Guys are you serious? Rare? Cyrix 133mhz was one of the el-cheapo 'pentium imitations' back at the time and a lot of my friends used them because of the price. Most of them (CPUs, not friends 😜 ) ended in garbage - which must also be a reason that there are very few of them around nowadays.
As for the reason they ended in garbage - I'll leave it to you. 😜
you are probably thinking of the 6x86 which certainly was commonplace. but the 5x86 133 is a different story.
and we're really comparing 486 level chips here since this is socket 3. but I hear you on how the Chris hardware was very poor relative to a Pentium for 3D gaming. the Cyrix FPU actually is slower than that of the K6.
Guys are you serious? Rare? Cyrix 133mhz was one of the el-cheapo 'pentium imitations' back at the time and a lot of my friends used them because of the price. Most of them (CPUs, not friends Sticking Tongue Out ) ended in garbage - which must also be a reason that there are very few of them around nowadays.
As for the reason they ended in garbage - I'll leave it to you. Sticking Tongue Out
Well, perhaps you are confusing Cyrix 5x86 and 6x86, the 5x86 processor was used in 486 motherboards afaik, and was propably the fastest performer there. 6x86 processors on the other hand were designed to be Pentium replacements, and were otherwise faster than Intel Pentiums, but had slower floating point units. Also I think they put out quite a lot of heat, and your friends may have had bad cooling for them?
EDIT: A bit late 😜
😁
Wow, marvelous comparsion!!! I admire you - you had to devote much time on these tests. What I see cx5x86 contains really speedy fpu - (of course as for 486) and superior memory bandwidth.
There is only one thing that you miss in this comparsion - game benchmarks. Sometimes one gametest can clarify situation better than many synthetic tests bonded together. If I were you I would add Quake 1 timedemo.
Again, great job! Regards! (:
@rg100
Thanks for the info. If I get stuck I'll re-read all 8 pages on the other posting. I was planning on using an ISA video card if PCI doesn't work correctly. I'd only be using the floppy to run the benchies, so IDE/SCSI is not required. I probably won't get to this for a bit as I want to finalize my selection of benchmark programs for the next version of the benchmark comparison PDF.
@Tetrium
Cyrix MII PR433... That's a tough one to find. Even tougher is the MII-466. I'm not nearly as interested in collection, but more focused on nostalgic usage. Where did you post your regro rig photos?
@udam_u
Thank you for the possitive encouragement. I will add Quake1 timedemo and 3Dmark99max to the list along with a few other benchmark programs I deem important. I have about 60 programs to go thru before compiling version 2 of the pdf. I found Bytemark's benchmark program compiled for DOS/4GW that I might add as well.
I currently have my Cyrix 5x86-133 running at 3.68V. It was a struggle to figure out why it wouldn't boot into Win98SE/NT4. I degraded the BIOS settings and began turning on features one-by-one. Apparently the L2 cache needs to be set to Write-Thru for the Cyrix 5x86-133 to boot into Windows, however Write-back for the Cyrix 5x86-120 booted fine. I did a quick Floating-Point test in WinNT using the TaskManager -- Winamp CPU utilization dropped from 60% to 50% with the 133 Mhz when compared to the 120 Mhz cpu.
@feipoa, I am very interested in reading the next version of your benchmark comparison PDF, because I am looking forward to seeing what programs you are going to use. I notice you have about 60 to go thru. If any of these are fun to mess about with, and they are difficult to locate on the net nowadays, and they were free when they were last released on the net, would you consider uploading some of them please? Thanks a lot if possible.
I have a few questions to clarify about the Quake1 Timedemo that you guys are running,
1) Quake1 in DOS, Quake1 GL, or WinQuake? I am assuming the DOS varient whereby you boot into Win98SE, then shutdown into MS-DOS mode, correct? I have the DOS shareware 1.06 for testing, but may also have access to the others mentioned.
2) Timedemo demo1? I am assuming demo1 is the most commonly quoted frame rate benchmark?
3) There are many screen resolutions to choose from. In terms of the actual pixel resolutions, which one is the most commonly quoted? 1024x768, 800x600, 640x480, 320x240, or 320x200? rg100, which one have you been quoting?
It's Quake for DOS, shareware version 1.06. I do the following:
1) Run Quake at the DOS prompt. No need for any command line arguments. I always use DOS 6.22. If you are using W98, then you could press F8 at boot up time, and then select "5) Command prompt only" mode, to get to DOS.
2) Start a new game. No need to mess about with resolution sizes. Just leave it on default.
3) Make sure the game screen is full size. That's how I have done my tests. Other old and crusty benchmarks on the net have the screen size "2 notches" less than full screen. To get the game screen to be maximum size, press the + key a couple of times.
4) Press and hold down the CTRL key, and then press the ~ key. This should display the console.
5) type this -> timedemo demo1, then press enter. demo1 seems to be the most quoted that I have seen.
6) quickly press the CTRL and ~ keys again, so that when the demo runs, you don't see the console screen.
7) watch the demo until the FPS rate is displayed.
This extra information to streamline the process is very helpful. However, as I have already altered the default screen resolution, I do not recall what default is. Could you confirm which resolution you are using as default (i.e. run Quake, hit ESC, Video Options)? Having a number associated with the 'default' resolution is better for documentation as well. Thanks.
It's the very first resolution option in the list available - the smallest (and "grainiest") resolution possible. It's 320x200.
BTW, when you test this game on highly OC'd hardware, I wonder how quickly you will come across this problem - when you go to start a new game, it crashes with some kind of stack or page fault problem. That's why I like this demo. It's sensitive to OC'd CPUs and cache. Also, another interesting problem is running the Doom shareware demo. If the CPU and/or cache is too OC'd, the demo behaves oddly. It's really interesting. The man runs down the wrong corridors, and does daft stuff.
To run the Doom shareware (version 1.9s), you do this at the DOS command prompt:
doom -timedemo demo3
As with Quake, I ensure that the screen is full size before running the demo. You can do that by starting a new game, and pressing the + key a couple of times. Then, quit the game, and then launch the demo from the command prompt, as above.
@feipoa
I have about 60 programs to go thru before compiling version 2 of the pdf.
Gosh second revision is going to be really huge! (: In first turn you used to use benchmarks that measure overall CPU performance (cache, fpu), memory performance tests and some basic video performance tests. What kind of benchmarks are you going to use in second revision?
I currently have my Cyrix 5x86-133 running at 3.68V. It was a struggle to figure out why it wouldn't boot into Win98SE/NT4. I degraded the BIOS settings and began turning on features one-by-one. Apparently the L2 cache needs to be set to Write-Thru for the Cyrix 5x86-133 to boot into Windows, however Write-back for the Cyrix 5x86-120 booted fine. I did a quick Floating-Point test in WinNT using the TaskManager -- Winamp CPU utilization dropped from 60% to 50% with the 133 Mhz when compared to the 120 Mhz cpu.
I encountered similar problem during overclocking my cx5x86@150MHz. Firstly taught by experience with AM5x86 I turned off L1 cache. After this operation CPU correctly boots to DOS. Then I turned on cache and set it to write back mode which caused stability problems. However when I change setting to write thru processor regained stability. Finally I solved this problem by adding additional amount of thermal grease - CPU became stable in write back mode... Maybe your processor also isn't fully coated with a thermal grease? Did you test a little bit higher voltage - for example 4v?
did a quick Floating-Point test in WinNT using the TaskManager -- Winamp CPU utilization dropped from 60% to 50% with the 133 Mhz when compared to the 120 Mhz cpu.
Great! I think that everythink is all right - 133MHz has 10% faster clock than 120Hz.
Regards! (:
wrote:you are probably thinking of the 6x86 which certainly was commonplace. but the 5x86 133 is a different story.
You could be right. I never saw, or heard of '486 Cyrix'. To be perfectly honest, I always though that once over 100 Mhz, it stops being a 486 and starts being a 'mmx' CPU, or a similar imitation. K5@133 is of course, something else.
@udam_u
It very well may be related to the lack of thermal grease. I am not at the stage in testing where I want to grease-up the cpus, esp. a Cyrix 5x86.
In an effort to create the most definitive 486, socket 3 benchmark comparison chart, I will be adding the following:
CPUs:
Cyrix 5x86-80 (Do these even exist/work? We'll find out soon!)
Cyrix 5x86-133 w/ALL Enhancements OFF
Cyrix 5x86-133 using a Biostar MB-8433UUD v3.0 Motherboard
Intel DX2-66 (one of the most popular 486-era cpus)
AMD DX4-120 (another popular cpu before the x5 emerged)
AMD X5-200 (ADZ or ADW, whichever one I can get working)
Cyrix DX2-50 (I might add this)
AMD DX2-66 (I might add this)
Benchmarks:
***DOS-7.10***
Cpu Index v2.3
PiDOS
Landmark v2.0
Justin Benchmark DOS
Bytemark v2, 32-bit DOS
Doom 1.9s Timedemo, demo3
Quake 1.06sw Timedemo, demo1
***Windows 98SE at 1024x768x16bit***
SuperPi
Justin Benchmark WIN
Ziff-Davis Winbench96 (CPUMark32 & Graphics WinMark)
Ziff-Davis Winbench99 (CPUMark99, FPU WinMark99, DirectDraw)
WinTune98
Sandra99 Benchmarks (CPU, Graphics, Memory)
PassMark v4.0
3DMark99 Max
Landmark v2.0, Ziff-Davis Winbench96, Doom, Quake, WinTune98, PassMark, 3DMark99, and Bytemark were all very popular benchmark tools in the 1996-1998 era. I was going to add another very popular benchmark app, Final Reality Benchmark v1.01, however it crashes about 3 minutes into the benchmark, at the robots. I am not sure if this is related to the video card drivers, the directX version, or a bug in Final Reality. I might try upgrading to DirectX 8.0, or getting the DirectX 5.0 beta hack working in NT4 and testing it again. Final Reality requires DX5+.
I would really like to mess about with those 2 winbench progs. Were they free, when they were released? If that is true, please can you make them available? I can't find them on the internet. Thanks a lot if possible.
http://xtreview.com/id-314-v-Winbench-99.htm Link to winbench 99, seems it was shareware when new
EDIT: Remember to check that file with some AV prog, I haven´t tried to install that myself, I just made sure it starts to download!
Thanks a lot! 😀 I have downloaded it, and will test it tomorrow morning. Looking at feipoa's list of interesting benchmarks, I have managed to locate most of them on the internet. The ones that I am still hunting about for are:
PiDOS
Bytemark v2, 32-bit DOS
Justin Benchmark WIN
Ziff-Davis Winbench96 (CPUMark32 & Graphics WinMark)
The Dos Justin Benchmark program contains the Windows executable as well. It is also a command console executable.
The Justin DOS app (BENCHMRK.EXE, 16-bit, I assume) displays a memory bench and a floating point benchmark index value, while the Windows one (Benchmark.exe, 32-bit, I assume) will display Integer, Floating-point, Text Processing, I/O Processing, and Memory Access time for whatever algorithm is used. I really need to do more tests with this program on multiple cpus to see if it has any value. It is freeware, so I have enclosed it.
I do not see a copyright on the PiDOS. I found it in a forum, not too long ago. I can't recall which one now, I thought it was this one. You will want to edit your Pi.bat file to reduce the number of digits. Its default at 100,000 takes too long. I find that setting it to 25000 /f requires 19 sec, which is what I will bench with.
I found the Bytemark program compiled with DOS 4GW here,
http://www.tux.org/~mayer/linux/compare/index.html
Apparently, the compiler your choose makes a big difference. The program displays units of iterations/second, as well as index values normalized to a Pentium90. Without testing this particular Bytemark program yourself on a Pentium90, I would not place too much weight on the normalized values compared to the P90. I found a list on the web, for example, that gives the Cyrix 5x86-120 an integer rating of 0.97 (1.00 would be a P90), while I get only 0.90 with this particular Bytemark compilation. FYI, FP_FAST needs to be turned off on the Cyrix 5x86-120 for this program to compute the Neural Net, while the Cyrix 5x86-133 can have FP_FAST set ON.
As for the Ziff-Davis Winbench96, after reading the licence agreement, I am somewhat apprehensive to upload it here. The licence agreement states that the user shall not transfer the software, nor the licence agreement. The company is still sorta around. You might be able to find it if you do a search for WINBEN.ZIP, WINBEN.EXE, WINBEN96.ZIP, WINBEN96.EXE, WB96.ZIP, WB96.EXE. I no longer recall where exactly I found it.
Thanks a lot for the freeware links! I managed to find Winbench96 version 1.0. I'll test them all out today. Should be fun! 😀
wrote:In an effort to create the most definitive 486, socket 3 benchmark comparison chart, I will be adding the following: […]
In an effort to create the most definitive 486, socket 3 benchmark comparison chart, I will be adding the following:
CPUs:
Cyrix 5x86-80 (Do these even exist/work? We'll find out soon!)
Cyrix 5x86-133 w/ALL Enhancements OFF
Cyrix 5x86-133 using a Biostar MB-8433UUD v3.0 Motherboard
Intel DX2-66 (one of the most popular 486-era cpus)
AMD DX4-120 (another popular cpu before the x5 emerged)
AMD X5-200 (ADZ or ADW, whichever one I can get working)
Cyrix DX2-50 (I might add this)
AMD DX2-66 (I might add this)
A few hints:
The AMD 486's basically come in 3 varieties, all 3 may influence benchmark results. They have 3 different cache arrangements: 8kb WT, 8kb WB and 16kb WB. The AMD 5x86's are all 16kb WB btw, but the AMD 120 often has 8kb WT or WB. 😉
Also no Intel DX4 (WT or WB) nor POD-83?
On another note...Intel DX4-100 @ 2x66Mhz...Intel 5x86-133? Would be interesting to see what Intel might have done this way 🤣.
Btw, thanks for the download links! I´ve been looking for a (ANY!!) pi benchmark that can run from a bootdisk for a while now.
One question, does this version of superpi display partial seconds (like "19.13s") or only whole seconds (like "19", or just "20")?
