Congratulations! You've optimised your system.

Refering again to the Ultimate 486 Benchmark Comparison, your numbers are right were they should be for an AMD X5-160. The good news is that your motherboard is not adding an automatic 2/3 multiplier to your PCI bus. I can tell this from your PCPbench score, which is heavily reliant on the PCI bus.

Speedsys and pcpbench are just two tests and are not good average representations of CPU performance. Test your system in various games or in Windows. If you want more CPU power, you need to consider a POD100 or an IBM 5x86c. Check to see if your motherboard has 3 jumpers for the FSB. If it does, there's a good chance it has undocumented FSB settings for 60 and 66 MHz. I can only hope that your motherboard/BIOS adds a 1/2 or 2/3 multiplier for 50/60/66 MHz FSB operation. If you are able to get an IBM 5x86c running at 2x66, expect your Speedsys L1, L2, RAM to jump to 190, 73, 56 MB/s, respectively, or 195, 56, 44 MB/s for a POD100.

Thanks! 😀

As a nice piece of late night thoughts, I've already read some threads on these CPUs and systems... and something looks quite weird about PCPbench.
and its scores on different systems. This one,
UMC chipset PCI 486 mobo. AMD P90 CPU options?

> I've got it working with the P75 chip, at 200 MHz.
> ATM, PcpBench gives me 11.7, with the RAM Read setting = 1. If set to 0, it becomes unstable.

The rocking Speedsys results for this setup are 74.99 for the cpu and 56.97 for the memory. But where in the heaven is the bottleneck that makes it just 11.7 in PCPbench. The 'typical' AMD@160 system that I've launched gives 11.3.
The CPU increase is clear 25% here, the memory increase is also considerable. So there must be some very nasty 'bottleneck' that hinders it all for PCP.

Maybe I was jumping to conclusions that the video cards' speed loses meaning here and the CPU is what's important.
Since my last memory throughput increases were yielding nothing already for the PCPbench score, it's not the memory that hinders progress and not the CPU really... that +25% is something to make a difference!
So what's that mystical hindrance might be?

Yes I've considered the POD (83 variation). It needs to be a stably overclockable POD, otherwise it gives the score of 60 exactly as AMD@160... The memory score of 37.32 is, well, not the performing of anime, vampire type of activity but isn't too nice, either. Sure it has the Intel FPU! Too bad it's great for 3D graphics only. and even the full P1 boxes tend to su.. perform poorly with 3D (even the 233MMX.) Some P-II 400 types seem to do the job, mainly.

So of the more reasonable 486 challenges, the PCPbench remains undefeated, no one seems to get the meagre 12 score.
what kind of Slow spell does it cast upon one's computer?

*
it's not the FPU that's slow (POD would beat all other scores!) and not the raw CPU power (otherwise AMD@200 would fly) and not the memory (we've seen very good scores already, to no avail.)
slow VGA cards afterall?

Can you let me know what PCPbench score you get at 133 MHz?

I beleive rg100 was not using this setting when he got that score: VESA Modus 100 (640x480 8bpp LFB). He ended up with a score of 9.0 with an X5-200 using a Matrox Millennium G200. I have not bothered to investigate if it is the graphic card slowing things down. PCPBench is only 1 benchmark. I get 9.6 at 160 MHz when everything is optimised on a UMC board. What benchmark score do you get in Quake1 at 320x200 and at 640x480?

Many factors alter benchmark scores, but you'd probably need to dig into the source code and various compiler optimisations to really understand what is going on with PCPbench.

Generally, you have these to track down:

Chipset - this is where the cache and RAM controllers are located; how fast can it communicate reliably?
CPU - what architecture is used?
BIOS - what cache and memory wait states are added? What other PCI- and CPU-based features are set in the BIOS?
FSB frequency - what speed is set? This determines how fast the CPU communicates with the memory.
PCI BUS frequency - the frequency the graphics card is running at and communicating with the CPU/memory
Memory - EDO, FPM, and what speed rating?
SRAM Cache - what speed rating and how much cache? Cache can only cache so much memory.
Graphics card - what card is used and how does it interact with the benchmark program?
Benchmark program - what type of algorithms are being run and to what degree does it rely on all the above?
Compiler - what optimisations are used?

One comment on finding a POD which will work well at 100 MHz is to try and modify the onboard voltage regulator so that the chip runs at a slightly higher voltage. This task has been placed on my vast list of 486 hardware optimisations to attempt. I'll get to it one day.

Mithloraite wrote:

Thanks! :) […]
Show full quote

Thanks! 😀

As a nice piece of late night thoughts, I've already read some threads on these CPUs and systems... and something looks quite weird about PCPbench.
and its scores on different systems. This one,
UMC chipset PCI 486 mobo. AMD P90 CPU options?

> I've got it working with the P75 chip, at 200 MHz.
> ATM, PcpBench gives me 11.7, with the RAM Read setting = 1. If set to 0, it becomes unstable.

The rocking Speedsys results for this setup are 74.99 for the cpu and 56.97 for the memory. But where in the heaven is the bottleneck that makes it just 11.7 in PCPbench. The 'typical' AMD@160 system that I've launched gives 11.3.
The CPU increase is clear 25% here, the memory increase is also considerable. So there must be some very nasty 'bottleneck' that hinders it all for PCP.

Maybe I was jumping to conclusions that the video cards' speed loses meaning here and the CPU is what's important.
Since my last memory throughput increases were yielding nothing already for the PCPbench score, it's not the memory that hinders progress and not the CPU really... that +25% is something to make a difference!
So what's that mystical hindrance might be?

Yes I've considered the POD (83 variation). It needs to be a stably overclockable POD, otherwise it gives the score of 60 exactly as AMD@160... The memory score of 37.32 is, well, not the performing of anime, vampire type of activity but isn't too nice, either. Sure it has the Intel FPU! Too bad it's great for 3D graphics only. and even the full P1 boxes tend to su.. perform poorly with 3D (even the 233MMX.) Some P-II 400 types seem to do the job, mainly.

So of the more reasonable 486 challenges, the PCPbench remains undefeated, no one seems to get the meagre 12 score.
what kind of Slow spell does it cast upon one's computer?

*
it's not the FPU that's slow (POD would beat all other scores!) and not the raw CPU power (otherwise AMD@200 would fly) and not the memory (we've seen very good scores already, to no avail.)
slow VGA cards afterall?

In general I think the 486 platform has very limited capabilities to be a "good" 3D performer. Of course "good" needs to be defined but using a Socket 7 system makes things much much easier and less frustrating if your goal is not manipulating a system over and over again but to play the games troublefree.

feipoa wrote:

Can you let me know what PCPbench score you get at 133 MHz?

I beleive rg100 was not using this setting when he got that score: VESA Modus 100 (640x480 8bpp LFB). He ended up with a score of 9.0 with an X5-200 using a Matrox Millennium G200. I have not bothered to investigate if it is the graphic card slowing things down. PCPBench is only 1 benchmark.

Sure PCPbench it's just one of many, but it appears to be quite mysterious. Even with all 'main resources" in the experiments already made here being present in abundance (CPU& FPU power, memory bandwidth, VGA performance) it manages to get real stuck at some 11 score.
Could it be not a "bad optimization" but a sign of some real flaw in the mentioned 486 systems... Something that we might identify and even overcome/improve?

Because the sum of CPU-FPU-memory-VGAperformance looks quite simple. Yet the mysteries! 😀

The improvement of my setup was as follows:

1. AMD@133 w/slow default memory & bios settings: 49.46 CPU, 7.6 PCPbench.
That also included WB L2 caching w/256kb cache (in no way enough to cache the 64mb RAM completely). Also I read about stuttering and slowdown in some games with this method mentioned in some threads...

2. w/all better performance bios routines enabled and also WT caching: 9.0 PCPbench. Neat!
Memory is still at all defaults.

3. AMD@160: 59.85 CPU, 10.9 PCPbench. That's with Matrox Mil-2 PCI 4mb.
Diamond Viper Riva 128 4 mb. gives 10.8 (slower card or?..)

4. @160 and a slight increase in memory, at last (throughput of 32.9 instead of the default 30). That's 11 on PCPbench. (Diamond)

5. same CPU, with fastest memory, throughput of 47.7: PCPbench 11.3
Both video cards produce it.

So what does seem missing out of this and 200Mhz or POD@100 system that doesn't allow 12?
to any of these machines.

P.S.
Besides VGA, cache might be some reason maybe. But I've seen it can be not at all fruitful to performance-
by udam_u on Quake bench result by RG100:
"I'm sure that your system is faster than mine. You have reached better result in Quake which is more impartial than any other simple benchmark like 3Dbench and Shiny. How much cache have you got? I have got 512kB cache in my rig."
and RG100 was using just 256 cache

What PCPBench score do you get with the AMD X5 at 133 MHz with BIOS-optimised cache/memory?

I did a few tests on my 486 and indeed the memory/cache performance did not influence the PCPBench score.

Board: Asus PVI-486SP3 (SIS 496/497)
CPUs: Intel DX4 @120 MHz, AMD x5 @ 160 MHz
VGAs: S3 Virge DX/GX, Matrox Millennium, Nvidia Riva TnT, Tseng ET6000

No matter which VGA and CPU I use, the PCPBench score stays around 10.5 FPS on my board. Real games do perform considerably faster on the 160 MHz AMD than the 120 MHz Intel though.

@Mithloraite: Congrats, these are fantastic scores for a SIS 496 board! 😀

I'm not sure it would be worthwile squeezing any more out of a 486, since a Pentium at a similar clock speed yields about twice as many FPS in PCPBench.

Try running PCPbench with an AMD X5 at 3x50 and ensure that the PCI bus is run at 50 MHz. I have found that PCPBench is heavily PCI bus speed reliant.

You are right about the bus speed dependency. Unfortunately I'm not able to test 50 MHz on my board since PCI/VLB stops working at that speed.

I have a 496/497 board running fine at 50mhz fsb. But what it gains in PCI performance it loses in cache/memory performance because I had to raise the cache waitstates. I think this board would fly if I had nice 12ns cache modules instead of my 12ns tag/15ns data combo.

FGB wrote:

I think this board would fly if I had nice 12ns cache modules instead of my 12ns tag/15ns data combo.

That is what I originally thought too, but it didn't seem to help at all. At 50 MHz, I still had to slow my cache down to the slowest setting, and I was using 10ns TAG, 10ns cache. This was on a UMC board. Switching to 12 ns TAG, 15 ns cache also required the slowest cache setting. I know the 10 ns cache performs well at high speed because I use the same TAG piece on a 430TX board at 83 MHz. I'll play around with my SiS boards more when I finish up the 686 benchmark comparison and find a case for it. I need to check it for SCSI bus mastering in W2K. If it passes this test, I may rate it above my priced UMC boards, esp. if it works with all Cyrix 5x86 special features.

feipoa wrote:

What PCPBench score do you get with the AMD X5 at 133 MHz with BIOS-optimised cache/memory?

Sure I will perform this test and also will have to add a couple of the ubiquitous Virge DX cards to the results. Not too much is to be expected here probably (besides the compatibility) but one of these has 35ns memory. Surplus or?..
Just currently moving the system to another case.
Please be aware of these AT boxes...

They tend to look nice, UTT seems to have made millions(?) of these and the plastic quality is OK (very white). But the metal frame is not firm or thick enough. With my sample the geometry gets easily incorrect so even plugging cards is troublesome... due to measurements slightly displaced.

Another type is seemingly much more firm and correct of form 😀

feipoa wrote:

What PCPBench score do you get with the AMD X5 at 133 MHz with BIOS-optimised cache/memory?

It's time to fix this pending question now 😀
Also some experimental data on other video cards beside Millenium II & Riva 128 is added.

With all optimized settings for bios, cache and memory the result for Speedsys is

the PCPbench score being 9.4

Launching AMD X5-133 P75 as P90 @160MHz gives it a lift of 1.9 to the score of 11.3
adding +20% to CPU overclocking adds exactly +20% to PCPbench 😀

This test could be the next after stage 2 (with 9.0 score) where memory was not optimized. It leads to the notion that the memory optimization by itself gives mere 0.4 fps. But it allows for something better with better CPUs, up to a certain limit probably present in the chipset.

*

As for the Virge cards, these are not a complete loss. The slower one could produce 10.0 under Win95 and 10.4 under DOS.
The two later models both achieve 10.9 in Windows and "full" 11.3 under DOS.

Also noting that all cards are hindered under Windows. Was it made to make things slow? 😀 In this case Millenium 2 scores 11.1

*

Since there are 4 cards hitting the same limit of 11.3 fps we may think it's not a video card problem.

Thank you for posting that. It seems the SiS pcpbench scores are about 1.5 points better than the UMC scores. Less PCI wait states?

feipoa wrote:

Thank you for posting that. It seems the SiS pcpbench scores are about 1.5 points better than the UMC scores. Less PCI wait states?

yes. Something good is obviously present here. (like " a small aura of good" in AD&D games). But what is the nature of this advantage, now, might not be perfectly clear. I've carefully checked this thread and the real high overclocking result seems actually quite simple and predictable. It's with 180 (not 200MHz)
UMC chipset PCI 486 mobo. AMD P90 CPU options?

If I'm not mistaken it was an UMC board and the score is *12.5*. This would account for the chipset being not quite important... for the +12.5% CPU overclocking (from 160 to 180MHz) we do see a simple 12% PCPbench increase.

So if I could add this type of CPU overclocking to my CPU, this would be 11.3fps +12.5% so I could expect some 12.7 fps. It's simple and no secrets are around the PCPbench or the chipsets as it may seem. To sum it up, only a very fitting board might do it with the divider of 3/2, 60 MHz for cpu, 40 for PCI/memory. Such boards are rare! 😀

It would mean that if I come across such a board or a CPU that is stable _at least_ with 50Mhz bus... That might be a chance to try something of interest. Simple 50x3 CPU with 33Mhz for PCI/memory will probably be the same as 160Mhz w/ 40 PCI/memory bus or slightly worse.
Only overclocking it further to 60x3 with 40 memory bus can do a good trick.

I possess a few boards that do have a divider but just a standard 1/2 one. It's all useless and leads solely to despair 😀

So where are all the "P75+" CPUs might be?.. Those that work just like that with 50MHz bus and overcolockable 😀
I mean the real ones, not repainted by the Chinese as "plus".

P.S.
this particular LS board has 2 jumpers for the bus settings; so the useless bus of 50MHz seems to be the real technical limit here.

It is important to check which mode PCPbench is run in to properly compare the scores. The results will be very different depending on the mode.

feipoa wrote:

It is important to check which mode PCPbench is run in to properly compare the scores. The results will be very different depending on the mode.

to be perfectly exact (if such thing exists at all 😀) let me quote RG100

"Also, PcpBench mode 100, in NOT LFB mode = 11.8!
Edit: PcpBench (mode 100, not LFB) = 12.5! That's if I set the cache timings to 2 1 1 1."

what's the deal with this LFB... some buffering afair but what is its role I didnt' research... Is it slow or quickens things, for starters? 😀

For the user who posted the photos of the Chaintech 486SPM, did you solder on the PS/2 DIN connector yourself? Do PS/2 mice function on this motherboard? Also, did you solder on the coin cell battery without removing the charging lead?

I'll reattach your photo below. It no longer seems to be available.

Does anyone have the actual manual for this motherboard? The jumper settings as listed on stason are incomplete. EDIT: it appears that there are several variants of the "486SPM" made by different manufacturers with different layouts.

http://www.elhvb.com/mobokive/Archive/Chainte … 4spm/index.html

That helped some, but not completely. Does anyone have a high resolution top-down image of this motherboard so I can see where all their jumpers are, that is, for a known working motherboard.

When I got this board, the jumpers had been scrambled every which way and nothing made sense. While I followed the online literature the best I could, I could not determine where JP46 and JP47 should be set at. Stason lists these as "Factory configured - do not alter", but does not note what position they should be in. JP46 & JP47, according to Stason, are the two jumper headers above JP32.

It was also not clear to me if JP1 and JP2, which set the DMA channel, can be left floating.

Unfortunately, I have been unable to get this board working. The barrel battery has been removed and only very minor signs of acid are apparent. I tried using an external battery, those bulky 3.6V Lithium batteries you sometimes found on 386-era boards, but the boad did not turn on. The manual also does not mention, which of the 4 EXT BAT pins are GND and Vcc. GND was easy enough to figure out, but that leaves 3 other pins.

I guess I will try to reflash the BIOS while waiting to see if anyone has a high resolution image of this board.
EDIT: I've written off this board as dead. I've tried everything. I am still wondering if the poster of the board with a PS/2 DIN has a working PS/2 mouse on this motherboard.