The magnificient four

[JoJo_ReloadeD]

Here is my last overcloking job, just an attempt of getting the performance limit on every PC class, from the XT era to the 486 era.

Not trying to beat a world record or anything, just trying to see how far these machines can go for the thrill of it. Those are the 4 machines:

- XT Class. NEC v20 @ 15.2 mhz:



Performance achieved: Slightly better than a 80286 @ 8mhz.

- 80286 class. Harris 80286 @ 27.5 mhz:



Performance achieved: Comparable to the 80386sx @ 33mhz.

- 80386 class. Am386 @ 50 mhz:



Performance achieved: Comparable to the 80486dx @ 33mhz.

- 80486 class. Am5x86 @ 166 mhz:



Performance achieved: Comparable to the Am5k86 @ 100mhz

In all cases the benchmarking software is the same:

- Landmark system speed test 6.0
- Trixter's Topbench 0.38a
- Superscape 3d Bench 1.0

If anyone has any doubt about the systems or wants to see some software running in any of them just ask and I will gladly try it and do captures and/or videos.

 

[Chuck(G)]

How stable are these setups, long-term?

I learned a long time ago never to buy RAM or CPUs from a hard-bitten overclocker.

 

[JoJo_ReloadeD]

How stable are these setups, long-term?

I learned a long time ago never to buy RAM or CPUs from a hard-bitten overclocker.

I have checked those over an hour with 3dbench and worked flawlessly, altough I had to add heatsinks to the 386, 286 and v20.

 

[kyodai]

I thought the V20 was rated "8-16 MHz"? So obviously 15.2 MHz does not sound THAT impressive to me.

 

[Xacalite]

For the fastest XT, I would search for a V30.
For the fastest 386... I guess you overclocked a 386DX/40, right? Indeed, it's the fastest "true 386", but there was also a bunch of 386-pin-compatible chips with internal cache, I'm not sure what category they should compete in.
Also, NexGens only support 386 instruction set, so...

 

[Chuck(G)]

A V30 is a 16-bit BIU chp--you can't plug it into an XT without some sort of adapter card--and then, it really wouldn't be a real XT, would it?

I once bought a bunch of used DDR from a fellow who was a freakish overclocker. It was impossible to get the stuff to work at the rated voltage and speed. You had to use elevated voltage on those, which leads me to believe that overclocking stuff is not the best way forward.

An hour is not much--try running the things for a month solid at the overclocked speed and then report back--I'd be curious to hear what the outcome was.

 

[Xacalite]

A V30 is a 16-bit BIU chp--you can't plug it into an XT without some sort of adapter card--and then, it really wouldn't be a real XT, would it?

Sure, but the OP mentioned "XT Class" rather than "IBM PC XT", so I think anything below 286 is acceptable.

 

[GiGaBiTe]

I once bought a bunch of used DDR from a fellow who was a freakish overclocker. It was impossible to get the stuff to work at the rated voltage and speed. You had to use elevated voltage on those, which leads me to believe that overclocking stuff is not the best way forward.

Of course overclocking components (especially RAM and CPUs) is going to cause them to degrade far more quickly than under normal use. I'd say overclocking is fine while it's still easy to get replacements for said equipment you're overclocking.

This is why you have to be careful on Fleabay when buying used CPUs that were popular with overclockers because you could get a land mine.

 

[Anonymous Coward]

I'm curious about the am5x86 at 166MHz. Most people overclock this chip using a 40MHz bus to get 160MHz. How did you get to 166? Are you using an 83MHz oscillator?

Which motherboard are you using for your 50MHz 386? The CPU itself doesn't have too much of a problem reaching 50MHz, but usually 386 motherboards can't handle it.

 

[SpidersWeb]

Which motherboard are you using for your 50MHz 386? The CPU itself doesn't have too much of a problem reaching 50MHz, but usually 386 motherboards can't handle it.

Not sure what he's using, but I've had a couple of boards now that were 386/486 hybrids made in the later 486 years - the 386 CPU being a low cost option but the ability for high clocks, VLBus, etc was there along with a 486DX upgrade socket on occasion. Normally no-name brands, most of mine were non-runners usually with an OPTi chipset.

 

[JoJo_ReloadeD]

I'm curious about the am5x86 at 166MHz. Most people overclock this chip using a 40MHz bus to get 160MHz. How did you get to 166? Are you using an 83MHz oscillator?

Which motherboard are you using for your 50MHz 386? The CPU itself doesn't have too much of a problem reaching 50MHz, but usually 386 motherboards can't handle it.

You are right, I'm using a 40mhz bus to get 4x40 = 160, but all the tests show that the processor is around 162 or 163 mhz. It's a Kingston Turbochip, I don't know if that's the reason...

This is the motherboard used on the 386:
http://stason.org/TULARC/pc/motherboards/E/ELITEGROUP-COMPUTER-SYSTEMS-INC-386-UM-386.html#.ViSLdK3tlBc

I had to change the oscillator (from 80mhz to 100mhz) and the cache chips for a faster ones. They were rated at 20ns and needed at least 15ns to work without errors.

 

[Gabucino]

You could compare it to my 5x86-160MHz. I use Speedsys (and for pre-386: CheckIt) for benchmarking: http://data.gabucino.hu/5X86/speedsys.gif

More comparisons here: http://gabucino.hu/inv.htm

 

[JoJo_ReloadeD]

You could compare it to my 5x86-160MHz. I use Speedsys (and for pre-386: CheckIt) for benchmarking: http://data.gabucino.hu/5X86/speedsys.gif

More comparisons here: http://gabucino.hu/inv.htm

Here is the 386 and the 486:

 

[Xacalite]

I'm curious about the am5x86 at 166MHz. Most people overclock this chip using a 40MHz bus to get 160MHz.

Yes, it was common back in the era to overclock 5x86 ADZ variant (designed to work in higher temperature) to 160 MHz, the other variant - ADW - probably wasn't so tolerant.

There were even fake chips with "160 MHz" markings on them.

 

[Anonymous Coward]

Be aware that with the 386 the CPU benchmark number is distorted (inflated) when an FPU is not installed.

Also, it would be nice to see your memory graphs.

 

[JoJo_ReloadeD]

Be aware that with the 386 the CPU benchmark number is distorted (inflated) when an FPU is not installed.

Also, it would be nice to see your memory graphs.

 

[Anonymous Coward]

Is it possible to squeeze out any more memory performance on either system, or is everything pretty much running on the edge of stability already?

 

[JoJo_ReloadeD]

Is it possible to squeeze out any more memory performance on either system, or is everything pretty much running on the edge of stability already?

I'm afraid they are at their limit.

The v20 has 70ns memories (needed memories faster than the 100ns that originally had its motherboard), but I was unable to get any more speed from it.
The 286 has 60ns memories, resulting on a 0ws system.
The 386 has 60ns memories, resulting on a 1ws system. To get a 0ws system I think I would need 40ns memories but think they doesn't exist.
The 486 has 60ns memories, resulting on a 1ws system. Don't know it there are fpm simms faster than 60ns. In any case I don't have any.

By the way, I forgot to mention that the 486 has a Matrox 2mb pci card + 2 Voodoo2 cards. Here is it running Half Life:

 

[GiGaBiTe]

Yes, 50ns FPM SIMMs exist, at least the 72 pin varieties do.

I bought a Kingston 32 MB FPM SIMM in the late 1990s for my Macintosh LC III, which I still have working. It uses Toshiba memory chips rated at 50ns.

I'm surprised that Half-Life runs on your 5x86-166 that well. I used to have a similar setup to yours and I couldn't get the game to run beyond slideshow frame rates. You can see the game does get a bit unstable though because the train starts jerking side to side at some points.