ph4nt0m wrote:

Another world's fastest 486. […]
Show full quote

Another world's fastest 486.

Cyrix 5x86-120GP 3.6V S1R3 (year 1995 week 41) @ 133MHz (2x 66MHz) 3.7V
LuckyStar LS-486E rev. D. with dual banked 256Kb cache (3-1-1-1 read timings)
64Mb EDO RAM (2x 32Mb Kingston Kingston KTM8X32L-70ET)
Diamond Monster Fusion 16Mb PCI (3Dfx Voodoo Banshee)

What makes it special is all 9 15ns 32Kx8 ISSI IS61C256AH-15J cache chips have been desoldered and replaced with 8 10ns Cypress CY7C199D-10VXI + 1 8ns EliteMT LP61256GS-8 for the tags. The 15ns chips didn't work at 66MHz. First the 8ns EliteMT chip went in and the cache started to work at 3-2-2-2. Then the 10ns Cypress chips made it work at 3-1-1-1. I have also replaced all those poor Tayeh 10uF/16V capacitors with tantalum 100uF/10V and 33uF/25V, and some 100nF MLCCs with 5uF. The power regulator was modded to do 3.7V instead of 4.1V in addition to 3.45V. A custom cooler with a dual ball bearing fan which runs off 5V instead of 12V to reduce noise.

Now, that's what I call dedication! Thumbs up for modding hardware to achieve these specs!

Very nice to see another Cyrix 5x86 running at 2x66 MHz! I'm surprised that the 8/10ns cache allowed for 3-1-1-1 timings. I'm also surprised that you are able to run any S1R3 chip at 133 MHz. Makes me wonder why they switched to S0R5 chips, which I had assumed was for higher yield at 120 MHz+. What is the datecode on the back of your chip? It looks like your motherboard has a socket for a DIP-40 KBC. It would be pretty neat if you could mode the board for a native PS/2 mouse port.

On my Biostar MB-8433UUD system, I had modded the board for use with 1024K of L2 cache, however after years of on/off testing for stability, I was finding that with 1024K, I had set the memory read wait state to 2 WS to achieve the stability I desired. Rather than accept this performance penalty, I put the L2 cache back to 256K. I modded another MB-8433UUD board with additional jumpers so that I can swithc between 256K, 512K, and 1024K (all double-banked) without having to do any soldering each time I switched the cache quantity. On my to-do list is to see if I can use 512K and still achieve the 1 WS on memory read.

I pulled my IBM 5x86c-133/2x system out of the closet and re-ran Speedsys and the Quake 1.06 timedemo1 at 320x200. quake.exe -nosound -nomouse

I've attached the results. It reported 19.1 fps. As this is DOS, branch prediction is enabled. I'm using a Matrox G200 with Voodoo2. CPUMark99 reports only 6.14, but in Windows, I have branch prediction disabled.

Your Quake timedemo result seems a little slow for your hardware. Are you using Phil's benchmark pack w/Quake or Quake 1.06 demo by itself? I am using Quake 1.06 by itself.

Strange that Speedsys is misreporting the processor speed. Does your BIOS identify it as a Cyrix 5x86-133?

Did you investigate the 4DPS BIOS I supplied in the other forum?

Looking at the speedsys charts, it appears as if our L2 RAM is similar, but your system has improved main memory results.

That's pretty badass. I've seen 486 boards from Russia with surface mount cache chips like yours on DIP adapters. How much did you pay for the 8ns part?

treeman wrote:

any chance of a quake score?
how come you soldered the new cache back in or was it the previous owner? No sockets that size or they are smd, can't make out on picture

very nice clean system!

These chips are all SOJ-28, so I have soldered them with a heat gun.

Quake 1.06 timedemo1 = 17.1 fps
GLQuake 1.09 timedemo1 = 20.6 fps (640x480 16bpp)

Seems not so high. I have to check the settings, drivers and maybe try different cards for GLQuake.

aries-mu wrote:

You even altered the motherboard and components electrically almost entirely! I'm not an electronic guy, I would never have replaced capacitors with new ones having different specs!

Curious: What kind of hard (or solid state) drive have you installed? Please tell me some SD or CF...

That's absolutely alright to replace electrolytic capacitors with tantalum of a higher capacitance. BTW these tantalums intalled are NOS made in 1995. I guess they cost maybe 1 USD each back in the day, so getting them installed on this inexpensive board by the manufacturer was out of question.

ST9546A is an old 2.5" Seagate 540MB 4500rpm IDE drive for notebooks. Not very fast indeed, but it will do for now.

feipoa wrote:

Very nice to see another Cyrix 5x86 running at 2x66 MHz! I'm surprised that the 8/10ns cache allowed for 3-1-1-1 timings. I'm also surprised that you are able to run any S1R3 chip at 133 MHz. Makes me wonder why they switched to S0R5 chips, which I had assumed was for higher yield at 120 MHz+. What is the datecode on the back of your chip? It looks like your motherboard has a socket for a DIP-40 KBC. It would be pretty neat if you could mode the board for a native PS/2 mouse port.

Year 1995 week 41. It seems IBM had a hard time getting good yields on S1R3. Do 133MHz 5x86 S1R3 even exist? So they switched back to S0R5 for a few months until 5x86 was replaced in production by 6x86. It's a bit sad because AMD manufactured their 5x86 until 2004. Cyrix 5x86 die shrinked from 650nm to 350nm could be very nice.

I have installed both sockets for KBC. The board POSTs with some 40-pin KBCs, but there are missing resistors and jumpers near the socket which have to be populated in order to enable full functionality.

Anonymous Coward wrote:

That's pretty badass. I've seen 486 boards from Russia with surface mount cache chips like yours on DIP adapters. How much did you pay for the 8ns part?

I have salvaged the 8ns chip from a dead MVP3 based mainboard. They are out of production for a long time. The only new 10ns 5V 32Kx8 async SRAM chips I could buy online from non-Chinese distibutors were made by Cypress or Alliance. The date code on the Cypress chips I have installed is year 2018 week 31.

Is a 5x86 considered a 486 ?
or should it be "Worlds fastest Socket 3 Motherboard" ?

feipoa wrote:

I pulled my IBM 5x86c-133/2x system out of the closet and re-ran Speedsys and the Quake 1.06 timedemo1 at 320x200. quake.exe -nosound -nomouse

I've attached the results. It reported 19.1 fps. As this is DOS, branch prediction is enabled. I'm using a Matrox G200 with Voodoo2. CPUMark99 reports only 6.14, but in Windows, I have branch prediction disabled.

I see it's timedemo2 on your screen shot.

Intel486dx33 wrote:

Is a 5x86 considered a 486 ?
or should it be "Worlds fastest Socket 3 Motherboard" ?

AMD 5x86 is identical to 486 by architecture. There are 486DX4 chips with 16Kb write back cache. Some AMD 5x86 were labelled 486DX5.

Cyrix 5x86 is a cost reduced version of 6x86. It has only one ALU like all other 486 chips. Pentium Overdrive for Socket 3 has two ALUs like regular Pentiums.

ph4nt0m wrote:

... […]
Show full quote

...

aries-mu wrote:

You even altered the motherboard and components electrically almost entirely! I'm not an electronic guy, I would never have replaced capacitors with new ones having different specs!

Curious: What kind of hard (or solid state) drive have you installed? Please tell me some SD or CF...

That's absolutely alright to replace electrolytic capacitors with tantalum of a higher capacitance. BTW these tantalums intalled are NOS made in 1995. I guess they cost maybe 1 USD each back in the day, so getting them installed on this inexpensive board by the manufacturer was out of question.

ST9546A is an old 2.5" Seagate 540MB 4500rpm IDE drive for notebooks. Not very fast indeed, but it will do for now.

...

Thanks!

And I wonder: are your PCI slots running at 33 MHz? Let's say you wanted to set your CPU at 120 MHz (or even use an Am486 DX4-120 Enhanced WB), would you have been able to set the FSB at 80 MHz, CPU at 120 MHz, and the PCI slots at a real 40 MHz speed (instead of those weird downclocked configurations like 27 MHz or something)?

Thanks again.

aries-mu wrote:

And I wonder: are your PCI slots running at 33 MHz? Let's say you wanted to set your CPU at 120 MHz (or even use an Am486 DX4-120 Enhanced WB), would you have been able to set the FSB at 80 MHz, CPU at 120 MHz, and the PCI slots at a real 40 MHz speed (instead of those weird downclocked configurations like 27 MHz or something)?

Thanks again.

They are running at 33MHz indeed. There is a jumper to set the PCI bus speed to 1/2 of the system bus speed. The ISA bus could be clocked at 1/3 or 1/4 of the PCI bus or 7.159MHz. Only the last option is feasible. The clock generator doesn't support system bus speeds higher than 66MHz.

wow, thanks so much

Is there a thread with the best motherboard jumper and bios settings for each CPU type ?

Example :

CPU
● AMD 5x86-133 (4 x 33 MHz FSB)

Timings
Cache
W1
W0
Etc....

I benched my amd 5x86 160 with pretty stock common parts, it is no match for the cyrix 133

amd 133 clocked at 160
MB-8433UUD-A V3
1 x 32 ram 72pin chip with 256 cache
s3 vision card speedsys says 2mb I remember it as 4mb ah well
running of cf card

ph4nt0m raised a good point which i never thought about, we seem to be using different quake versions, I use the one from inside the dosbench package which Im pretty sure I got from vogons with the menu which has speedsys also, so maybye those results are not that great benchmark when comparing different versions?

but clearly the cyrix at 133 outperforms the amd at 160

ph4nt0m wrote:

I see it's timedemo2 on your screen shot.

That is because once the timedemo1 finishes, the timedemo2 starts automatically. The 19.1 fps is from the timedemo1 result, not timedemo2. How do you guys get timedemo1 to stop once the benchmark concludes? For me, the timedemo's keep playing in loop.

I am not using the benchmark from the Phil/Vogons pack. That pack has been tweaked to make it harder on Pentium-class systems, thus lower results show up on 486 systems. I"m using the original Quake Shareware version 1.06. Timedemo1.

feipoa wrote:

ph4nt0m wrote:

I see it's timedemo2 on your screen shot.

That is because once the timedemo1 finishes, the timedemo2 starts automatically. The 19.1 fps is from the timedemo1 result, not timedemo2. How do you guys get timedemo1 to stop once the benchmark concludes? For me, the timedemo's keep playing in loop.

glquake -width 640 -height 480 -bpp 16 -fullscreen -nomouse -nojoy -nosound -noipx +timedemo demo1

Quake 1.07 doesn't run on 386 systems. That's why we benchmark with 1.06

I downloaded the standalone quake 1.06 and ran the timedemo manually from console, got exactly the same score 16.6

But like feipoa said it doesn't stop and automatically goes to demo2, the one I always have been using from dosbench stops after demo1 by itself.

I guess then those scores are compatible 16.6 on x5 vs 19.1 on cyrix/ibm 133

However my cache is stock 12ns from memory so maybe could squeeze 1 more fps out with faster cache chips and perhaps L2 write back enabled bios.

On stock timings I get a high 14 fps or low 15 fps, its been a while

Really? I've heard from people that the Quake shareware 1.06 in Phil's benchmark pack runs slower than stock Quake 1.06. Hopefully somebody will mention how to stop the autoplay.

maybye for a pentium, but not a 5x86?
I guess you could try phils benchmark vs stock quake and see if the cyrix gets a different score

treeman wrote:

I downloaded the standalone quake 1.06 and ran the timedemo manually from console, got exactly the same score 16.6 […]
Show full quote

I downloaded the standalone quake 1.06 and ran the timedemo manually from console, got exactly the same score 16.6

But like feipoa said it doesn't stop and automatically goes to demo2, the one I always have been using from dosbench stops after demo1 by itself.

I guess then those scores are compatible 16.6 on x5 vs 19.1 on cyrix/ibm 133

However my cache is stock 12ns from memory so maybe could squeeze 1 more fps out with faster cache chips and perhaps L2 write back enabled bios.

On stock timings I get a high 14 fps or low 15 fps, its been a while

These are my settings for AMD 5x86-133-P75 over clocked to 160mhz.
I am having ram and cache performace problems.
I am using benchmark tools from Philscomputerlab.com
Computer works fine just problem playing MP3’s.
I am not sure if its the bios, CPU, or ram , or audio driver ?
How can I check ?

heh you forgot the quake score first

OK. Apparently the +timedemo demo1 flag works for the DOS demo 1.06 as well. I could have swore I tried this years ago and it didn't work; so I assumed it was for later edition of GLQuake . On my AMD Am5x86-160 system, I ran the timedemo with the +timedemo flag and thru the console directly and the scores were the same.

I'm using an S3 968-based VLB card in my Am5x86-160 system and receive 17.6 fps, however I've seen a similar comparison by another user in which using a Trio64 upped their scores another 0.3 fps. So perhaps 18.0 fps is possible with optimal conditions.