I can't wait when computer case manufacturers will start making decent compact double motherboard cases. Then you just put 2 systems in one on the 486DX2-66 and one on Pentium III 800Mhz. After that you are covered on all games from 1994 to 2002.

clueless1 wrote:

So maybe I could take the average 386/40 from Phil's chart and place that into this chart so someone knows what a typical 386/40 scores in specific benchmarks.

I'd sorted results by some param. Then made a search to find a last "heap" of close results for that CPU. By this way you may find best typical results, instead of abstract average. For example, Pentium 100 in Doom has 57 fps by this method.

Tertz wrote:

clueless1 wrote:

So maybe I could take the average 386/40 from Phil's chart and place that into this chart so someone knows what a typical 386/40 scores in specific benchmarks.

I'd sorted results by some param. Then made a search to find a last "heap" of close results for that CPU. By this way you may find best typical results, instead of abstract average. For example, Pentium 100 in Doom has 57 fps by this method.

Nice idea. I'm just not sure how to duplicate this "heap". Maybe Median is good enough? The median P100 score is 54 (average is 50).

Maybe you can fold your data and Phil's together. Cache tweaking seems like a logical extension, and this will keep those results in perspective.

gdjacobs wrote:

Maybe you can fold your data and Phil's together. Cache tweaking seems like a logical extension, and this will keep those results in perspective.

I've thought about that, but I don't know how to do that without creating a ton of work, and I'm trying to keep this simple and easy for people to participate in, or no one will bother.

What I am imagining is actually using a small subset of Phil's numbers: just median results of common 386-486 performing cpus. Of course, there has to be enough data points in a specific processor/speed to use. And what I'd do with these values is simply use them as guideposts among real data so users could tell what their fast cpu with disabled L1 cache performs like.

Something like this. The red values are median results taken from Phil's chart.

clueless1 wrote:

Something like this. The red values are median results taken from Phil's chart.

test.jpg

Yeah, that should work nicely. As it's data for a real machine and not a composite mean value, you can also include specifics in a footnote. That way people can refer back to the appropriate entry on Phil's table.

gdjacobs wrote:

clueless1 wrote:

Something like this. The red values are median results taken from Phil's chart.

test.jpg

Yeah, that should work nicely. As it's data for a real machine and not a composite mean value, you can also include specifics in a footnote. That way people can refer back to the appropriate entry on Phil's table.

Actually, the red results are composite values (medians, though, not averages). edit: what I mean is I used a median() formula, so each result is not necessarily from the same machine, and some results might be a median between two results depending on whether there was an even or odd number of candidates. But I think I like where you're coming from. I could take a real result, that's closest to the median for that CPU, to represent that CPU in this table. I wonder which would be preferred (a real result or a theoretical median).

clueless1 wrote:

gdjacobs wrote:

clueless1 wrote:

Something like this. The red values are median results taken from Phil's chart.

test.jpg

Yeah, that should work nicely. As it's data for a real machine and not a composite mean value, you can also include specifics in a footnote. That way people can refer back to the appropriate entry on Phil's table.

Actually, the red results are composite values (medians, though, not averages). But I think I like where you're coming from. I could take a real result, that's closest to the median for that CPU, to represent that CPU in this table. I wonder which would be preferred (a real result or a theoretical median).

The median will be the value of a real result for an odd number of entries. Unless you have results concentrated at either end of the distribution, the median result should be fairly central.

Yeah, you beat my edit in the previous post. Sorry about that.

clueless1 wrote:

Yeah, you beat my edit in the previous post. Sorry about that.

No problem. I've been robo posting or else I would have moved the topic to your other thread. I'll pick it up there.

clueless1 wrote:

I envision having a number of "reference" benchmarks scattered throughout. So maybe I could take the average 386/40 from Phil's […]
Show full quote

I envision having a number of "reference" benchmarks scattered throughout. So maybe I could take the average 386/40 from Phil's chart and place that into this chart so someone knows what a typical 386/40 scores in specific benchmarks. And the average 486/25, etc.
....
Something like this. The red values are median results taken from Phil's chart.

test.jpg

Well, i had nearly the same idea a few months ago. I was already buying & building up my Super-socket system, little inspired from Phil's video 'time-maschine', and i also want to know, what is possible with my K6-2(CXT) even in comparison with the properties Phil has used.

To compare my underclocked K6-2 with the three modes (L1+L2 On, L1 Off and L1+L2 Off), i've extracted the most reasonable data from Phil's VGA chart for the following CPU's and average them. For fast comparisons, I've compared only the two benches 3DBench2 & PCPBench:

386DX-25, 386DX-40, 486DX-25, 486DX-33, 486DX2-66, 486DX4-100, Pentium 60

I've also compared the values of Phil's test values, when he was running the WingCommander Comparison with 386DX-25/-33/-40 and corresponding 3DBench2 & PCPBench values.
Additionally, i've also added Phil's test values from his video 'time-maschine' and the underclocked K6-III+ with 133MHz for the '386-mode' and '486-mode'.

(The values for 386DX-16 and 386DX-20 are interpolated from the average value of DX-25 and have to be seen carefully.)

Finally, i've compared these values with my benchmark results, and i've made additionally the WC-test. All tests i've made at the lowests clockspeed. Additional tests with higher clockspeeds were only with full enabled caches.

I know this is maybe not exactly accurate for statistics, but it gives a approximate number for example which speed is necessary to play WC.. 😎

Great stuff, really awesome to see that the database is being put to good use 😀

Using it as a reference was my main goal when I started the project.

PhilsComputerLab wrote:

Great stuff, really awesome to see that the database is being put to good use 😀

Using it as a reference was my main goal when I started the project.

Awesome. Glad to have your blessing. 😀

melbar, very interesting! We've definitely got the same process in mind. 😀

clueless1 wrote:

I'm just not sure how to duplicate this "heap". Maybe Median is good enough?

5 pcs of P100 have 57 fps. As this is real numbers and "heap" represents kind of typical numbers, - I'd choose this. Anything lower may be thought as bad configs.

That might be the better way. Create a ten cell binning of the numbers, strongest bin wins.