I wouldn't count on 200mhz in the first place. I've only ever seen it stable at 160MHz for the highest (4x40) and 150mhz seemed unstable for me (3x50), i'd hate to think what 4x50 would do. I'm not exactly fond of smelling magic smoke.

I never understood why no company ever released a 486 to Pentium upgrade adapter (the shitty one from Intel doesn't count). We had these things for 8088s, 286s, and 386s called "dinguses". Was there something so different about Pentium that made a socket adapter too difficult to produce?
Ideally what I'd want though is a newly fabricated 486 CPU that had really high multipliers but bus speeds of 33 and 40MHz.

Yeah, I can completely agree. Can anyone confirm that the chart @ wikipedia is correct?

http://en.wikipedia.org/wiki/Am5x86

It's saying that the DX5-133W16BGC stepping requires only 3.3V, while every other stepping requires 3.45V. Technically this would mean that this stepping can run @ 133MHz at less voltage, so theoretically throwing 5V at it may get it higher/more stable at/above the 200MHz mark. Does anyone have this stepping and can confirm?

numeriK wrote:

Yeah, I can completely agree. Can anyone confirm that the chart @ wikipedia is correct?

http://en.wikipedia.org/wiki/Am5x86

It's saying that the DX5-133W16BGC stepping requires only 3.3V, while every other stepping requires 3.45V. Technically this would mean that this stepping can run @ 133MHz at less voltage, so theoretically throwing 5V at it may get it higher/more stable at/above the 200MHz mark. Does anyone have this stepping and can confirm?

My guess is that what you would ideally want, is a 5x86 with a very late manufacturing date. Iirc these chips were made till at least 1999!

But even if you were to manage 266Mhz, it would, I reckon, drastically reduce the lifespan of both your chip and your motherboard. Not to mention it could be a risky business (the board might blow something up attempting 266Mhz).

But even if you were to manage 266Mhz, it would, I reckon, drastically reduce the lifespan of both your chip and your motherboard. Not to mention it could be a risky business (the board might blow something up attempting 266Mhz).

If the mobo runs 60 or 66mhz FSB even stably, i would not be afraid of using it for multi 4.
And the mobo does not care what internal speed the cpu does run on.
The cpu is so low power, i doubt with normal cooling these days it will burn out. but 5v can make it unstable of course. a steady sweetspot voltage should be better that upping the vcore all the time.
And Mr phasechange or Teccie does help.

I too dream of this thing, but i have only one mobo for the task. (my Mach2 is ready for chilling on the bench with the amd dudes)
But soon will have 20 5x86 P75 cpu's for testing the ceiling. 😁

JaNoZ wrote:

If the mobo runs 60 or 66mhz FSB even stably, i would not be afraid of using it for multi 4. And the mobo does not care what int […]
Show full quote

But even if you were to manage 266Mhz, it would, I reckon, drastically reduce the lifespan of both your chip and your motherboard. Not to mention it could be a risky business (the board might blow something up attempting 266Mhz).

If the mobo runs 60 or 66mhz FSB even stably, i would not be afraid of using it for multi 4.
And the mobo does not care what internal speed the cpu does run on.
The cpu is so low power, i doubt with normal cooling these days it will burn out. but 5v can make it unstable of course. a steady sweetspot voltage should be better that upping the vcore all the time.
And Mr phasechange or Teccie does help.

I too dream of this thing, but i have only one mobo for the task. (my Mach2 is ready for chilling on the bench with the amd dudes)
But soon will have 20 5x86 P75 cpu's for testing the ceiling. 😁

20! 😁
You should take pics with a camera of the monitor so you can see the boot speeds of the CPU's 😁

In theory 233 or 266MHz might be possible with an am5x86. The Intel Pentium MMX chips were fabricated on the same .35 micron process and were able to reach those speeds. I'm sure there must be more to it than that though. I also recommend getting a 5x86 with a 2000 manufacture date.

I don't think it would be worth it. Even if you could get a CPU and motherboard that play nice together at speeds over 200mhz, you likely wouldn't see much difference if any when running apps on such a system. The law of diminishing returns would factor in here. The buses on the motherboard would be the most limiting factor to performance and you really can't do anything about those. It would be like adding 100hp to the output of the engines of an aircraft carrier. The extra 100hp would be there, but wouldn't add anything noticeable to the overall performance of the ship.

I agree that it wouldn't be practical, but it would be totally worth it to me. Sure there will be some bottle necks, but not all applications make heavy use of I/O.

I would like 233mhz 486 !
Quake loves this higher FPU performance compared to 133-180 models.

Memory states play a big factor to Quake's speed on a 486 too.

If my math serves me correctly:

15ns L2 cache has a maximum theoretical speed of 66MHz
10ns L2 cache has a maximum theoretical speed of 100MHz

60ns DRAM has a maximum theoretical speed of 66MHz @ 2/2 WS
50ns DRAM has a maximum theoretical speed of 80MHz @ 2/2 WS

So, upgrading both the L2 and DRAM to lower frequencies will definitely give you more headroom, however - like I said before - it's not the DRAM or cache that's currently limiting me.

I can boot and be 100% stable @ 60MHz x 3, and boot and be far less stable @ 66MHz x 3.

It just seems like the CPU itself doesn't play nice at and over the 200MHz mark. Even when I'm at 50MHz x 4 the machine isn't 100% stable... I need more CPU's to test! 😎

numeriK wrote:

It's saying that the DX5-133W16BGC stepping requires only 3.3V, while every other stepping requires 3.45V.

I have a 133W16BGC chip. It didn't even turn on at 4x50 and 5 V. It also has a mid-1998 production date. It is really hit-or-miss for which AMD X5 will run well at 200 MHz. There is one forum member, rg100, who has a chip which runs quite well at 200 MHz. The best chance for pushing 233 Mhz might be to test his chip, but I do not think above 200 MHz will work. But if you are going to try it anyway, it is best to begin with your L2 cache disabled.

If your PCChips M919 v1.5 is anything like the v3.4, it doesn't handle 66 MHz well with any timings, where 'well' = at least Windows98 stable. If it is FPU you are after, previous tests seem to indicate that an AMD X5-200 is slower than an IBM 5x86 2x66.

Anonymous Coward wrote:

Ideally what I'd want though is a newly fabricated 486 CPU that had really high multipliers but bus speeds of 33 and 40MHz.

Sounds like a Cyrix MediaGX. They could run at 33 x 9.0. It is very similar to a Cyrix 5x86 and adds MMX. It can also use SDRAM. Results will be on the Ultimate 6x86 Benchmark Comparison (MediaGX testing complete, working on VIA chips now and have run into a little snag).

What if am5x86 works fine at higher frequencies but is just sensitive to higher bus speeds? Without an interposer it is not possible to test this theory unfortunately.

Anonymous Coward wrote:

What if am5x86 works fine at higher frequencies but is just sensitive to higher bus speeds? Without an interposer it is not possible to test this theory unfortunately.

Are there any socket3 boards that have a 1/2 FSB to PCI divider? If so that would help a ton.

I know what you're saying regarding higher bus speeds, but I have run at 60MHz x 3 with near 100% stability. I've also got speedsys results from 66MHz x 3, but that's about it. Could be a cache or ram issue at that point (15ns and 60ns respectively). However, I don't want to say that's definitely the issue because my PCI bus is all out-of-whack with only a 2/3 divider.

numeriK wrote:

Are there any socket3 boards that have a 1/2 FSB to PCI divider?

The 1/2 PCI multiplier was the most common for 486 boards; perhaps it was easier to implement than 2/3.

Well my 8433UUD board is dead (RTC), so I swapped it's BIOS with the M919. MUCH better BIOS! (I forget how awesome it was).

So as of now I'm using the M919 w/ the 8433 AWARD BIOS, not the AMI. It's actually quite stable and feels faster... preliminary benchies!

AM5x86 @ 60Mhz bus, 3x multi, ~187Mhz, L2 cache disabled (purposely), 64MB EDO @ 0/0WS, system is surprizingly stable (as of now)

Speedsys 4.78 score: 67.07
Memory Bandwidth: 181.14 MB/sec
L1 cache: 131.85 MB/sec
Memory Throughput: 75.05 MB/sec
3Dbench: 100.0fps
Quake timedemo2 = 19.0fps

Booted up at 66Mhz bus, 3x multi, ~205Mhz, L2 cache disabled (purposely), 64MB EDO @ 1/0WS... these are the highest scores I've ever received on this AMD chip:

Speedsys 4.78 score: 74.50
Memory Bandwidth: 200.80 MB/sec
L1 cache: 146.37 MB/sec
Memory Throughput: 83.26 MB/sec
3Dbench: 111.1fps
Quake timedemo2 = crashed while loading... boo!... CPU isn't stable enough at ~200Mhz+

You'd be better off replacing the RTC in your Biostar MB-8433UUD board. It is not difficult and the new RTC will last 15+ years. Using the wrong BIOS in another motherboard often appears OK at first glance, but funny issues start creeping in on you.

To establish stability in any 486 system, I like to go through the rounds of installing Win98SE w/IE6 and the unofficial service pack using a SCSI or other PCI bus mastering hard disk controller card. Run at least 3DMark99Max, 3DWinbench97, and Quake2 in software mode. If that is successful, install WinNT 4.0 on another partition using NT's default boot loader. Install all service updates, including IE6. Run Winbench96/99 and WinTune98. Play an mp3 for 2 hours continually. If that goes well, try for a Windows 2000 installation using the defaut boot loader again. Same drill of tests. You can also try CPUburn or CPUstab. Stay away from the onboard IDE controller because PIO mode really bogs the system down.

My MB-8433UUD will pass all this abuse and this system has been on 24/7 for many years. The M919 will likely fail at one of these tests. v3.4 of the M919 has PCI SCSI bus mastering problems in W2K and occassionally in Win98. I'm not sure if it has PCI IDE bus mastering issues as well, but this would be a good test.

Many 486 boards will show signs of weekness or fault with long term stability tests. A lot of these faults can be attributed to BIOS settings, at other times, not. For my interests, the benchmark results are not of interest until the system is demonstrated stable.

If you can get your v1.5 m919 working at 3x60 with all the noted tests, I'd be content with that. If you eventually install more cache, remember that some boards need to have their L2 cache speeds reduced (40 MHz+), while others do not. Depending on your system requirements, setting up a fast, solid 486 system can be a real challenge. Enjoy the process!

For my interests, the benchmark results are not of interest until the system is demonstrated stable.

While I can respect where you're coming from, I disagree. A benchmark is exactly that. Many of the best 3DMark systems in the world are only stable long enough to run 3DMark, usually due to thermal issues/lack of enough liquid nitrogen. So, according to you my results are not valid?

What I'm testing here is the limits of a 486. Benchmark results prove these limits can be achieved, for a short period of time or not. I'm not looking for a 200Mhz+ 486 that's 100% 24/7 stable - to be quite honest that's most likely impossible - but who truely knows? Thus the point of this thread.

I'm playing with 60-66Mhz bus speeds and ~205Mhz CPU speeds, and I'm getting complete runs through various benchmarks, yet they're irrelevant?

Like I said, I can appreciate where you're coming from but that simply isn't the point of this thread. If I want a 100% stable system I'll go back to factory defaults... 😵