Your results and due diligence are impressive. I am looking forward to the results with 128 MB FPM RAM to match that of my test system. If possible, one test with 1:1 divider and one with 1:2/3 divider. Were you going to run the DOS tests too?

If you tried to run a fully loaded Windows98SE with only 32 MB of RAM just so that all the RAM is cacheable, I suspect the system would be unbearably slow. We are currently running barebones 98SE installations.

It is too bad that the manufacturers of this Biostar didn't allow for up to 1024 KB of L2 cache and a WB/WT option. I haven't been able to find 1024 KB w/TAG on eBay for a long time now. From my test results posted in "Quick question about 486 mobo's cache," there doesn't seem to be any noticeable speed difference between L2 in WB or WT mode.

feipoa wrote:

Upon looking at the Biostar MB8433UUD settings, the POD83 should have one jumper that is different from the 5x86.

Can I double-check this please? I want to quickly test my POD83 on the Biostar, to see if it can reliably run at 100 MHz. I checked the jumper settings online, and spotted 2 differences between the P24T and P75 settings. However, this webpage stated it was for versions 1 and 2, and the version we are using is 3.x

I have "ringed" the 2 differences

feipoa wrote:

Your results and due diligence are impressive. I am looking forward to the results with 128 MB FPM RAM to match that of my test system. If possible, one test with 1:1 divider and one with 1:2/3 divider. Were you going to run the DOS tests too?

It was a lot of nerve wracking fun. 😀 I'm aware that this huge overclock can't be doing the hardware much good, in the long term. And I've only got one ADW that has got a potentially magic production date. I'm going to let the ADW "cool down" for a bit, and temporarily mess about with an OC'd POD83 for today. I'll also look for some 64MB FPM sticks...

BTW, I was a little surprised at PassMark v4's 2D Graphics mark score of ~52. I was expecting a value a little bit higher than that. Your OC'd ADZ @ 160 score is 54. Even your ADZ @ stock speed beats my score. I wonder, could this be due to the extra 8MB of RAM onboard your Matrox G200? Could it be due to the ADZ architecture, compared to the ADW?

When I installed that older Matrox driver, I noticed that 2 Matrox processes were present in memory. I decided to leave them there, while the tests were running. I think one was called mgactl. Also, I just realised something. I left the Matrox driver's settings on default. I didn't specifically activate anything. I think there's an option called Z-something or other. Sorry, can't remember what they call it. You can see it inside the Matrox display panel property area.

Edit: I tried both of my POD83s OC'd to 100MHz, but the Quake timedemo failed. I tried setting the BIOS timings to "slow", but that didn't solve the problem. I guess that neither of my POD83s are "magic".

Edit 2: In an attempt to get the OC'd POD83 working, I set the BIOS set up area option called "L1 cache mode" to write-through. However, if I run chkcpu.exe at the DOS prompt, it tells me that the CPU is in write-back mode. I guess I need to alter a jumper on the mobo to tell the mobo not to use it in write-back mode. I wonder which jumper that is?

Edit 3: If I run LandMark v2.0 in DOS with my OC'd POD83, I notice that my Matrox G200 gets a score of 13845 for the video. For your OC'd POD83 with a G200, it is 16948. I wonder if that's because of the extra 8MB of onboard memory you have for that VGA card, or are you using any "video TSRs", to get any kind of boost for your video card performance?

Your jumper settings for the POD83 are correct, but remember to put the voltage onto 5 volts and the frequency to 40 Mhz for the OC.

I think the beating of the 2D score was randomness. Try rebooting and running it again. We really cannot compare our true scores until you put 128MB of FPM RAM inside there (you can use 4 sticks if that is all you have) and you set 1:2/3 for the PCI divider. This is a CPU/graphics comparison rather than a world's fastest 486 system comparison. That can come next.

I just received my ADW. Please let me know what your date code and CPUID are.

I don't think a 486 is fast enough to benefit from Z-buffering. Someone correct me if I'm wrong. I also left all default background driver-related apps running.

Were you able to boot into Windows with your POD100 and run the tests? The only tests I can't run with my POD100 are Ziff-Davis Winbench96 Graphics WinMark and Quake Timedemo, however from all the data I've gathered, I have extrapolated those values. Try running your POD100 on WT mode. In WT mode, I can run all tests. Looking at your EDIT2, you've already done that. Try putting your POD jumper settings onto the same settings as your AMD X-5, but at 40 Mhz FSB and 5V. That should force it into WT mode and let you run all tests to agree with my CPU test #Q, in WT mode. The POD doesn't listen to the CLKMUL command, so 2x or 3x jumper on the MB is unimportant.

What is Video TSR? I don't think my extra 8MB of video ram is doing much except at higher resolution settings.

OK, thanks a lot for the info. Please note that I did not increase the mobo's voltage jumper setting for the POD83, because I thought this CPU had its own built in VRM which "converted" the mobo's generated 5v down to 3.something that this CPU operates at. I would imagine that you cannot force the POD to operate at a higher voltage. I think these things are designed so that they can be to put in to mobos that only operate at 5v, and the clever built in VRM that these CPUs have reduces this power down to about 3.something.

Regarding my successful ADW chip, its production date inscription is: A 9630DPE, and the CPUChk.exe DOS utility reports: CPU-ID Signature of 0004F4.

I have not attempted any Windows tests with the POD100, because I was a little disappointed at its failure to complete the "crucial" (because it's sensitive to OC'ing) Quake timedemo test. However, I will set the mobo's jumper settings to exactly match that of the AMD 5x86 P75 X5, and I will also set the mobo's CPU jumpers to 5v (although I wonder if this action is redundant.)

Regarding my phrase "Video TSR", I clumsily meant something along the lines of some kind of "frame buffer booster" utility which might also stay in memory, such as activating LFB or maybe activating some important VESA mode. Actually, forget I mentioned it! 😉

feipoa wrote:

Your jumper settings for the POD83 are correct, but remember to put the voltage onto 5 volts and the frequency to 40 Mhz for the […]
Show full quote

Your jumper settings for the POD83 are correct, but remember to put the voltage onto 5 volts and the frequency to 40 Mhz for the OC.

I think the beating of the 2D score was randomness. Try rebooting and running it again. We really cannot compare our true scores until you put 128MB of FPM RAM inside there (you can use 4 sticks if that is all you have) and you set 1:2/3 for the PCI divider. This is a CPU/graphics comparison rather than a world's fastest 486 system comparison. That can come next.

I just received my ADW. Please let me know what your date code and CPUID are.

I don't think a 486 is fast enough to benefit from Z-buffering. Someone correct me if I'm wrong. I also left all default background driver-related apps running.

Were you able to boot into Windows with your POD100 and run the tests? The only tests I can't run with my POD100 are Ziff-Davis Winbench96 Graphics WinMark and Quake Timedemo, however from all the data I've gathered, I have extrapolated those values. Try running your POD100 on WT mode. In WT mode, I can run all tests. Looking at your EDIT2, you've already done that. Try putting your POD jumper settings onto the same settings as your AMD X-5, but at 40 Mhz FSB and 5V. That should force it into WT mode and let you run all tests to agree with my CPU test #Q, in WT mode. The POD doesn't listen to the CLKMUL command, so 2x or 3x jumper on the MB is unimportant.

What is Video TSR? I don't think my extra 8MB of video ram is doing much except at higher resolution settings.

Isn't there a Pentium optimized version of Quake? Maybe there's a benchmark that goes with it that the POD will run?

@sliderider
Not sure if there is an optimized Quake version, but I do not think the lack of such a version would render the POD100 untestable. As I mentioned, I was able to complete the Quake test w/POD100 in WT mode and in WB mode at 83 Mhz.

@rg100
I don't usually trust voltage regulators to up regulate when I don't know the minimum voltage requirement to regulate, i.e. Vin,min. If Vin,min = 3.45V and your input voltage (from the motherboard) is 3.35V, there could be a problem. Setting the voltage to 5V for the POD reduces uncertainty. Vin,min on the Biostar's voltage regulator is 4.5V, for example. If, for example, the POD83 has the same requirement for their build-in voltage regulator, then you'd definately need to run your motherboard's voltage at 5V.

I've seen voltage regulators on motherboards that say the output is 3.3V and when measured, it is 3.51V. This likely has to due to the mothermourd manufactor's using low tolerance resistors (+-5-10%).

My ADW is: A 9641DPA. Overclocking to 200 Mhz failed for me today. Mine has CPUID 04E4, however one of my ADZ's have 04F4. Can you check your other ADZ's and see if any of them are 04F4. And confirm that these also do not overclock to 200 Mhz?

I added thermal paste to a Socket 370 cooler heatsink and ran the fan with the ADW. Voltage was set at 5V and I set all BIOS settings to conservative. The Graphics cards I tried were Matrox PCI, Diamond VLB, and Trident ISA. I also tried 3.7 V. No luck. It might be a factor of the M919 rather than the CPU, or it may be that you have a magic cpu.

Due to my failure to overclock the AMD to 200 Mhz, I am now totally reliant on your results w/128MB FPM RAM and 1:2/3 setting to finish off my comparison sheet (I'm actually still waiting on an AMD DX4-120 WB).

POD83 is actually supposed to have it's CPU voltage set to 5v, it's an overdrive chip 😉

Success! @feipoa, your advice about setting the mobo's voltage to 5v was essential. Thanks very much! I now have the POD83 overclocked to 100 MHz, and it passes the "crucial" Quake test, in write-back mode, with all of the BIOS timings set to their most aggressive settings! (Actually, DRAM read is set to 1, and not 0, but everything else is "maxed out".) Full screen Quake timedemo gives me a score of 20.7 FPS. Time permitting, I may run the whole batch of tests again (DOS 6.x, Win 3.x, Win95, Win98SE) using this OC'd POD-100.

I understand that your ADW failed its 200 MHz overclock. Please can you elaborate? When you say failed, did it a) not POST, or b) it POSTed OK, but your OS would not boot, or c) your OS booted OK, but the Quake timedemo failed.

I will check my ADZs when I can find them (I put them in various boxes with the paste still covering the CPU so I can't identify them without scrapping away paste on lots of CPUs), and at the same time I will try and locate 128MB of FPM, (and I'll also look for my M919 again! - I jam about 15 mobos next to each other inside typical office archive boxes, it's the only way to save space), but I can definitely confirm two things right away:

a) none of my ADZs OC'd to 200 MHz, on any mobo.
b) I couldn't get my M919 mobo to OC to 200MHz using my "magic" ADW.

I am inclined to think that if your ADW 200 MHz OC failed to produce a BIOS POST screen, then it is due to the M919. I am also inclined to think that if you tried your ADW in your Biostar, you will have much more success. I hope you can try this experiment! 😀

Edit: Please note that I tried the POD83 using two different types of mobo jumper settings. Both produce identical (and successful) results for the Quake timedemo test. The first jumper configuration appears in the illustration several posts above, for the POD CPU. The second jumper configuration that I tried for this POD CPU was for the AMD X5 jumper configuration, and this configuration also appears in that illustration above.

Ah, this has just made me realise something. For both of these jumper configurations, I ran CPUChk.exe, and it reported that the POD was in write-back mode. I'll have to mess about with this again, to see if I can force the POD to use write-thru mode. This may come in handy, if the write-back mode causes the OC'd POD to behave oddly with some Windows tests...

@rg100

Before you do that, can you run the gongshow of DOS/WIN98SE tests with 128MB FPM RAM, 1:2/3 PCI, and X5-200 so I can finish up this list? Also, I don't think the POD83 will survive continual use at 100 Mhz. I'm glad you got it running at 5V -- it looks my theory on the regulator was correct. Oddly enough, the actual pentium chip runs at 3.x volts. Intel should have put in a more versitile regulator.

Your Quake score seems low for L1 on WB mode. My POD83 at 83Mhz gets 20.8 fps in Quake. Check both CTCM and chkcpu to confirm it is in WB mode. In WT mode, I get 19.7 fps. Actually, I don't need to ramble off these numbers, you have the updated list.

When I say my ADW failed at 200Mhz, I mean the screen never changed from black to anything but black. I don't think the M919 runs well at 50 Mhz FSB. I could only sometimes get the X5-150 running on 3X.

The fact that you couldn't get your M919 to OC to 200 Mhz with the magic ADW is discouraging. From what you've told me, I'm left to assume that an AMD X5-133ADW w/CPUID=04F4 and a Biostar MB-8433UUD-A v3.x is the way to go for a 486 at 200 Mhz. Unfortunately, I can't rip open my Biostar MB again, at least not for some time. I really need to get a backup for testing. I'd settle for a v2.0 if anyone has one.

Don't always trust CPUCHK.exe, it sometimes disagrees with CTCM for write-back/through. Always use both apps. to check the status of the L1 cache.

EDIT1: It would also be worthwhile if you checked your ADZ's to see if any of them are 04F4. If you tried all of your ADZ's and none of them are 04F4, then there may still be hope for the ADZ afterall.

EDIT2: Whenever my precision surface mount resistors come in to modify the MB-8433UUD's voltage regulator to 3.7 V (for the Cyrix 5x86-133), I can try the ADW/ADZ on the Biostar at 200 Mhz. Unfortunately, these particular SMD's were non-stock for the precision I wanted. The M919 can get away with low tolerance for its regulator, but the Biostar's is pretty precise.

I will look through more of my boxes of retro stuff to look for 128MB of FPM! You're right, my Quake score does seem low. I wonder if the CPU is in write-back mode, or not. I am trying to determine if the POD is in WB mode. Unfortunately, I am not able to interpret these DOS utilities, to figure out if the POD is in WB mode, or not. Please can you look at the screenshots below. BTW, what version of CTCM are you using? If it differs from the two I have tried below, please can you upload a copy, if it is freeware. Thanks a lot.

CTCM version 1.6r3. Oh dear, it locks up at the MMOVI instruction test.

CTCM version 1.7a. Oh dear, it locks up at the MMOVI instruction test.

Cachechk version 4

ChkCPU

BIOS settings. I wonder if they are set to "maximum performance"?

Any reason why your IBC DEV SEL is set to Slow? IIRC, this is for ISA refresh. POD100 should still work on Medium. Also, any reason why your Slow Refresh (RAM) is not Enabled? This should reduce the frequency of the RAM refresh waveform (for dynamic RAM). From what I've seen, it doesn't seem to improve performance, but from an engineering standpoint, it is less waistful on resources. The RAM needs to support slow refresh; both my FPM and EDO do.

You want to use the pre-1.7 CTCM, aka CTCM to see if L1 is WB or WT. I have attached it for you.

Thanks a lot! I downloaded that older version of CTCM and ran it. It states that the POD is running in write-back mode. Also, I have been looking for sticks of RAM. I found another "random stick" of EDO. I tried it, and I was able to get some of these DOS benchmark utils running with the BIOS DRAM option set to "0" (not "1"). Quake worked, but only once. I got 21.9 FPS, full screen. Also, I set those two BIOS options you mentioned to "Enabled" and "Medium".

Your 3DBench, Quake, and Doom POD-100 scores easily beat mine. Maybe that PC-Chips mobo of yours is a really good board? And maybe you've got really good RAM? 😀

CPUs O and P from the chart I sent you, if you look at the superscript, you'll see that these CPUs were tested in the Biostar. CPU Q in WT mode was from the M919.

OK, sorry about that! Some of those scores are really good. While I rummage about for more sticks of RAM (all of which are EDO at the moment), I have run all of the DOS tests on DOS 6.22, with the overclocked POD83. BTW, my config & auto start up files are empty.

Current BIOS settings. I think they are maxed out.

CTCM. Shows that the oc'd POD83 is running in write-back mode.

Symantec Sysinfo v8.1

PC-Config v9.33 (including % of Pentium 100)

PC-Config v9.33 (graphics)

CpuIndex v2.3

PiDOS [25k digits]

Landmark v2.0

Chaikin Benchmark DOS v1.0

Bytemark v2, 32-bit DOS

Roy Longbottom Dhrystone v1.1 - DHRY1OD (VAX MIPS Rating)

Roy Longbottom Linpack - LINPCOD (MFLOPS)

Roy Longbottom Whetstone - WHETCOD, MWIPS (MFLOPS), and N1 to N8.

Speedsys v4.78

Cachechk v4.0

3Dbench v1.0. Hmmm, could be better, perhaps?

Doom v1.9s timedemo3 (realtics) - full screen. Sorry about the photo. I took 8, and they were all unreadable. My unusually poor score is 1779.

Pcpbench v1.04 - default mode 100 - 640x400. I wonder why a) the image looks "letter boxed", and the score is a bit low?

Quake v1.06 timedemo1 (fps) - full screen. That was lucky - with the BIOS timings "maxed out", it didn't crash. Usually, I have to set the BIOS DRAM read option to "1", to prevent Quake timedemo from crashing, if I overclock the POD83 in write-back mode.

feipoa wrote:

The fact that you couldn't get your M919 to OC to 200 Mhz with the magic ADW is discouraging. From what you've told me, I'm left to assume that an AMD X5-133ADW w/CPUID=04F4 and a Biostar MB-8433UUD-A v3.x is the way to go for a 486 at 200 Mhz. Unfortunately, I can't rip open my Biostar MB again, at least not for some time. I really need to get a backup for testing. I'd settle for a v2.0 if anyone has one.

I have managed to get an ADW OC'd to 200MHz on another mobo. It's an SYL8884PCI-EIO. It talked it about here, about 2/3'rds the way down on that page. I understand that your Biostar 3.0 mobo is being used as part of a production server. Would you consider upgrading this server hardware, to say any Pentium mobo? That way, you could free up the Biostar for some valuable and exciting retro testing! 😀

I have looked through a lot of my retro boxes, and only found EDO RAM. However, I know I have some 32MB FPM sticks somewhere. Begin Edit: I meant to say 32MB, not 64MB. I think I've got some 64MB FPM sticks, but I'm not sure where they are. End Edit. I've only got 4 of them, and so finding *all 4 of them* could be difficult. Tomorrow, I might do some 128MB EDO testing. What I will do is run a "big benchmark", like WinBench 99 using an ADW at stock speed, with 32MB of EDO RAM. This takes about 30 minutes, as there are over 100 tests involved. Then, I will add in the rest of the RAM, to make it 128MB, and rerun this test. I believe that the results will show a significant performance drop, because of the large amount of uncached RAM being used by Windows. However, I will be happy if I am proven wrong about this! 😀

I had a mess about with WFWG 3.11, using the OC'd POD. WinTune 2.0's disk testing didn't like SmartDrive enabled, so I put a REM comment next to this line inside the autoexec.bat file. After I did that, WinTune 2.0 completed all of its tests.

WinTune 2.0, RAM

WinTune 2.0, video

I have run two tests. One with 32MB of EDO, the other with 128MB of EDO. I'm using the ADW CPU. I appreciate that I'm not using FPM yet, because I haven't found four 32MB sticks (or two 64MB sticks), but I think I've discovered something relevant. I ran WinBench 99 version 2.0, and chose 118 tests to run. I omitted the disk tests, because they seem to cause problems with my compact flash device. (These omitted disk tests appear in the results table below as "Error".)

The column below on the left contains the test results with 32MB in the mobo. The column on the right contains the test results with 128MB in the mobo. I'm not sure why they are labelled "133" and "133 B" - please ignore that.

A red coloured test result indicates the worst result, while a blue coloured test result indicates the best result. Looking at these left and right columns, it seems that the 32MB tests column (on the left) has performed the best.

128 MB of EDO RAM will be fine for the comparison. I haven't noticed a difference between 60 ns EDO and 60 ns FPM. Some of your POD100 scores are worse than mine on the Biostar. The only other setting in my BIOS I have is the Bus Park Option - Enabled. I don't recall what this does. Your Quake Time Demo seems low. In WT mode I get 23.6 at 100 Mhz. In WB mode, it should be around 24.9 fps. Your PcpBench and 3Dbench scores also seem low. Try running at 1:2/3 and see what happens.

The Windows tests of interest that should show dropped performance with using more RAM than is cacheable are CPUMark, PassMark, WinTune98-Memory, SuperPi, and maybe FPUmark. I think DOS tests will yield the same result. It looks like you only ran Winbench99 graphics mark.

EDIT: I took this from the BIOS Survival Guide by J. P. Rodrigue
"PCI Bus Parking. Sort of bus mastering; a device parking on the PCI Bus has full control of the bus for a short time. Improves performance when that device is being used, but excludes others. Try with NICs and Hard Disk Controllers."

You're right, some of those POD scores (Quake, PcpBench, 3DBench, and also Doom) are rather poor on my Biostar. It could be because of the

* difference in the board revision (3.0 to 3.1)
* difference in the BIOS firmware
* difference in the quality of the RAM
* difference in the VGA card, somehow?

Who knows? It's interesting, but nevertheless a little bit disappointing. Oh well! 😀 Regarding my WinBench 99 version 2.0 tests, I know that I ran 118 tests, but I don't know why the comparison table doesn't include all of them. Also, I've given the following a couple of days of careful thought - please would you consider either

* testing your ADW on your Biostar, or
* removing the X5-200 column from your data table, or
* adding my 32MB based test results

Now that I realise the performance will automatically drop if I add more RAM, I am reluctant to run another round of risky overclocked tests. The main reason is that no one would want to set up a system like this. If you remove the X5-200 column from your data table, you could put it back in if you decide to test your Biostar mobo at a later date. If you include my 32MB results, you could add a superscript at the bottom of the table to say that it used 32MB of RAM. I think that would make sense.

Whatever decision you take, I'm very grateful for the info and advice given about doing these tests, and also I'm sorry that I've decided against running the additional "uncached and artifically slow" and also somewhat risky 128MB tests. It's possible that I've only got one "magic" ADW, and I need it to still be alive in a couple of years time, when I begin some YouTube videos about old PC hardware. Thanks a lot.

a little OT, but what is the Quake1 timedemo benchmark command line?

keropi wrote:

a little OT, but what is the Quake1 timedemo benchmark command line?

Start a new game, and then at the console, you type timedemo demo1
To get to the console, press CTRL ~
After you type timedemo demo1, quickly press CTRL ~ again, to dismiss the console.

The tests done recently have been in full screen mode. To get to full screen mode, when you are in a new game, press the + key a couple of times.