The Ultimate 486 Benchmark Comparison

[sliderider]

We need a decoding guide for the production numbers on the underside of the Cyrix/IBM/SGS-Thompson 5x86 chips.

[feipoa]

The undersind of the SGS-Thompson 5x86 chips are the same as the Cyrix pieces. Are you looking for any specific information? The information contained on the back identifies the fab location, dye run, week, year, and lot.

IBM made the chips longer than Cyrix did. The newest IBM 5x86 I have is from Week 38, 1996, however I have heard rumours that IBM made this chip into 1997 [source: Anonymous Coward]. Cyrix stopped making the chips Week 7, 1996. The Week 38, 1996 IBM chip did not overclock well.

[popper]

@feipoa

I am very impressed! Thank you (and all the helping hands) for your work - this comparison ist exactly what i was searching all over the web for so long....


Does anybody know the story behind intel's "DX scheme"? I mean: Why the hell did they call their DX4 DX4 and not DX3 as the multi implies?
DX = 1 x multi
DX2 = 2 x multi
DX4 = 3 x multi

Was there a DX3 intended to be created?

[feipoa]

You are welcome! I eventually plan on redoing the comparison on a different motherboard, add some benchmarks, and add all possible 486 CPUs.

To answer your question,

From Wikipedia, Intel named it DX4 (rather than DX3) as a consequence of litigation with AMD over trademarks.

From micro.magnet.fsu.edu, Since Intel reserved the "DX3" designation for a 2.5 times clock multiplier that never reached the marketplace, they had to create the newer moniker, the "DX4"

[popper]

Thank you!

and: Thank you!

It seems that all questions about naming their products relate to marketing - interests; but there is still a broken logic, when they wanted to call it "DX3" without giving it that multi.

[idspispopd]

IMO multipliers are relative. Just look at (Super-)Socket 7: 1.5x gets interpreted as 3.5x (for Pentium MMX and similar clocked CPUs), 2.0x as 6.0x (K6-2/3), and the K5 taking both 1.5x and 2.0x as 1.5x, 2.5x as 1.75x and 3.0x as 2.0x.

[popper]

This post is a note to self
6) 486 CPUs to add: UMC U5SX-33, UMC U5SX-40, AMD X5-150, Intel DX2-133, AMD DX2-133, Intel 486 DX-50, Intel 486 DX4-100-WT, AMD DX4-100-8KB-WB, AMD DX2-66-16KB-WB, Intel DX2-66-8KB-WB
I will not be doing this anytime soon.

Although i noticed "not...soon": i am very interested!


I was watching some videos on YT one member of our vogons-forum once created. One of them brought the UMC U5S in focus, which was a very nice addition to my knowledge. Interestingly HighTreason benched the U5S ("new hooker") with the 16 Bit Topbench, where we can see very high results - higher then any 486 class CPU within the same MHz-range.
Additional i think i can remember that he spoke about a "pentium-like" architecture for the UMC U5S design (in some way). According to wiki the CPU has significantly higher troughput at integer division.... but that can't be the only reason, can it?

Feipoa / HighTreason, do you have any further (empiric) information about that strange CPU (which obviously is no clone)?

[feipoa]

I have yet to test my UMC CPU.

My next course of action is to fully test all my 386 motherboards and CPU combinations for the fastest possible 386 motherboard with a Ti486SXL CPU. I am about 10% finished with that job. Following this course, I would likely begin the remake of the 486 benchmark comparison. In this manner, I can include 386 chips on the benchmark comparison. Unfortunately, my house is in the planning stage for renovation/repair again so my vintage hobby is on hold.

I also have a fat stack of untested benchmark programs that I would like to evaluate as possible candidates for the remake of the 486 comparison.

I would be very surprised if the UMC U5SX has "Pentium-like" architectural features.

[Turboman]

How do you tell what Intel Overdrive 100 is WB? I have 169 pin SZ957, 168 pin, SU004, SZ926, and SZ959, and have you ever thought about testing the Kingston Turbochip?

[Tertz]

It's not clear for Quake is "console off" in table means "viewsize 120", or text console's removing during testing by ~.
Also it's interesting what is speed in DOSBox of top 486 - Intel 486 DX4 100 MHz at its standard frequency. DOSBox should work in Win9x.

---
To check emulation speed in DOSBox 0.74 may be used Speed Test by Agababyan. (inside DOSBox run it as: speedtst >res.txt).
DOSBox's settings: scaler=none, core=dynamic, cputype=486_slow, cycles=max; other settings are default
Additionally would be good to test with core=normal, but it's less important.

If you did not use DOSBox: to mount a folder as drive 'C' in DOSBox is needed to input 2 strings in DOSBox's .conf section [autoexec]:
mount c c:\folder\
c:
To create default .conf: rename or delete the current .conf, then run DOSBox and it will create default .conf
---

It's interesting to see can 486 100 MHz give at least 286 level in DOSBox. If someone has this CPU, would be nice to see your results. Other interesting CPUs (at standard frequencies only and preferably on Intel chipsets) in DOSBox are: Intel 486 DX2 66 MHz, Pentium 166 MMX, Pentium 100, Pentium 200 MMX, etc.

[feipoa]

@Turboman
I think all Intel DX4 overdrives are write-through.

@Tertz
"console off" = text console minimised by pressing ~

[Tertz]

"console off" = text console minimised by pressing ~

Just in case. Often "viewsize 120" is used for timedemo benchmarks with Quake, like there. For Quake 1.06 320x200 demo1 difference coefficient is ~1.133. So your P100 in testing with 26.9 fps would get 26.9/1.133 = 23.74 fps.

[max1024]

I have just on default settings run speedsys on Pentium Overdirve 83Mhz, Shuttle HOT-433, 128Mb EDO Ram (4 banks), 256Kb chache, there is the result. Is it normal for such config? I will plan to make some tweaks for RAM in bios and try to change FSB to 40Mhz. Is it posible to cahnge the multiplier on POD83? or it is locked.

Also I've installed W2K SP4 and run Aida 64 chache and memory benchmark.

Is it 256 cache works? Aida 64 do not detecs it, as and cpu-z to.

[feipoa]

Which version of Shuttle Hot-433? You can only down multiply the POD83 by removing the fan.

The throughput results for your POD83 system are below optimal. Please refer to the data table in this thread to determine what the optimal values should be. i.e. L1/L2/RAM @ 163/47/37.

Your cache is working as determined by the Speedsys chart. A lot of Windows-based benchmark programs cannot properly detect direct-mapped cache.

If you are planning to run Win2000 on this motherboard, I think you will find that POD83 to be less than impressive. The AMD X5-160 would be faster with the GUI.

[chrisNova777]

big thanks to the orignal poster for this comparison + visual reference!

[Skyscraper]

I could not find a general 486 Speedsys thread so Im posting some results here.

Asus PVI-486SP3 with 256KB cache, an unknown AMD DX4-100 probably a NV8T and 2x8MB double-sided FPM.
The CPU cache is jumpered as WT as I think the CPU dosnt support WB.
I can not upgrade to a newer BIOS than 0205 because of a 12V flash chip which isnt supported by newer BIOS versions.



Cachechk: "Setup Defaults".

Speedsys: "Setup Defaults".

Optimizing the BIOS settings "DRAM Speed" from "faster" to "fastest" and "DRAM Write Cycle Post" from "1" to "0" added alot of memory performance.


Cachechk Optimized settings.

Speedsys Optimized settings.

It was ages ago I played with a fast 486 so I do not really know if the numbers looks OK or not.
I hope the performance is as it should.

[Tertz]

Skyscraper
Would be nice if you tried Speed Test in DOSBox on default BIOS settings. There is no clear data still about DOSBox emulation on 486 CPUs.

[Skyscraper]

Skyscraper
Would be nice if you tried Speed Test in DOSBox on default BIOS settings. There is no clear data still about DOSBox emulation on 486 CPUs.

This system only has DOS 6.22 installed at the moment, sorry.



If I load "BIOS deafults" and set all settings manually I get even more memory throughput than if I load "Setup defaults" and set everything manully eventhough the settings are the same...
The tiny bit of lost CPU score compared to the results above is because I added a AWE64, I think.

The memory performance seems rather good for FPM @33 Mhz FSB

[matze79]

who wants dosbox on a 486er ?

[Tertz]

who wants dosbox on a 486er ?

Those who want XT era games linked to timer. Maybe 286 level is possible too. Those who want CGA composite, Tandy music, etc nontypical hardware. Those who have other compatibility issues wich DOSBox may possibly overcome. Those who want booter games and don't want to write floppies for them.
It's not even obligatory to setup Win9x to use DOSBox, as it works in HX DOS Extender. But in Win9x it may to work faster.