of a matter of opinion, and most BX boards are going to fall within a couple % of any of the numbers reported for the P2B, so it should give a good cross-section for the platform.

As for the problems with the TBred 1700, yeah, sound like there's something wrong with that particular chip. Have you looked closely for any chips or cracks on the core substrate? That was a pretty common cause of Sudden Athlon Death Syndrome, and it could cause all sorts of strange problems without completely killing the chip.

Old Thrashbarg wrote:

As for the problems with the TBred 1700, yeah, sound like there's something wrong with that particular chip. Have you looked closely for any chips or cracks on the core substrate? That was a pretty common cause of Sudden Athlon Death Syndrome, and it could cause all sorts of strange problems without completely killing the chip.

Yes! I just checked this CPU, and on the top left and also top right sections of the head part, you can see that it has "crumbled away". 😮

I have finished testing. 😅 Please see the O.P. and also the 2 other Speedsys posts for all the results. I'm going to see if I can put all of the results in to Excel spreadsheet charts. But before I do that, if someone wants me to run any further tests, or suggest a better way to "chart" the results, then please let me know. Thanks.

Edit: Would something like this be OK?

Edit 2: Please note that I have resorted all of the Speedsys screenshots, so they now appear in testing order.

Old Thrashbarg wrote:

While the BP6 is a good board, for sure, there wouldn't be much point to testing it in this comparison, since its main advantage is SMP capability, and the tests are being run under DOS/Win98. In single CPU use, the P2B is every bit as fast as the BP6, and is arguably a bit more stable.

Yes but I think I read somewhere that because of this, you may get increase in memory throughput - beating even DDR boards in some cases. Right?

Retro - are the 3dmark 2001 se scores updated? The score between 7600 and 8867 seems pretty lousy for that video card and KT133a. I mean, I've had more then 10 000 score with a radeon 8500 which is a much worse video card then the ti4200! And that was done on a Asus A7V133-c w/ palomino 1800+. I was expecting something between 11000 and 15000 in your case.

It also seems that an overclocked celeron <1.7 Ghz achieves (almost) better score then the 2Ghz Athlon, which is almost incredible.

Yes but I think I read somewhere that because of this, you may get increase in memory throughput - beating even DDR boards in some cases. Right?

I've never heard of anything like that, nor have I ever seen any indications of such in any benchmarks I've done over the years. I don't see how it could even be possible... the 440BX has the same memory controller in it, regardless of whether it's hooked to single or dual CPUs.

Now, the 440BX did beat most of the P3 DDR boards that floated to the market, but that's mainly because the BX was a good chipset, and the Via Apollo 266 was shite.

Man, I hate to be the Scrooge here, but I completely don't understand the point of the review. (I will say it was very well done through Retro.) Comparing a Celeron, a T-Bred and a Palomino on different motherboards and different chipsets is like comparing apples to oranges. The comparison would be no more or less valid by throwing in a P4. The i865 chipset has Win98 drivers too. Depending on the generation, of course one of them will be faster!

Considering all of this is old hardware, the speed doesn't matter so much as the reliability, compatibility and consistency of how it performs. You really just need a speed "appropriate" setup for the generation of games you are going to play. If raw speed was important, just get a modern quad-core.

That being said, IMO, the "best" is easily the BX motherboard. There is a reason VMware chose to emulate the BX chipset. It was one of the most reliable chipsets ever. And funny enough, it's being compared to the KT133A, which is one of the worst.

elfuego wrote:

Retro - are the 3dmark 2001 se scores updated? The score between 7600 and 8867 seems pretty lousy for that video card and KT133a. I mean, I've had more then 10 000 score with a radeon 8500 which is a much worse video card then the ti4200! And that was done on a Asus A7V133-c w/ palomino 1800+. I was expecting something between 11000 and 15000 in your case.

It also seems that an overclocked celeron <1.7 Ghz achieves (almost) better score then the 2Ghz Athlon, which is almost incredible.

I was so careful to get every score recorded correctly. I'm very confident that every score value in the O.P. is correct. If the values seem "wrong" in some way, then I need to investigate why these values are lower than expected. The BX results seem "healthy", so there's no need for any immediate investigation other than to play some games! The AMD mobo results were a bit disappointing. After I finished the tests, I tried to figure out a way of OC'ing this system. The mobo has FSB dip-switches. They go up to 115 Mhz, but 115 mhz is unstable. 110 mhz is OK. Curiously, these FSB OC'ing dip-switches also go down to 90 Mhz. Why is that? The maximum CPU that this board can accept is a t-bird 1400 (100 fsb). It can also accept a Palomino 1500 rated CPU (100 fsb), which has a real speed of 1300. The top speed 100 fsb t-birds are rare and very expensive. I don't have one, but I do have the 100 fsb 1500 rated Palomino. I don't know which is faster, but I guess the Palomino is good enough.

I tried a few fast CPUs eg t-bred 2400+ and even an XP-M. They both POSTed!, but the BIOS POST clock speed always said 1500 mhz (and no more), and the BIOS POST screen always froze when it said "Checking NVRAM..."

Before I discuss the problems with the KT133A, I just want to talk a bit more about the AMD chipset test mobo. There are no options inside the BIOS set up area which will allow me to select the CPU's multiplier. So, if I use a "factory unlocked" t-bird (eg 1400 mhz), or a specially "user made unlocked" palomino, I don't see how I can increase its multiplier. I tried using cpumsr.exe on a "factory unlocked" t-bird (which had it's L1 bridges closed by factory default), but the cpumsr program would not allow me to increase the multiplier. Do I need to use wpcredit before using cpumsr? I tried setting the 3rd bit of field 55, but then I realised that was for the VIA KT133A chipset, and not the AMD 750 chipset!

OK it's time to comment about the "problematic" KT133A mobo, and the Ti4200 graphics card. Why were the scores low? Well, I can only make guesses -

1) I reinstalled the VIA 4-in-1 4.35 chipset driver package, and selected every install option, including "turbo mode" for AGP. I don't know if that was the correct thing to do. (I reinstalled this VIA driver package because I wondered if my results were lower than expected.)

2) I removed a damaged t-bred 1700+ rated CPU, and erased its results from my O.P., because part of its cache was physically damaged, and was interfering with these tests. I replaced it with 3 different CPUs, for a total of 4 different tests seen in the O.P. My guess is that these 3 CPUs cannot all be broken, so at least one of these CPUs should be OK.

3) The QDI Kinetiz BIOS (flashed to version 4.1, dated January 2004!) is not all that effecient for windows 98? Maybe the mobo is not well designed? (I'm really guessing here!)

4) The Ti4200 driver is not good. I'm using the nVidia driver from the nVidia website. It's only version 45.something. I think I can go as high as 61.something. I remember having problems with some games with the 61 driver, and so I used the 45 driver instead. The 45 driver runs everything without error. What's really curious about this 45 driver, is that if you go to Display Properties inside Control Panel, there are 2 specific MSI tabs of information, but I got the driver from the "generic" nVidia webpage. Strange!

I have just double-checked the driver version. I went to the MSI website, and went to their download page -

http://eu.msi.com/index.php?func=downloaddeta … 130&prod_no=887

I looked at my graphics card, and there is a sticker on it that says G4Ti4200-TD64. The webpage above is correct for this product. The highest driver version is 45.something. That's exactly what I am using. Maybe the Ti4200 doesn't work very efficiently with the KT133A? It's interesting, if you scroll down there is info about a driver NOT recommended for KT133 and also a driver Recommended for KT133. But these are old drivers, and I guess they can be ignored.

Also, maybe I need to look on the net to see if I can find a test already done using a KT133A mobo + a Ti4200, and then compare my results.