I currently have a millennium I in my P133.
As soon as I get my hands back on my mystique, i'll add the input (I guess P133 is 1.5x compatible)

My MB is socket 5, 430FX, 256KB PB.

I only got AGP Matrox cards.

Really nice results! Great work, as always!

vetz wrote:

...
Hopefully the results will show which chipsets and boards have the best performance and how much cache and type of RAM (EDO/SDRAM) affects the results. The Pentium 100 was chosen since it is supported in all boards.
...

Can I request additional tests with the same CPU: The slowest the board is capable of with L1/ L2 caches enabled, using Turbo switch, if available. And fastest the board is capable with the same CPU overclocked.

That's 2 additional tests, maybe too many? Thought I'd mention it because I was just doing these tests for an AT build...

Great thread! There's some really great information coming out of these benchmarks - I wish it was around when I was doing my various builds.

I use an ASUS VX97 so I'm glad to see it up near the top of your list too!

I added the result I got during Phil's benchmark and now I'm looking forward to testing my two other boards based on 430HX and 430TX chip sets.

Please note:

Phil uses a different method for SUM of FPS.
It would be great to lock cells people are not supposed to change (Sum of fps, Doom FPS, etc.).

RacoonRider wrote:

I added the result I got during Phil's benchmark and now I'm looking forward to testing my two other boards based on 430HX and 4 […]
Show full quote

I added the result I got during Phil's benchmark and now I'm looking forward to testing my two other boards based on 430HX and 430TX chip sets.

Please note:

Phil uses a different method for SUM of FPS.
It would be great to lock cells people are not supposed to change (Sum of fps, Doom FPS, etc.).

They aren't locked? 😒

Interesting how the VIA MVP3 with 1024K cache falls so far behind the 430FX with only 256K cache in 3DBench2.

Is the pipeline burst cache in the 430FX responsible for the socket 7 clobbering its same-chipset, socket 5, older brother? Do you have any motherboards which allow for asynchronous SRAM and a PB COAST module on the same board? It would be interesting to see the changes with these benchmarks. Was there already a thread on this? /me thinks there was.

Were there any 430HX boards which allow one to plop in an AMD K6-3+ with just a BIOS mod and no interposer board? I suppose this was one of the glories of the 430TX.

Very nice work Vetz. You certainly have a nice stash of socket 5/7 boards. One day I'd like to run a similar comparison with my holdings of 486 and 386 motherboards.

feipoa wrote:

Is the pipeline burst cache in the 430FX responsible for the socket 7 clobbering its same-chipset, socket 5, older brother? Do you have any motherboards which allow for asynchronous SRAM and a PB COAST module on the same board? It would be interesting to see the changes with these benchmarks. Was there already a thread on this? /me thinks there was.

Were there any 430HX boards which allow one to plop in an AMD K6-3+ with just a BIOS mod and no interposer board? I suppose this was one of the glories of the 430TX.

Very nice work Vetz. You certainly have a nice stash of socket 5/7 boards. One day I'd like to run a similar comparison with my holdings of 486 and 386 motherboards.

Yes, basically the cache type has alot to say. If you look at the results my FIC PT-2003 board has been tested with 256kb, 512kb Pipeline Burst cache and 256kb async cache. I also have a ASUS P55 board (430FX) which supports both types of cache (COAST and DIP sockets), but the keyboard DIN plug has been damaged and needs to be replaced.

I believe the Gigabyta 586HX revision 2.0 and above supports the AMD K6-3+ with just a BIOS update (according to info on the web). I have the 1.56 which doesn't work 🙁 The problem is that you won't get 100mhz FSB. Maximum in a HX/TX is probably 83mhz. So if you're going to use a AMD K6-3+ to its fullest potential you need a SS7 board. If you're going to limit your build to a 233mhz MMX then by all means, go for a HX or TX board 😀

Thanks 😀 I'm mostly trying to find Socket 5 boards. I find the early Pentium boards interesting. It was a time when they were crazy expensive and most people used 386/486's.

RacoonRider wrote:

I added the result I got during Phil's benchmark and now I'm looking forward to testing my two other boards based on 430HX and 4 […]
Show full quote

I added the result I got during Phil's benchmark and now I'm looking forward to testing my two other boards based on 430HX and 430TX chip sets.

Please note:

Phil uses a different method for SUM of FPS.
It would be great to lock cells people are not supposed to change (Sum of fps, Doom FPS, etc.).

Cheers! Nice result from a 430FX board 😀

I know he uses a different method for SUM. He omits the 3DBench2 score. From my knowledge this was done because it created stranger results on some slower 386's and faster Pentium III+ machines. 3DBench2 is made for slow Pentiums so I felt omitting didn't fit for this benchmark test.

I've tried to lock the those cells, see if it works now. I don't have that much experience with Google Spreadsheets.

j^aws wrote:

Can I request additional tests with the same CPU: The slowest the board is capable of with L1/ L2 caches enabled, using Turbo switch, if available. And fastest the board is capable with the same CPU overclocked.

That's 2 additional tests, maybe too many? Thought I'd mention it because I was just doing these tests for an AT build...

What do you mean, "the fastest with the CPU overclocked?" You want to overclock the Pentium 100 CPU? Or do you mean the fastest CPU you can use in the board?
Regarding slowest capable speeds, I*m doing a benchmark project on my Socket 5 boards since they can be slowed much more than the Socket 7's and give much more insight to how these boards can work as a very flexible substitute for a 486. I probably won't go back testing all the other boards right now though.

I only got AGP Matrox cards.

I'll test my AGP Matrox cards in a MVP3 and see how big the difference is.

As soon as I get my hands back on my mystique, i'll add the input (I guess P133 is 1.5x compatible)

My MB is socket 5, 430FX, 256KB PB.

Great! It should perform very well with that PB cache 😀 Rare to have a Socket 5 board with PB cache. I know they were available, but it was the most expensive option. Mostly Socket 5 430FX boards are just missing extra multiplier options to make them Socket 7 boards. For instance my ASUS board just goes up to 2.0x multipler, but the upgraded model the P55TP4 have up to 3.5x

vetz wrote:

Yes, basically the cache type has alot to say. If you look at the results my FIC PT-2003 board has been tested with 256kb, 512kb Pipeline Burst cache and 256kb async cache. I also have a ASUS P55 board (430FX) which supports both types of cache (COAST and DIP sockets), but the keyboard DIN plug has been damaged and needs to be replaced.

Excuse me, I was so focused on the chipset and cache column I didn't pay sufficient attention to the motherboard type.

vetz wrote:

I believe the Gigabyta 586HX revision 2.0 and above supports the AMD K6-3+ with just a BIOS update (according to info on the web). I have the 1.56 which doesn't work 🙁 The problem is that you won't get 100mhz FSB. Maximum in a HX/TX is probably 83mhz. So if you're going to use a AMD K6-3+ to its fullest potential you need a SS7 board. If you're going to limit your build to a 233mhz MMX then by all means, go for a HX or TX board 😀

From the results of the Ultimate 686 Benchmark Comparison, the AMD K6-III+ 500 on a 430TX board at 6x83 was only a few percent slower than an AMD K6-III+ on an MVP3 board at 5x100. It probably has a lot to do with a) the increased PCI, hence graphic, frequency on the 430TX board, and b) the TX board's BIOS not adding a possibly necessary wait state for the memory/cache control.

vetz wrote:

I'm mostly trying to find Socket 5 boards. I find the early Pentium boards interesting. It was a time when they were crazy expensive and most people used 386/486's.

I too have a strong interest in this era of motherboards, however I do not want to invest in another en devour to find the best of the socket 5 era. It is either too costly, too time consuming, or requires too much storage space.

vetz wrote:

I'll test my AGP Matrox cards in a MVP3 and see how big the difference is.

I've been curious about this for some time, particularly regarding the G200/G250 AGP vs. G200 PCI.

Vetz wrote:

What do you mean, "the fastest with the CPU overclocked?" You want to overclock the Pentium 100 CPU? Or do you mean the fastest CPU you can use in the board?

I mean testing the board by using the same class of CPU, i.e. P54C Pentiums (non-MMX), and getting a Min/ Max of what the board is capable of.

E.g. take the fastest P54C Pentium and downclock it to 100MHZ for your benchmark, then further downclock it to slowest possible that runs the benchmark to get the Min (keeping L1/L2 enabled, but using Turbo switch if available).

Conversely, overclock this Pentium to the fastest the board allows to run the benchmark and get a Max. So, you'll get a Min/ Max and your baseline to compare. What's 'best' is not necessarily the fastest listed - one can look at the board with the largest scope as consideration for the 'best'.

Vetz wrote:

Regarding slowest capable speeds, I*m doing a benchmark project on my Socket 5 boards since they can be slowed much more than the Socket 7's and give much more insight to how these boards can work as a very flexible substitute for a 486. I probably won't go back testing all the other boards right now though.

That would be an interesting project. I don't have any Socket 5 boards to test, but I can agree on the flexibility. I stumbled onto a Socket 7 board that's flexible enough to replace my 486 test board with its Turbo feature. And with L1/L2 enabled:

Pentium Min/ Max from 15MHz -> 240MHz, and runs all the benchmarks too.

I don't know which boards do this (they are not limited to Socket 5), which is why it would be interesting in this test.

feipoa wrote:

Were there any 430HX boards which allow one to plop in an AMD K6-3+ with just a BIOS mod and no interposer board? I suppose this was one of the glories of the 430TX.

This one works with K6-2+, might work with K6-3+ as well.

Because the K6-2+ and K6-III+ have on-die cache, this might even the playing field? With slowing down through caches, I can only speak for ALI and VIA SS7 ATX boards. You pretty much always get a 386DX equivalent with L1 and L2 cache disabled. With L1 disabled and L2 enabled the performance you get depends on a lot of factors, but mostly on the CPU and you end up with a 486DX up to a DX2. The slower ones are Pentium and AMD. The faster ones are Cyrix and MMX.

vetz wrote:

Current results: https://dl.dropboxusercontent.com/u/57506833/vogons/diverse/motherboards_P100.PNG […]
Show full quote

Current results:

Pretty consistent with numbers, that can be found in c't: Mercury/Neptune < Triton Async < Triton PB.

vetz, I just contributed benchmark results for all my Socket 5/7 motherboards to both your and Phil's benchmark. I wonder why no one else is interested enough to run a couple of tests.

RacoonRider wrote:

vetz, I just contributed benchmark results for all my Socket 5/7 motherboards to both your and Phil's benchmark. I wonder why no one else is interested enough to run a couple of tests.

Thanks!

I sorted the results and the two HX boards are barely still on top 😀 Your results seems pretty consistent with mine. I'll post a new updated picture in the first post later.

vetz wrote:

I'll test my AGP Matrox cards in a MVP3 and see how big the difference is.

Results with an G200 AGP with Chaintech 5AGM2 (VIA MVP3)

3DBench2: 95,5
PCPBench: 29,8
Doom: 1222 realticks (61,12 FPS)
Quake: 25,1 FPS

Clearly AGP Matrox cards are faster than PCI in VGA modes for 3DBench and Doom. PCPBench and Quake scores stay the same. Can't be compared directly unfortunately.

While it is unlikely that this will increase performance at all, do you have a G250 to test?

feipoa wrote:

While it is unlikely that this will increase performance at all, do you have a G250 to test?

I don't own a G250, but I do own a G400 AGP and I just benched it. Results were the same.