Interesting that the K6-2+ and 3+ used so much power, they were touted as being cooler and more power efficient than the non-plus variants if I remember correctly.

Those power consumption figures got to be total system power draw at the wall, right?

Yes, the power consumption is of the total system draw at the wall, I haven't the right tool to measure the consumption of the cpu only.
The power draw for the k6-3+ could be lower but the Asus p5a supports v2.0 as minimum voltage.
The k6-x+ cpus anyway are really cold, the only cpus I had issues because they were so hot after shutting down the system were the winchip 1 and the cyrix with higher voltage.

I had to say that difference between K6-2 and K6-3+ is so uninspiring that it's not worth it in most cases. And they all have 550-600Mhz limit.

Pentium MMX is (clock-f0r-clock) a beast, particularly in FPU. I'm disappointed at the Cyrix scores. They had a reputation for a much stronger ALU vs Pentium and Pentium MMX, but the difference in these results is very small, less than 5% vs Pentium (non-MMX) in Doom and only really shining in Winstone99 (but even there nowhere near enough to justify the "PR" numbers) and 3dBench. Also odd how the MII is actually slower than the 6x86MX in every single benchmark.

Not so surprised at the relatively high power draw from the K6Plus CPUs - not only are they being overvolted, but they also have far more transistors on them for all that cache. Tbh I'm pretty impressed that a CPU with 21M transistors (K6-3+) actually draws less power than an identically clocked CPU with 9M transistors (K6-2).

The Serpent Rider wrote on 2020-09-01, 08:13:

I had to say that difference between K6-2 and K6-3+ is so uninspiring that it's not worth it in most cases. And they all have 550-600Mhz limit.

Yes, that is surprising/disappointing. It looks like the 128kB of the 2+ gives a pretty consistent 5% improvement, the 3+ a 10% improvement, no more. The only real reason to go for the plus is that it clocks higher; it's a rare K6-2 that runs at 600MHz, but most K6-3+ will do so and might even go up to 700MHz (never managed 720MHz, but it certainly has been done).

Idt Winchip: I found the cpu being unstable I do not know if the mb doesn't fully support it or if my cpu is somewhat damaged, it throwed some random general protections errors, I was able to complete all the tests.

That's because you need 3.5v to run it stable, according to specs. They can work at default Pentium voltage, but it's not guaranteed.

dionb wrote on 2020-09-01, 08:14:

Pentium MMX is (clock-f0r-clock) a beast, particularly in FPU. I'm disappointed at the Cyrix scores. They had a reputation for a much stronger ALU vs Pentium and Pentium MMX, but the difference in these results is very small, less than 5% vs Pentium (non-MMX) in Doom and only really shining in Winstone99 (but even there nowhere near enough to justify the "PR" numbers) and 3dBench. Also odd how the MII is actually slower than the 6x86MX in every single benchmark.

About the different performance of the v2.2 counterpart, it has been noticed already in the past here (Subtle differences between Cyrix and IBM 6x86s), they probably had to change something to achieve a lower voltage.
Also it's weird (cheater?!) the result of the rise mp6 just on super pi... I wonder why... it's faster even compared to a k6-3+ at 600mhz!

The Serpent Rider wrote on 2020-09-01, 08:19:

Idt Winchip: I found the cpu being unstable I do not know if the mb doesn't fully support it or if my cpu is somewhat damaged, it throwed some random general protections errors, I was able to complete all the tests.

That's because you need 3.5v to run it stable, according to specs. They can work at default Pentium voltage, but it's not guaranteed.

I tried both voltages 3.3 and 3.52, it proved to be unstable at both voltages, I suppose something is wrong with that cpu, I also tested on another mb and had the same issues...

Nemo1985 wrote on 2020-09-01, 08:42:

[...]

Also it's weird (cheater?!) the result of the rise mp6 just on super pi... I wonder why... it's faster even compared to a k6-3+ at 600mhz!

The mP6 was a weird design with three MMX units and a heavily pipelined dual-issue FPU. Rise claimed that these features could lift performance above P2-levels, but IRL they didn't come close in general computing (see Quake scores...). Quite possibly SuperPi manages to do something that was uniquely suited to the quirks of this architecture.

dionb wrote on 2020-09-01, 09:24:

Nemo1985 wrote on 2020-09-01, 08:42:

[...]

Also it's weird (cheater?!) the result of the rise mp6 just on super pi... I wonder why... it's faster even compared to a k6-3+ at 600mhz!

The mP6 was a weird design with three MMX units and a heavily pipelined dual-issue FPU. Rise claimed that these features could lift performance above P2-levels, but IRL they didn't come close in general computing (see Quake scores...). Quite possibly SuperPi manages to do something that was uniquely suited to the quirks of this architecture.

That's an explanation, in fact superpi is the only benchmark where it gets the best results, quake\quake2 it's behind the k6-2...

The only real reason to go for the plus is that it clocks higher; it's a rare K6-2 that runs at 600MHz

K6-2 officially reached 550 Mhz, so 600 is not a significant problem to achieve on CXT. K6Plus more or less limited to the same clock limits due to integrated L2 cache. Although extreme overclocking limit is higher.

I'm going to do another roundup but at 250mhz, with a mvp3 motherboard with wider voltage support, if anyone is interested.

I have come concern about the chance to overclock some cpus, so I'd like your opinions:
The cyrix higher voltage i'm going to use is this one: https://cdn.cpu-world.com/CPUs/6x86/L_C ... 02.7V).jpg
I will need to overvolt it to V2,9, should be doable, maybe it will be unstable on more demanding tests.
I have now 2 idt winchip cpu rated for 225 mhz and the 200 counterpart, I hope one will be able to reach 250 mhz.
Since the winchip2 is rated for 200 mhz I hope it will be able to reach 250, I have a beefy heatsink ready for rumble.
Pentium mmx 233 should be able to reach 250 without issues, but I wonder if the vanilla pentium S will be able to do the same, the tillamook doesn't boot on that motherboard, unlucky.

For the other cpus there should be no issues since they all have higher standard clock.
As for video card I would stay with the pci v3 to have some continuity the first round up, but if you think it is more suitable I can swap to a geforce 2 mx400.

Just chiming in to say, you might want to patch Quake to 1.08 before doing the benchmarks.

The patch provides a small FPS boost compared to 1.06.

Joseph_Joestar wrote on 2021-01-03, 15:51:

Just chiming in to say, you might want to patch Quake to 1.08 before doing the benchmarks.

The patch provides a small FPS boost compared to 1.06.

That's something to take into consideration, thank you for the tip! Let's see what the other fellow vogonians think, is it more important to keep the continuity of the round up or use different version of the software?

Updating this comparison with more cpus, different hardware and different conditions, I consider this redux more realistic than the others, since the motherboard was able to support the vcore of the cpus while the Asus p5a was more limited.

A direct comparison at 200 mhz with 100 mhz bus was not possible, due to the 2x multiplier not being supported from K6-2 CXT and Winchip2a.
So this comparison is made at 66x3.

Motherboard: Aopen AX59 Pro 1mb cache
Ram: 128 mb with fastest timings
Videocard: 3dfx Voodoo 3 2000 PCI
Hdd: Compact Flash Agfaphoto 8gb 233x with DMA enabled

Cpus:
Rise MP6 iDragon
IBM 6x86MX v2.9 - Cyrix MII v2.2
Intel Pentium, Intel Pentium MMX and Tillamook
Amd K6 (v2.9 and v2.1), Amd K6-2, Amd K6-2 CXT, Amd K6-2+, Amd K6-3+
IDT Winchip, IDT Winchip 2, IDT Winchip 2A, IDT Winchip 2B (not fully tested due to probably cpu gone bad).

The performance between Pentium MMX and Tillamook was obviously the same (tested on dos), it is just interesting to check how the power consumption decreased.
I tested the k6-3+ at his rated v1.6 voltage but I also tested it at the lowest supported voltage from the mb (v1.3) without any issue.

That's the version at 100 mhz fsb.
Please notice the * on winchip 2a, since it wasn't available the 2x multiplier setting it has been tested at 2,33, hence the increased performance.

I had some testing issues, the intel pentium was unable to end the final reality benchmark, it was closing without any warning, I tried different cpus but same issues.
With the k6-2 non cxt version the 3dmark 2k refused to run complaining a missing file, again different cpus were tested without any solution.

I just post also this table just to give an idea of the performance difference between 66mhz of bus and 100.