Lots of questions! 😀 I'll try and answer them as best I can -

>> If you could replicate these conditions, that would be fantastic.
I will do this, but the problem I have ATM is time. Please can you give me about a week to get this done. Thanks a lot.

>> For the Windows98SE tests, can you use 1024x768x16bit, DirectX 6.1a, and with all the latest security updates?
I can do this. In fact, what I can do is install W98SE, and then apply the unofficial service patch. Will that apply the latest security updates?

>> Is the X5-200 long-term stable in Windows 98SE?
I do not know the answer to this. TBH, I am a bit doubtful!

>> I have used DOS 7.10 included with Win95 and 98's. I doubt this will have any impact on your results using 6.22. What do you think?
Good point. Perhaps I should run all of my tests using DOS 7.10? Please can you refresh my memory about something. When using W98SE, what "tweak" do I need to do, in order for W98SE to automatically boot in to DOS only? ATM I use TweakUI to display a list of boot up options, but I'd rather use a simpler approach, and simply edit some Windows .INI file, to tell Windows not to loads its UI.

>> Did you confirm that the ADZ does not overclock at all to 200 Mhz and that only the ADW does? What led you to this determination?
I have never been able to OC an ADZ to 200 MHz. I have only been able to OC an ADW CPU to 200 MHz. I tried everything: loosened up all BIOS timings. Nothing worked. I think I tried 2 ADZ CPUs. Both failed the 200 MHZ OC test. This testing isn't very "complete". I wonder if there is someone out there who has successfully OC'd an ADZ CPU. Actually, come to think about it, I find it a bit surprising that no one else has managed to get any AMD 5x86 133 MHz CPU OC'd to 200 MHz.

>> Why are you not using LFB mode in pcpbench.
Good question. I should be. Some things I get a mental block over. I keep getting a mental block over the issue of LFB. I think you can activate it by using SciTech's Display Doctor, more specifically one of its utilities called Univbe. It's either that, or something called FastVid. Sorry I can't remember this. I'm a bit stupid, TBH. I'll mess about with this a bit later, and get LFB activated, and then rerun those tests again.

>> Does ctcm, and chkcpu16 show your POD83 as L1 Write-thru even though you put Write-back for the BIOS setting?
ATM, I can't remember. I think there's something "funny" about the POD83. I think it has difficulty with some mobos. One of the things that is odd is how L1 works. I think if it isn't 100% compatible with the mobo, L1 goes in to WT mode, even if you set it to WB. Maybe double-check your mobo jumper settings?

>> I thought the POD83 was WB.
It is, but please see above.

>> did you set PiDOS to 25k digits?
I ran it using its default settings. I guess this isn't 25k digits. I can do this test again. In fact, I can rerun everything again, but using DOS 7.10.

>> I tested today my am5x86@180 on ga-486am/s mobo with fsb:pci divider set to 1:1. It booted to dos correctly from flopy drive. However, I haven't tried with hard drive. Did you try to boot from floppy drive?
That's very interesting. Were you using a PCI VGA? I am beginning to suspect that a high FSB value (eg 50 or 60) with a PCI divider value of "1:1" causes these mobos to fail, not because of the video, but because of something to do with how they access a disk at boot up time. I haven't tried booting from a floppy drive yet. That's another job I will do! I discovered something odd yesterday. If I set the Biostar mobo to "completely normal slow and safe" settings, and set the PCI divider to "1:1", Windows 95 freezes at the desktop if I use the integrated IDE for a HDD, ODD, and a FDD. That shouldn't happen. If I use an EIDE ISA IO controller, this problem is resolved.

>> I did not have to loosen up any cache or memory timings + The POD100 was stable enough in Windows to run all the Windows98SE-based tests I mentioned
That is fantastic! I'm unsure whether I can match that. I think that one of your PODs is "magic" - perhaps a bit like one of my ADWs is "magic".

>> There's also a few processors that I'm waiting to come in the mail to add to the list, like the Cyrix 5x86-80, AMD DX4-120, ...
Try OC'ing the AMD DX4-120 to 50 FSB (150 MHz CPU clock speed).

Edit:

>> One more thing, did you set PiDOS to 25k digits?
I've just double-checked this. The default is 25k digits, and that is what I have used.

I decided to run some of the aforementioned "big list of windows tests" compiled by Vogons member feipoa, on Windows 95. Here are the results:

17) SuperPi v1.1 (32k digits test in middle)

18 ) Justin Benchmark Win v1.0 - all items, averaged over 3 trials - I ran this 3 times in a row.

19) Ziff-Davis Winbench 96 v1.0 - CPUmark32 and Graphics Winmark (make sure you right click your taskbar, properities, and uncheck "always on top" for the graphics winmark to work). Also, this test took a long time. During this time, I wiggled the mouse to see if it had crashed. It hadn't. I hope I didn't reduce the score by doing this. I doubt it.

CPU:

Graphics:

Ziff-Davis Winbench99 v1.1
20) CPUmark99 -v1.0, this is the standalone addition, not bundled with Winbench99

23) Sandra99 - CPU ALU/FPU,

Multi-Media ALU/FPU,

and Memory ALU/FPU.

I also ran WinTune97

Please note that PassMark version 4 would not run in Windows 95. I decided not to run WinTune 98, but to run WinTune 97 instead. I decided not to run 3DMark 99, because at the time of testing, I did not have DirectX 6 installed. Also, I decided not to run WinBench 99, but just run WinBench 96. However, I did run CPUmark99. Damn, I just realised that my Windows 95 desktop is only 640x480x8. Also, I noticed that when I used this mobo @ 200 MHz, the CD-ROM and FDD could not be seen by Windows 95. Begin Edit: Pilot error: forget the bit about the CD-ROM and FDD. It works fine.End Edit.

When I have time, I will do a fresh Windows 98 SE install on this OC'd mobo, and run all of the "big list of windows tests" compiled by Vogons member feipoa. I'll make sure the desktop is set to 1024x768x16.

I installed SciTech Display Doctor "Univbe". I reran PcpBench, and it now runs in "LFB mode". However, my 640x400 "mode 100" score has worsened. It's now only 9.0. Previously it was 11.something, when it was not running in LFB mode. Strange.

I remember running a utility called s3vbe318.zip (s3vbe20.exe), when I was testing an S3 (Trio64V+) card, and when I switched to LFB mode, PcpBench ran faster. So I dug out my Trio64V+ PCI VGA, and tried it. With LFB not activated, I get 11.4. With LFB activated, I get 9.0. 😕

I am not sure about the unofficial service pack for Win98SE and about what it contains. I installed IE6 and installed all the updates (security, drivers, IE updates, DX6.1a, etc) via the Windows Update website using a network connection.

It would be great if you could run the tests under the same conditions to reduce uncertainty.

What version is your ADZ and ADW? You can run CHKCPU16 to find out. Maybe the ADZ will OC using 4.5 V instead of 5V? Grease up a Socket370 cooler and slap it on :) I'm not there yet, but this is on the list.

The funny thing about the POD83 cache was that it was split, 16KB for data, 16KB for instructions. The BIOS would need to be aware of this. My PCChips M919 manual doesn't have a specific jumper setting for the P24T (POD83). I am instead using the 5x86 jumper settings. Upon looking at the Biostar MB8433UUD settings, the POD83 should have one jumper that is different from the 5x86. I think the M919 is treating the L1 cache as one 32KB or one 16KB in WT mode. I should double check this in cachechk. I might need to test the POD's in my Biostar.

I think I just talked myself into re-testing the POD83 at 83 Mhz and 100 Mhz in the Biostar MB-8433UUD due to caching issues which may degrade its real intended performance.

The reason that your PiDOS was defaulted to 25K is because I probably zipped/uploaded the version I changed the bat file for 25K.

The AMD DX4-120's only have 8KB cache so it is not so interesting to OC them compared to the Intel DX4 and AMD X5's which have 16KB WB cache.

The results of your initial X5-200 Windows95 tests are very encouraging, however the conditions are still not consistant, i.e. win95 vs. win98se. I am looking foward to the Win98SE tests. I am aware this is time consuming, so time permitting.

Your PCP Bench results are more realistic now. The X5-200 has a slow FPU, it is about equivalent to a Cyrix 5x86-120 when OC'd to 200 Mhz. I get a PCPBench score of 8.8 using the Cyrix 5x86-120 (MB-8433UUD) and 9.7 w/VLB card (M919).

I also do not recall how to default boot into Win98SE. I've been hitting F8, 5 at boot, however I think the change may be this,

EDIT c:\msdos.sys
BootWin=0

feipoa wrote:

I am not sure about the unofficial service pack for Win98SE and about what it contains. I installed IE6 and installed all the updates (security, drivers, IE updates, DX6.1a, etc) via the Windows Update website using a network connection.

+

feipoa wrote:

It would be great if you could run the tests under the same conditions to reduce uncertainty.

Very unfortunately, I can't do that, I'm sorry. For starters I don't have a NIC, and I also don't have a clue how to connect to the net using W98. It's been 10+ years since my last dial up connection on W98. Also, do MS allow this any more? I have a feeling that they have dropped support (including updates) for W98, quite some time ago.

So, I think that this update procedure is not possible, even if I had to ability to go online. Is there some other way we can establish a similar set up? I am prepared to do a fresh installation. Oh BTW, I'm going to be using a 2GB 133x speed compact flash drive. If you decide to do a fresh install, we could agree to just install W98SE, and nothing extra. That would keep things simple.

Begin Edit: I can also use a regular HDD, such as a Western Digital 7200RPM 80GB model. However, if possible I would rather use a CF, because then I don't have to use a ribbon cable, and I think that helps with overclocking. I'm sure I've had problems with 200 MHz OC'ing, when using a ribbon cable + HDD, as opposed to a CF inserted directly on to the IDE pins. End Edit.

feipoa wrote:

What version is your ADZ and ADW? You can run CHKCPU16 to find out. Maybe the ADZ will OC using 4.5 V instead of 5V?

I'll run that utility. I think I've got it somewhere. I remember running something called chkcpu.exe. I am guessing it's the same as chkcpu16. On the surface of the ADW CPU, it says 9630DPE. I might try my other ADWs, which have different production dates. Also, I'll double check my ADZs. ATM, they are not immediately available to check right now.

feipoa wrote:

Your PCP Bench results are more realistic now. The X5-200 has a slow FPU, it is about equivalent to a Cyrix 5x86-120 when OC'd to 200 Mhz. I get a PCPBench score of 8.8 using the Cyrix 5x86-120 (MB-8433UUD) and 9.7 w/VLB card (M919).

A PcpBench score of about 8-9 is fairly common, for any "fast" 486 mobo. I think what is holding back my score is the VGA. I believe this makes a difference. Going back to your earlier remark of "It would be great if you could run the tests under the same conditions to reduce uncertainty", it would be great if we were using the same VGA.

Please can you list your PCI VGAs? Do you have an S3 Virge 325, or an S3 Trio64v+? If necessary I could buy the cheapest PCI VGA you are using, because I believe this "common ground" is important for graphics testing. Please note that it's important to double-check that a PCI VGA works @ 50 FSB speed, with the BIOS PCI divider option set to "1:1". Only some of them do!

feipoa wrote:

I also do not recall how to default boot into Win98SE. I've been hitting F8, 5 at boot, however I think the change may be this,

EDIT c:\msdos.sys
BootWin=0

Thanks a lot. I'll give that a go. Once we've established a similar set up, I can get W98SE installed, and I'll probably run through every test including all of the DOS ones again, using DOS 7.10.

The systems do not have to be the same, but the closer a match they are the better. I do not have S3 Virge 325 or S3 Trio64v+, although I have quite a few 1997-2000 era PCI VGA cards. The only one in my pile that I can get working on a 486 is the Matrox G200 w/16MB RAM.

Is S3 Virge 325 capable of 50 Mhz? Does it support Direct3D, DirectDraw, and OpenGL? What about 1024x768, or 1280x1024? I'm not sure I'd have much use for this card outside of this one test.

I would just start with what you have at this point, as I am also going to attempt the X5-200 setup. In my opinion, the G200 is a great PCI card, one worth adding to any retro collection. Since I am already 16 cpus-tested into this project, it would be a lot more time intensive for me to swap VGA cards and retest everything with a new VGA card. For my X5-200, if I cannot get my PCI card working, I will put in an ISA card and negate graphics performance. Likewise, if I cannot get my PCI SCSI card working, I'll at least run all the DOS tests from floppies. And again, I might not get anything working. I still have 6 cpus to test before I get to the X5-200 attempt.

You can connect to the internet the same way in Win98SE as one does in in WinXP if you have an NIC and a router. Windows update still works with Win98SE and IE6sp1, although there are no NEW updates to download. In fact, Windows NT4 still works with Windows Update, but you are stuck with the final release of the security updates. It is probable that this unofficial service pack contains all such Win98SE updates.

Since most of these tests are CPU-RAM-Video only tests, it probably won't matter much which fixed disk you use. Theoretically, a 133x CF card will not outperform what is in my test bed (SCSI Ultra2 controller w/Ultra320 hard disk). I beleive a 133x can theoretically accomplish 20 MB/s or so, whereas Ultra2-LVD is 80 MB/s, and requires less attention from the CPU.

Re: S3 Virge 325. It is capable of 50 MHz, and it does support DirectDraw and OpenGL, but I don't think Direct3D works. 1024x and 1280x resolutions work OK. But please don't get one, because I'll get a Matrox G200 instead. 😀 It sounds like a good card! (BTW, I didn't want you to redo all of your CPUs tests, just the X5-200 test.)

Can you do me a favour please? Can you double-check that the Matrox G200 PCI card works OK at 50 MHz FSB, with the BIOS PCI divider option set to "1:1"? I really need to know if it gets to the Win98SE desktop, having successfully loaded its graphics driver. Thanks a lot for checking. As soon as you can confirm this please, I will buy one. There's a seller who has a whole pile of them in the UK. They're not exactly cheap, but these tests fascinate me, so for me it's worth it.

I had no idea you could still update Win98SE. Thanks for the heads up! However, if it's OK with you, I'll just install the unofficial service pack, and also IE6sp1. BTW, I've got a good idea. Once a get a Matrox G200, I'll set up the system to run at "stock speed", that is at 133 MHz. Then I can run a selection of tests. (Ideas please?) If they are extremely close to your ADZ "stock speed" results, then we will know that we have two similar base machines in which to compare further test results.

Matrox G200 and Adaptec 2940U2W PCI cards boot up into Windows 98SE with a 50 Mhz FSB fine using an AMD X5-150/3x and an M919 motherboard. The M919 does not have a PCI frequency divider option, so it either does it automatically or not at all. There is an ISA clock divisor setting though, and even at 1/4 FSB, the ISA cards had issues loading their drivers in Windows and had to be removed.

I could not get the ADZ to even show a screen at 200 Mhz. My ADW hasn't come in yet.

You should be able to get a G200 for less than $30 CND. I bought a backup recently for $20. You can get the 8MB upgrade module for about $10 as well, depending on your luck.

I have a hunch that the M919 mobo has an automatic divider. Is your Biostar mobo currently "in production" as a server? In other words, is it not available for testing purposes right now? If not, I'll go ahead and get the Matrox, but a really good test for this G200 video card would be to test it in the Biostar mobo with the BIOS divider option set to "1:1", and see if Win98SE can successfully load its driver when the desktop appears. That way we'll know for sure! 😀

I have other PCI VGA cards that boot in to Win98SE OK, but when you install their drivers, and set the resolution to 1024x768x16bit, that's when things become unstable...

The Biostar MB-8433UUD motherboard is occupied at the moment, though I can probably free it up sometime this week for a few hours if you wanted to wait on that G200. However, for me to boot into Windows 98SE, the Adaptec SCSI card will also need to be stable at 50 Mhz. If the system is not bootable, it might not be clear which, the Matrox or the Adaptec, has caused the 50 Mhz incompatibility. In short, you still might not get a definitive answer.

If the M919 is indeed auto dividing, 50 Mhz x 2/3 = 33 Mhz. This configuration will definately be stable and fast. To be fair, upping the FSB will up the benchmark results, which will sorta enable cheating in as much as raw CPU/graphics comparison is concerned.

I've found a Matrox G200 PCI VGA in one of my junk boxes! That's the really good news. The slightly odd news is that I did a quick test with it using Windows 95, by DL'ing and installing the w9x_623.exe driver from Matrox's legacy download section for old OSes such as W95, and after it installs, the screen says "General Protection fault".

Tomorrow I'll try 2 things. 1) Reinstall W95 from scratch, then reinstall the G200 card. If that fails, I'll reinstall W98SE from scratch, then reinstall the G200 card using the later w9x_683.exe package from the Matrox website. If that fails, then there could be a compatibility issue with the Biostar mobo and this G200. I'm really hoping that it will work OK, particularly with W98SE, as that is going to be the OS to redo all of the tests discussed so far.

Edit: Instead of installing the G200 w9x_623.exe driver software, I selected the G200 hardware from the W95 Add Hardware wizard, and installed the software from the W95 CD-ROM. That doesn't fix the problem, but does prevent W95 from displaying the "General protection fault" message. Or was it "Windows protection fault", I can't remember. Just before the desktop is displayed, the screen now says "Can't find display adapater". However, the G200 is listed in the Display properties, and its resources appear to be OK.

I'll reinstall W95 and also W98 tomorrow. I'm sure that will fix this problem...

Edit 2: Yes it does say "Windows protection error". I just saw it again. I tried to manually install the Matrox w9x_623.exe driver, by clicking on the "Change" adapter button found inside the Display properties. I wonder if it's something to do with the Matrox PowerDesk software, that gets installed alongside the display driver?

The Windows protection error may be due to poor pnp support on these older motherboards, some real/protected mode driver conflicts, or maybe a directdraw driver from Matrox sending out Pentium instructions. I really have no idea.

I have already come across this and have determined that the latest Matrox driver for Windows 98SE to work correctly is version 4.33C. I am not sure what the latest version for Win95 is.

The Matrox link to version 4.33c can be found here,
ftp://ftp.matrox.com/pub/mga/archive/win_9x/1 … 99/w9x_433c.exe

I tried several things in attempt to get a later driver version working with the G200 and Win98SE -- installing the driver only, no power desk; using the POD83; relaxing BIOS timings. It is curious that the latest NT4 driver 5.06 works with the G200 on the same motherboard in Windows NT 4.0.

Great, thanks a lot for that FTP link. I'll try it today. BTW, navigating around on that FTP site, I can find other things, such as older BIOSes for the G200. Have you upgraded your card's BIOS? It would be interesting to see if there are any speed differences between different versions. For instance comparing an early "simpler" release, to the final release.

Edit: It works! Version 4.33c works on Windows 95. Thanks a lot! I'll be able to start the full W98SE tests within the next 48 hours...

I didn't bother comparing the BIOS revisions. Both of my G200's are updated. I've been using version 3.3 B30 for many years already.

I just did a quick test with the Matrox G200 in Win95. I also installed DirectX 6.1. 3DMark 99 "lite" (not Max) works fine. It completes its tests and reports a score of 18. That's at a safe stock speed of 133 MHz. I'm nearly finished with my Win95 testing, and so I will begin a fresh Win98SE install soon. Then the testing fun will begin...

@feipoa

Which cx5x86 enhancements are most important? I have started doing 486 comparison and I would like to attach Cx5x86 100MHz results with and without enhancements enabled. Currently for version with enhancements enabled I use those settings:
BTB, FP_F, LSS.

I realized that not all settings are stable for this processor and also not all improve performance. Is it worth it to search another combination?

Regards! (;

For the Cyrix 5x86-100, Stepping 1 Revision 3,

*BTB_EN = 1 (for stability, requires that RSTK_EN = 0, LOOP_EN = 0, BWRT = 0 -- branch prediction).

*FP_FAST = 1 (Incease FPU performance by 15-20%)

LSSER = 0 (allows for reordering to/from load/store unit as opposed to serialising everything)

MEM_BYP = 1 (enables memory read bypassing so that data can be read from the write buffers prior to being written to external memory)

DTE_EN = 1 (enables the directory table entry cache)

LINBRST = 1 (enables a linear address sequence while performing burst cycles)

USE_WBAK = 1 (Write-back pins, in combination with CD = 0 and NW = 1 -- for enabling Write-back L1 cache)

WT1 = 1 (sometimes required for stability)

I would say that these are the most important for performance. Hopefully, your mobo already enables MEM_BYP, DTE_EN, USE_WBAK, LINBRST, WT1. Items marked with * should yield the most noticable results. Note that my recomendations are different for Stepping 0, Revision 5 cpus. Good luck!

So I am nearly finished with the 486 comparison sheet, I am just waiting on an AMD X5-133ADW, AMD DX4-120 (8KB WB), and AMD DX2-66 to come in. There is a lot of data.

I was wondering if anybody had any ideas on the best way to represent the data, aside from or in addition to, another chart? The chart currently has 22 cpus and 82 tests.

I was thinking about running the gongshow of tests on a socket 7 Pentium 100 using the same hardware from the 486 tests (RAM, graphics, hadddrive, scsi controller, etc) and normalizing all the 486 cpu data to the Pentium 100, then multiplying by 100 to get a relative percent of Pentium 100 performance. Then take the average of all ALU-based tests and FPU-based tests per CPU to get just two overall performance values per cpu (ALU and FPU). Alternately, I can average all ALU and FPU performance tests into one number, whereby the relative FPU tests are weighted to the number of ALU tests.

Again, the purpose of the cummulative tests is to show each CPUs performance on the same system with identical settings (cache, ram, hardware, Windows install, etc). I've seen benchmark comparisons which might amount to a dozen cpus, but on different systems with different cache/ram settings. To my knowledge, this is the most comprehensive socket3 benchmark comparison list compiled on the same system.

@feipoa

1) Thank you, that you've shared your knowledge! (: I was lucky by choosing most important settings. I will check if motherboard enables some settings automagically. (;

2a) I think that presenting data in form of graphs (one for each test) would be a great suplement to your classic table. (;

2b) I like the idea of normalizing all 486 cpu data to a Pentium 100. I prefer version with two separate numbers (one for ALU tests and the second one for FPU tests) rather than one overral score. However I'd most like to see both! ^^
One averaged number facilitates the process of selecting winner but hides simultaneously very important informations. I think that most users would be happy to know which CPU is better when it comes to work with floating point numbers and which is better when it comes to work with integer numbers. Finally it would be nice to know which CPU is generally the best.

Regards! (:

I have finished my Windows 98 SE tests. The system was set up like this: 32MB of EDO RAM, 256Kb of cache, PCI Matrox G200 VGA using an old Win98 driver, FSB @ 50 MHz, AMD X5 CPU running at 200 MHz, PCI divider set to "1:1". Screen resolution was 1024x768x16-bit. I found the system to be slightly less impressive (speed wise) to the same hardware running Windows 95. I prefer Windows 95, in this kind of situation. Also, it's a pity I haven't got 512Kb of cache. That way, I could increase the RAM to 64MB, and have a more responsive Win98SE system.

SuperPi v1.1. Also, Sandra shows the CPU speed.

Justin benchmark WIN v1.0 (3x)

Ziff-Davis Winbench96, CPUMark32 v1.0

Ziff-Davis Winbench96, Graphics WinMark v1.0

Ziff-Davis Winbench99, CPUMark99 Stand-alone v1.0

Ziff-Davis Winbench99, FPU WinMark99 v1.1

WinTune98 (3x)

WinTune98 (3x)

Sandra99, CPU

Sandra99, Multimedia

Sandra99, memory

PassMark v4.0 (30 day evaluation copy), 2D graphics.

PassMark v4.0, memory.

PassMark v4.0, math.

3DMark 99 (lite, not max version)

Regarding the 3DMark 99 result above - this software displayed a warning about "disk swapping", because I only had 32MB of RAM, and not the minimum ideal amount of 64MB. Also, during this test, I pressed the PRINT SCREEN key, so I probably slowed the test down a tiny bit. Actually, I forgot to paste this image in to a file, so that was a bit stupid of me! 😦