You can get chips from http://www.cpu-world.com/forum
Theres a LOT of trading with old cpu's going on there for very reasonable prices.
Bought a lot of chips from there myself 😉
It's excellent for getting larger quantities and getting to pick exactly which ones you want.

Mau1wurf1977 wrote:

Still the K6-2 seems to be the chip to get as you can go from 166 to 500 MHz by changing 2 jumpers. This should cover a decent speed range!

A K6+ chip would be even better, since you can change between 2x and 6x multiplier via software. For bus changes you'd still have to move jumpers though.

5u3 wrote:

Mau1wurf1977 wrote:

A K6+ chip would be even better, since you can change between 2x and 6x multiplier via software. For bus changes you'd still have to move jumpers though.

Interesting!

I have a K2+ here, but it won't work in my board...

Can you switch the multi via DOS as well?

The K2+ has on chip L2 cache. I wonder if you could disable L2 and L3 cache through BIOS. Because if L3 cache can't be disabled I won't have access to the "slowest" setting which would be a shame...

Excellent project and presentation. 😀 Keep up the great work!

BTW, what software package did you use for your table? It looks very good.

retro games 100 wrote:

Excellent project and presentation. 😀 Keep up the great work!

BTW, what software package did you use for your table? It looks very good.

Thanks! Pretty happy with the Results. These kind of "slow down results" are what I initially had in mind!

I found a thread of yours where you benchmarked 4x 486 cpus with 3dbench. It was very useful for my table...

The last discussion we had, you where right! Looking at the CPU test results won't give you an accurate gaming performance estimate.

The reason is because the AGP graphics system is much faster than a ISA video card. Therefore games will run better than the CPU benchmark score suggests. On a real 386, the ISA video system will add a delay or slow down to what the CPU can do. This isn't the case with an AGP system and therefore not nearly as much CPU performance is needed!

I am still looking for more 386 and 486 3dbench results and gaming videos on YouTube but there are hardly any. I might post a thread where fellow members can post their 3dbench results and we compile a database.

Another benefit of these "slow down" machines is that loading time is very fast, so you don't need to wait long between levels. Everything happens really quick and smooth, yet the actual gameplay runs at the correct speed.

I did the Table in Open Office, latest version. Nothing fancy!

Updated table!

There was one incorrect result (PIII 500 score for Norton SI with both caches enabled)

I also added MHz numbers to the right column. The figures should be pretty accurate, but keep in mind these are for Games only!

If you are doing CPU calculations, they won't match as my Retro machine has a very fast AGP video card, so games will run faster than the CPU indicates...

Great project!
How many of those benchmarks are 16-bit programs? Since slow down is usually needed to run old 16-bit games, I recommend to focus on 16-bit benchmark programs. I also recommend to run them with and without emm386 or similar memory managers since there might be some difference when the CPU is in real or V86 mode.

It is much more difficult to show the speed benefits of disabling L2 cache only. Newer titles and benchmarks are more likely to show a significant effect.

aitotat wrote:

Great project!

Thanks!

How many of those benchmarks are 16-bit programs?

That's a good question... Personally I like to be able to play games from the early 90s. I grew up with the Sierra adventures, Lucasarts Adventures, Wing Commander Series and Dynamix games. So these are "top of my list".

In a nutshell if you are looking to play games meant for a average 286 and slow to average 386SX I recommend you go with the Slot 1 solution.

If you like a bit more grunt for games that run well on an average to fast 386DX and average 486DX to slow 486DX2, then you should pick the Super Socket 7 platform.

The choice of CPU is also important on the Super 7 platform. E.g. the Cyrix 6x86 is a really fast cpu once you disable the cache. It can be used to have a perfect 486DX2-66 which is brilliant. However the lowest you can go is a very fast 386-DX40 which is too fast for many speed sensitive games such as Wing Commander or Test Drive III.

I haven't got a Pentium chip, but hoping to add some results down the track. I am also looking at sourcing a Cyrix chip that can do 100 FSB. The 366 model of the MII is seems to be a good candidate...

Great Hierophant wrote:

It is much more difficult to show the speed benefits of disabling L2 cache only. Newer titles and benchmarks are more likely to show a significant effect.

Yes the difference is really only noticeable when you start playing with demanding SVGA games...

From here onward I can continue with far less tests, because I was able to draw these conclusions:

- When L1 and L2 are disabled, clock speed and FSB have little impact on performance. Two tests (One with max. clock and one with min. clock) are sufficient to give you an idea what "slowest machine" you can hit with your setup.

- When only L2 is enabled, clock speed and FSB have do have an impact on performance. So for this setting, a few tests at farious frequencies are recommended. Three tests (max. clock, average clock and min. clock) give you a good idea of what is going on!

Here my inital results investigating the impact of cpu choice and FSB setting. I will later also add multiplier settings (I can select 2.0 to 5.5 on my K6-2+) and investigate if it's worth it or not (the FSB scaling seems to be much more "cost effective".

You can really see that:

- When L1 and L2 are disabled, clock speed and FSB have little impact on performance. Two tests (One with max. clock and one with min. clock) are sufficient to give you an idea what "slowest machine" you can hit with your setup.

- When only L2 is enabled, clock speed and FSB have do have an impact on performance. So for this setting, a few tests at farious frequencies are recommended. Three tests (max. clock, average clock and min. clock) give you a good idea of what is going on!

I thought fastvid only helped the Pentium Pro and that the Pentium II didn't exhibit the same DOS issue.

bushwack wrote:

I thought fastvid only helped the Pentium Pro and that the Pentium II didn't exhibit the same DOS issue.

Yea I don't know, but it had a massive impact on my PIII system:

On a PIII 500 GHz the score improved from 41.2 to 68.3

Question:

Been looking at the results and found something interesting. When L1 cache is disabled I can use the L2 cache on the SS7 system to boost performance (basically toggle between a 386 and 486).

However I can't do the same on the Slot 1 system. I believe this is because when you disable L1 cache, it might also disable L2 cache, even if you have L2 cache set to enabled.

The scores are identical for every single test...

So I was wondering if there is a way (maybe a software utility) to disable the L1 but leave L2 running.

Or are there Slot 1 / S370 boards that have L3 cache on the motherboard?

The K6-2+ is similar. It has on die L2 cache. The L1 setting in BIOS disables all on chip cache. The BIOS calls them internal and external cache.

Ran into my first game that gave me issues despite slowing down the machine with all my tricks.

The Heart of China would run fine with PC speaker but with SB it would give me a "Unable to initialize" message and googling tells me it's speed related. So I tried this in DOSBox and yea after around 20000 cycles you get this message. 20000 isn't much it's a 486DX2-50 or something...

But one time the game did play and I have no idea why. So I though well it's worth a shot and pulled out the AWE64 card and put in a cheapy clone card (ESS AudioDrive) and it worked right away. Quit the game. Tried again. Yup works. No idea what's up with the AWE but there is likely more to it...