infiniteclouds wrote:

This sounds really interesting. I'm not familiar - could you please expand on this? Is this through MTRRLFBE? Or some sort of BIOS setting?

No, it's done either through a motherboard jumper setting or BIOS setting. Maybe a DOS tool exists for this as well, but I haven't come across one.

infiniteclouds wrote:

Also, as I have an Ezra CPU (866mhz) and a 6BXC on the way I'm curious about the amount of SetMul options for it. It has less th […]
Show full quote

Also, as I have an Ezra CPU (866mhz) and a 6BXC on the way I'm curious about the amount of SetMul options for it. It has less then the Pentium I think?

VIA C3 =
L1D
L2D
ICD
BPD

No VPD, CCD or DCD settings for VIA?

Congrats; with that board and CPU, you can control FSB from DOS using Rayers SMB util, and CPU multiplier using Setmul. Even though you have what seems to be less parameters to control CPU speed compared to TR12 registers, you have enough already to scale smoothly, and with a higher top-end. Your 866MHz should easily clock to 1GHz, BTW.

Besides, the Ezra has more multiplier choices, and with the aforementioned parameters, covers a smooth and wide CPU speed range.

j^aws wrote:

Congrats; with that board and CPU, you can control FSB from DOS using Rayers SMB util,

Wow! I knew the CPU and board were flexible -- with a hidden jumper setting for 50FSB -- but controlling the FSB from software? That's really awesome. The other reason I'm excited about this build is how easy it is to swap CPUs in and out. Not having to remove/reattach/reapply heatsinks and thermal paste every time I want to test or use a new CPU makes me wish all CPUs came on cards!

Nope, but you have multiplier settings.

Despite having the External Cache Enable/Disable option in BIOS there isn't any motherboard/L2 cache... I don't think there are any when it comes to Slot 1s? There are 6 variables in terms of switches (including none).. they are, in order of fastest to slowest...

1) Nothing/All Enabled
2) BPD
3) ICD
4) ICD+BPD
5) L1D
6) L1D+BPD

Adding ICD to 5 or 6 yields no changes. Still, with 6 settings, 8 FSB speeds and 16 multipliers it's a little over 700 combinations -- it's going to take a while.

I see the main goal of this thread and spreadsheet as getting accurate 'performs like' settings of our different builds with caching and down-clocking tricks. We can see SpeedSys ratings are useless by themselves when trying to match another system -- Clueless sorts the benchmarks by Doom FPS for a reason... but I don't think it is totally random, either. Also, Doom seems to be pretty weird about what it likes and what it doesn't -- I get better Doom FPS on my K6-III+ 550mhz then I do on my 2.4ghz Athlon 64 4000+, for instance.

FSB plays a big part in the differences and the real legacy systems are using 33FSB or less, after all. In the build I mentioned at the start there are several ways to achieve the same clock speed.

I haven't done any Doom bench-marking yet but take 600mhz for example and look at the differences in results -- all on the same CPU.

@600mhz (133 x 4.5) SpeedSys: 315.45 PCPBench: 80.4 3DBench2: 320.8 […]
Show full quote

@600mhz (133 x 4.5)
SpeedSys: 315.45
PCPBench: 80.4
3DBench2: 320.8

@600mhz (100 x 6) SpeedSys: 314.38 PCPBench: 70.1 3DBench2: 275.2 […]
Show full quote

@600mhz (100 x 6)
SpeedSys: 314.38
PCPBench: 70.1
3DBench2: 275.2

@600mhz (75x8) SpeedSys: 314.37 PCPBench: 60.9 3DBench2: 236.6 […]
Show full quote

@600mhz (75x8)
SpeedSys: 314.37
PCPBench: 60.9
3DBench2: 236.6

@600mhz (66 x 9) SpeedSys: 315.11 PCPBench: 55.4 3DBench2: 215.1 […]
Show full quote

@600mhz (66 x 9)
SpeedSys: 315.11
PCPBench: 55.4
3DBench2: 215.1

@600mhz (50x12) SpeedSys:313.56 PCPBench: 45.6 3DBench2: 175.1 […]
Show full quote

@600mhz (50x12)
SpeedSys:313.56
PCPBench: 45.6
3DBench2: 175.1

Notice that even as the PCP and 3D2 scores almost half from the top marks to the bottom that the SpeedSys ratings float right around the same area. The closest benchmark to this SpeedSys on our tables is

Clueless' Pentium II 266 (66x4) SpeedSys: 309.88 PCPBench: 77.3 3DBench2: 195.0 […]
Show full quote

Clueless' Pentium II 266 (66x4)
SpeedSys: 309.88
PCPBench: 77.3
3DBench2: 195.0

Compared to my 66FSB setting that yields a slightly faster SpeedSys the Pentium II has ~-20FPS in 3D2 and ~+20 in PCP. I don't use FASTVID or MTRRLFBE though in any of my benchmarks. It can be hard to get a good comparison when that isn't a constant throughout other benchmarks.

Here's another comparison --

Clueless' K6-2 @ 366 (66x5.5) SpeedSys: 419.60 PCPBench: 79.8 3DBench2: 246.7 […]
Show full quote

Clueless' K6-2 @ 366 (66x5.5)
SpeedSys: 419.60
PCPBench: 79.8
3DBench2: 246.7

My 800mhz benchmarks all match this SpeedSys rating closest, at ~420. Here are those VGA results:

(133x6) PCPBench: 93.6 3DBench2: 367.2 […]
Show full quote

(133x6)
PCPBench: 93.6
3DBench2: 367.2

(100x8)
PCPBench: 79.5
3DBench2: 308.4

(66x12)
PCPBench: 61.1
3DBench2: 234.7

The 100FSB has a nearly identical PCP benchmark -- but the 3D2 score is much higher. Again, perhaps using MTRRLFBE would boost both the 66x12 scores up to be a very close match on both.

Please help and share if you find any patterns within the benchmarks we have so far and/or within your own!

@infiniteclouds -- yes, I sort by Doom because it's a real game, with real benchmark results (ie, not a 'fly-by'), free, and easy to run. Unfortunately, there is no single benchmark that will show how each system will perform exactly with various caches disabled. Different games/programs/benchmarks are affected differently. I've done some internal testing that include other benchmarks (Wolf3D, Ultima VII, Descent II, etc) and there is enough variation when you disable caches that even when looking across multiple game benchmarks, it's impossible to have a single, consistent conclusion. So to avoid complicating things needlessly, I just sort by Doom. 😉

Good point about FastVID and MTRRLFBE. To be clear, I never use those in any of these benchmarks (on the systems that would support them), and I hope no one else is. Although, I'd think the results would pop out of the spreadsheet if someone did.

One last observation I've made: when you disable caches on a system, it also tends to equalize the VGA performance. In other words, let's say you have a very fast PCI VGA and a very slow one. You bench them both at full system speed and note the big difference in VGA-related benchmarks. Now disable caches on the same system and retest. What I've noticed is that both graphics cards get the same benchmark results when you disable caches because they are so bottlenecked by the system without caching.

EDIT: I added the reference results to your tab of results, infiniteclouds. That was the other reason I stuck with Doom. I needed a lot of old reference results, so I used Phil's Ultimate VGA Benchmark sheet, and those were the benchmarks he used (minus speedsys).

clueless1 wrote:

One last observation I've made: when you disable caches on a system, it also tends to equalize the VGA performance. In other words, let's say you have a very fast PCI VGA and a very slow one. You bench them both at full system speed and note the big difference in VGA-related benchmarks. Now disable caches on the same system and retest. What I've noticed is that both graphics cards get the same benchmark results when you disable caches because they are so bottlenecked by the system without caching.

You mention a fast PCI vs a slow one... but do you suspect this would be the case as well with an ISA card vs AGP/PCI? Or do you think that using an ISA card could drop results even lower once already running at the slowest speeds with all caches and switches disabled?

That was the other reason I stuck with Doom. I needed a lot of old reference results, so I used Phil's Ultimate VGA Benchmark sheet, and those were the benchmarks he used (minus speedsys).

That makes sense. I'm just trying to figure out what's going on behind the results that yield similar or identical Doom FPS but very different SpeedSys ratings -- sometimes even on the same chip with different TSR configurations. Are there any other DOS games from the 386/486 era that have a timedemo?

@infiniteclouds --

I don't have any ISA cards, and don't know if I ever had, even back in the day. If I had to guess, I'd say little to no difference in performance between ISA and PCI once caches are disabled.

Wolfenstein 3D has a timedemo. Phil used to have it in his DOS benchmark pack, but removed it. I have it archived though. 😉 I believe there are valid reasons he removed it:
-possible licensing issues
-different versions of Wolfenstein executables will run different timedemos, making comparing results between different people challenging.

infiniteclouds wrote:

I'm excited about this build is how easy it is to swap CPUs in and out. Not having to remove/reattach/reapply heatsinks and thermal paste every time I want to test or use a new CPU makes me wish all CPUs came on cards!

I couldn't have been more wrong about this. The actual Slot 1 cartridge CPUs go in and out very easily... but once you pop one of these slotket adapters (MS6905) into there it is basically a permanent addition to the motherboard! The CPU and heatsink are attached to it already, of course... but it will not come out.

I've added my results for my Ezra - 714 combinations. I put them in a separate tab but you might want to change the formatting or whatever.

I believe there is some sort of pattern between all Socket 7 through 370 PCPBench results and actual speed in games that seems pretty consistent. I think in general though you're right and you can't create a setting for a specific 386 or 486 machine using caches and downclocking like I was hoping but rather have to tweak them for each individual game.

Athlon 64 x2 5050e 2.6 ghz @ 560 mhz (4 x 140) with l1 disabled.
pcplayer: 4.2
3dbench2: 16.2
doom: 12783 rt
quake: 2.4

High 386 scores.

Ultima 7 with the guard-walking-around benchmark of clueless1. I let him do 10 laps and tallied up the seconds. Sound was disabled, no compatible sound card on these.

K6-2 300
base: 5 s. / 230 fps
no l1: 28 s. / 41 fps
no l1/l2: 82 s. / 18 fps

Athlon 64 2.6 ghz @ 1.4 ghz
base: 5 s. / 230 fps
no l1: 41 s. / 28 fps

The athlon 64 with no l1 has it about right. The k6 is a bit off in both directions.

I tested my Core i7 980X @ ~4,3GHz (22x195) without L1 😀
3DBench 1.0c : 14.4 FPS
PCPlayer (normal/NOT 640x480) : 3.5 FPS
Doom (max details) : 12247 realticks
Doom (min details) : 3034 realticks
^Not sure which Doom setting is valid, so I tested both 😉

Speedsys 4.78 (extended memory would take too long to test, 64MB used 12GB in total) : 13.79

I'm using GeForce 7800 GTX (PCI-e), no FASTVID/MTRRLFBE/EMM386

EDIT :
Default scores (same settings, minus SetMul) :
3DBench 1.0c : 366.4
PCPlayer (normal/NOT 640x480) : 476.6 FPS
Doom (max details) : 645 realticks
Doom (min details) : 142 realticks

Speedsys 4.78 : 8615.75 (:D)

Same GPU, no FASTVID/MTRRLFBE/EMM386.

vvbee wrote:

Ultima 7 with the guard-walking-around benchmark of clueless1. I let him do 10 laps and tallied up the seconds.

The otherwise good athlon machine can't do sound in dos so it's not much use with this game. With l1 disabled, the k6 is still slightly too fast, and just lowering its multiplier a bit isn't enough. No doubt you can reduce the game speed some by downclocking more notably, but when you have to do it with jumpers it's a pain in the long run. A pentium 133 in the same machine with just l1 off seems decent, but might just run the k6 at 66 fsb or so, the pentium isn't as interesting of a cpu for me.

k6-2 @ 300: 41 fps
k6-2 @ 250: 41 fps (2.5 x 100)
p @ 133: 33 fps
p @ 100: 30 fps

Went ahead and put back the k6-2 to run it at 66 fsb.
k6-2 @ 200: 29 fps (3 x 66)
k6-2 @ 300: 31 fps (4.5 x 66)

Would indeed need to have the k6 at 66 fsb for this game. Well, not too bad.

clueless1 wrote:

I don't have any ISA cards, and don't know if I ever had, even back in the day. If I had to guess, I'd say little to no difference in performance between ISA and PCI once caches are disabled.

EZRA 866 @ 150mhz L1D+BPD (the slowest setting for this CPU)

Voodoo 3 2000 AGP = 8.6 3DBench
Tseng ET3000AX ISA = 7.0 3DBench

Voodoo 3 = 2.6 PCPBench
Tseng = 2.4 PCPBench

Both give approximately the same FPS (about 3.8 ) in Doom... but that 1.6 difference in 3DBench and .2 in PCP make all the difference in a game like Isles of Terra (Might & Magic III) where the animations were too fast on the Voodoo but perfect on the Tseng with the CPU/caches being equal. It seems an ISA video card is definitely useful for this kind of a machine. Now if I can just figure out a way to have them both in there at the same time and toggle between them...

@infinitecloudds

Cool! So I guess "little to no difference in performance" was not too far off. 3dbench is less demanding than PC Player, so the older the game, the more difference you'll notice.

Right -- for older games it makes a big difference with Wing Commander 1 and 2 seem about 25% slower than the Voodoo card with the same CPU/cache settings. I imagine an 8-bit VGA card would slow things down even more. In any case, if the plan is to use a slowdown machine to also play games from the 386 era and you have an extra ISA slot it's not a bad idea to get an older VGA card as well. I'm hoping there is a way I can keep both the ET3000 and Voodoo 3 in there at the same time but I'm not sure how to tell the system to ignore one of the ISA expansion slots at boot.

As an aside.... with the CPU maxed (133x7.5) and cache enabled the Tseng bottleneck really shows. PCP goes from 103 FPS to about 20. 3DBench goes from 402.6 FPS to 25, Doom goes from 119 down to 8.1 FPS.... and oddly enough Quake while still way slower than the Voodoo actually pulls off 20 FPS - albeit with a bit of static in the picture. This Tseng doesn't actually support VESA modes so it won't do anything above 360x480 in 3D... which suppose makes sense for a card I believe is from the late 80s? I'd still like to know why Quake performs so much smoother than Doom.

On a sidenote:
I recently did some comparisons with cache disabled and kept it disabled when finished for more tests later on. (P3 sig in my sig)
Well, 2 months later I wanted to run some Unreal and was surprised. Without L2 Cache the game actually ran pretty fine. around 40 fps for the most part.

Gotta run the rest of the benchmark suite next on it and hand out a chunk of numbers when finished. (both enabled, L1 disabled, L2 disabled, both disabled, multiple clock speeds)

I have one system I could test this with. What are the doom settings you are suppose to use? Do you use full screen or with the hud like you normally play? I also heard that even the control method you set in setup can change the results like mouse + keyboard is slower than just keyboard.

Baoran wrote:

I have one system I could test this with. What are the doom settings you are suppose to use? Do you use full screen or with the hud like you normally play? I also heard that even the control method you set in setup can change the results like mouse + keyboard is slower than just keyboard.

Full screen. If you download Phil's DOS benchmark pack, it makes it simpler. I run these with extended memory (no EMS) and keyboard and mouse. Thanks!

I added some benchmark results to the bottom of the spreadsheet. One reason why I started testing the system was because I wanted to know how much graphics card really matters when you have disabled caches. That is why there is double tests with 2 different graphics cards.
Basically that trident graphics decelerator turned my 486 into a 386 when it comes to most games. I think using a slower graphics card can be used to fine tune system speed when trying to run old games.
I used bios default settings with the tests.