Yes the GA-586HX is a bit of outlier here. It actually supports straight P233MMX (as well as K6 or MII) so no need to use Overdrive on it.

However, it gives just as big boost to the PODMT66x200 is to the P120 so I think the relative improvement will be the same.

It is just interesting that a mature chipset + tuning can boost the P120 by two CPU classes.

mpe wrote on 2020-03-30, 23:31:

And the winner is…. […]
Show full quote

And the winner is….

Pentium Overdrive 83

DSC_6010.jpeg

Despite only offering only a modest clock increase the all-new architecture makes all the difference when compared to the 486DX2-66. This unloved CPU really brings the most significant improvement. I expected it to rock the Quake test. Quake loves Pentium FPU and where Quake is compared no 486 can compete. 2.5x improvement is just massive. However, I was surprised it also won general-purpose Winstone 95 test - 1.6x improvement which is ahead of other overdrives.

The other overdrives were remarkably close to each other.
I'd put Socket 4 / Socket 5 on shared 2-3 spot by having in average 1.5x improvement
The Socket 7 is lagging slightly with 1.4x improvement

quake.png
winstone.png
It is not without surprise than in absolute terms P120 on 430HX chipsets beats both P133 and P150 on 430LX/NX. It just stress out how important choice of chipset was in the Pentium era.

Hello,

So I was very close, as my prediction was:
highest gain: Socket 5 (75 to 150)
then Socket 7 (120 to 200MMX)
then Socket 4 (60 to 133)
and lowest gain 486 to Pentium 83
😁

OK, but next time I will ask about the benchmarks that are going to be used.
I also knew very well, that using quake (was optimized for Pentium FPU) as benchmark, will show a huge gain for the POD 83MHz.

Really a nice competition, good job.

Did you have fans for these overdrives? If there's no fan installed, the processor is underclocked to protect itself.

pentiumspeed wrote on 2020-03-31, 18:32:

Did you have fans for these overdrives? If there's no fan installed, the processor is underclocked to protect itself.

Of course I did. They just look way too uniform with them so I prefer to take pictures with fans removed.

I think only the PODP83 has that feature (it disables multiplier without a fan).

I am doing just the opposite. I am starting with a Pentium and seeing how slow I can clock it down to.
Currently I am using this setup with an AMD K5-100mhz CPU.
By disabling all the caches on the CPU and Motherboard I have been able to slow it down to a 386@40mhz.
My goal is to take a Pentium CPU down to a 286@10mhz.

Which motherboards and chipsets do you think will work bests ?
And which bios type ?
And which settings ?

some asus socket 5 boards could do 40mhz FSB and i think even 20mhz with undocumented jumper settings.

mpe wrote on 2020-03-30, 23:31:

And the winner is…. […]
Show full quote

And the winner is….

Pentium Overdrive 83

DSC_6010.jpeg

Despite only offering only a modest clock increase the all-new architecture makes all the difference when compared to the 486DX2-66. This unloved CPU really brings the most significant improvement. I expected it to rock the Quake test. Quake loves Pentium FPU and where Quake is compared no 486 can compete. 2.5x improvement is just massive. However, I was surprised it also won general-purpose Winstone 95 test - 1.6x improvement which is ahead of other overdrives.

The other overdrives were remarkably close to each other.
I'd put Socket 4 / Socket 5 on shared 2-3 spot by having in average 1.5x improvement
The Socket 7 is lagging slightly with 1.4x improvement

quake.png
winstone.png
It is not without surprise than in absolute terms P120 on 430HX chipsets beats both P133 and P150 on 430LX/NX. It just stress out how important choice of chipset was in the Pentium era.

I was curious and installed Winstone 95 on my 2 486 machines.
I had my fun with this benchmark. I had to install a printer, without doing that, it crashed the program and the underlying Windows.

OK, but now I am a bit puzzled about your values:
Is that benchmark so much faster under Windows 3.1 than under Windows 9x?
You have got more than 100 for the DX/2 66.
I found a table of Notebook-tests, that were tested with Winstone 95 under Windows 95. The highest score was 118 with a notebook, that had a Pentium 90 CPU.

The Winstone 95 tests overall system performance. That means including storage.

The key to my high scores is a very high storage performance. I was using Adaptec 2940U PCI controller and scsi2sd v 6.0 device. This combo delivers roughly 8000 kB/s with negligible access time. Which is maximum possible on these systems.

Whatever contemporary systems used in 1995 was likely much slower. Especially if it was a laptop with a 4500rpm drive. I also had 32bit disk access and file access enabled which makes a big difference.

I think the coolest thing about your results is that the POD83 marginally outperforms the P60 and P75, showing that it's performing at true Pentium levels.

Did you happen to gather any other benchmark results? Doom would be interesting, not only because it's the industry-standard DOS game benchmark, but it also wouldn't be so FPU-skewed.

Intel486dx33 wrote on 2020-03-31, 19:53:

I am doing just the opposite. I am starting with a Pentium and seeing how slow I can clock it down to. Currently I am using this […]
Show full quote

I am doing just the opposite. I am starting with a Pentium and seeing how slow I can clock it down to.
Currently I am using this setup with an AMD K5-100mhz CPU.
By disabling all the caches on the CPU and Motherboard I have been able to slow it down to a 386@40mhz.
My goal is to take a Pentium CPU down to a 286@10mhz.

Which motherboards and chipsets do you think will work bests ?
And which bios type ?
And which settings ?

Some K5’s had an undocumented 1x multiplier, couple that with a 50mhz, 45mhz or even 40mhz bus setting found on some older motherboards and you may not even need to fudge Cache.

mpe wrote on 2020-04-05, 17:53:

The Winstone 95 tests overall system performance. That means including storage.

The key to my high scores is a very high storage performance. I was using Adaptec 2940U PCI controller and scsi2sd v 6.0 device. This combo delivers roughly 8000 kB/s with negligible access time. Which is maximum possible on these systems.

Whatever contemporary systems used in 1995 was likely much slower. Especially if it was a laptop with a 4500rpm drive. I also had 32bit disk access and file access enabled which makes a big difference.

Just for your INFO:
Micronics Board, Wide SCSI Disk from 1999
Compaq: Narrow SCSI Disk from 1997 (I think)
Re: The Ultimate 486 Benchmark Comparison

clueless1 wrote on 2020-04-05, 18:18:

I think the coolest thing about your results is that the POD83 marginally outperforms the P60 and P75, showing that it's performing at true Pentium levels.

Did you happen to gather any other benchmark results? Doom would be interesting, not only because it's the industry-standard DOS game benchmark, but it also wouldn't be so FPU-skewed.

Let me first comment on practicality of DOOM framerates. I know many like to use Doom for performance testing. However, I think it is actually a very bad benchmark once you enter Pentium or fast 486 performance levels. Reason being that the DOOM is famously doing only 16bit bus transfers when retiring frames to video RAM. Therefore faster systems with optimum transfers are penalised if they don’t handle 16 bit bus writes optimally. While it still scales with CPU performance somehow the simple engine is no task for a Pentium and a significant part of DOOM performance is lost on fast systems.

At the same time since the game framerate is capped at 35 which almost any fast 486 or Pentium system can achieve, it stops being really useful for benchmarking even as a real-world test. You are essentially measuring how fast a particular system can do suboptimal things...

I can demonstrate this on two 486 chipsets where one has almost two times faster memory/L2 and yet is beaten in DOOM realticks with the same CPU and VGA.

When using DOOM as benchmark I actually prefer to run DOOM in low-res mode (as in Phil's pack) with minimum window and low details. When running like this the effect of bus writes is diluted and it scales much nicely.

But yes, I captured DOOM scores in my test and results are as follows:

Basically PCI clock increase creates all the difference...

@mpe - thanks for posting those and your thoughts on why Doom isn't an ideal benchmark.

I wish there were more options for DOS gaming benchmarks. One game that I've played with a bit is Links LS 1997. It's really an amazing feat of programming when you consider it's DOS-native. It looks incredible and can play up to 1600x1200x32bpp if you have enough graphics card memory. It has a built-in benchmark that measures graphics and disk subsystem performance. Screen draws can be timed but it seems more CPU-dependent than graphics (I've not measured significant differences between slow and fast PCI graphics cards on a P200MMX). It might actually be a good candidate for your particular comparison (hand measuring screen redraws, not the built-in benchmarks). It's not something that's easily (at all?) obtainable digitally, I bought my copy off of ebay.

Thanks. I'll check the Links. I also plan to investigate some DOS chess games which have built-in benchmarks.

I actually like Quake as CPU/FPU/L2/MEM benchmark for what I do.

Quake doesn't depend that much on VGA card (in 320x200) as it is reasonably complex on Pentium even at low res. It scales well with both clock as well as FSB, L1/L2 size, L2/mem timings, everything.

You can argue it is FPU skewed. But the FPU in Pentium is really that much better when compared to 486 or K5/K6/5x86/MII, ... So it wins for a good reason and reflects superior hardware. Of course FPU is not that important for everyone. However, if you compare non-Pentiums or just Pentiums against each other Quake works very well.

I was thinking about modifying Quake source to render to off-screen buffer to make it even more useful.

Did you try the MMX Overdrive on the Plato or do you know whether it is running on the Plato? Physical it should but the BIOS needs the sufficient microcode patch.

The MMX overdrive works on the Plato mb. I only briefly tested it at 150/50 MHz, so it was rather slow, but it ran.

Pentium microcode cannot be patched by BIOS.

Yes, POD5V83 ran wonderfully on those late PCI486 boards with WB support. But the reality is almost anyone who owned such a board probably had at least a DX4 or probably a 5x86 already. So in my opinion it's not a realistic upgrade scenario.

Probably not the 5x86 (Nov 1995), but they could have had a DX4 (Mar/Oct 1994).

Obviously overdrives weren't meant to be upgrading high-end CPUs. However, when upgrading from DX4 the PODP83 would still win the competition with healthy 1.72x improvement. At least in Quake - not so much in Winstone or other non-FPU tasks (1.2x).

The PODP83 also performs quite well in older Socket 2 VL-Bus motherboards. Supposing it works which cannot be taken for granted.

Anonymous Coward wrote on 2020-06-23, 07:36:

Yes, POD5V83 ran wonderfully on those late PCI486 boards with WB support. But the reality is almost anyone who owned such a board probably had at least a DX4 or probably a 5x86 already. So in my opinion it's not a realistic upgrade scenario.

Maybe we misunderstood, the intel Plato is the Premiere II with Socket5 and Neptune chipset.

@mpe
Were there any problems (push F1 or whatever) while posting with the MMX Overdrive on Plato? Would be great if you could provide your BIOS version with flash tool.

The BIOS is 1.00.12.AX1T which suggests that my motherboard is from a Gateway computer.

The BIOS update can be downloaded here http://files.mpoli.fi/unpacked/hardware/rom/i … el/platord.exe/

I doubt BIOS can make a difference for CPU compatibility....