fixed some of the pics. Imageshack isn't very reliable....

One thing I thought was interesting about these results is how the K5 @ 133 actually beats the WinChip2 @ 240 for CPU integer performance in Sandra. K5's architecture (aside from FPU) is quite impressive, but it also caused major problems for clock scaling.. It is one of the fastest x86 CPUs from a performance per-clock perspective (ignoring SIMD).

I've tried overclocking it but it is only able to handle about 15 more MHz. And it doesn't have many multiplier options so it's hard to up the bus speed and keep it stable.

Winchip 2 is also a very poor overclocker and only has whole number multipliers.

If it's not too much to ask, can you toss DOSQuake START.BSP 320x200 timerefresh detault settings results in there too?

I thought about doing a Quake timedemo but I decided not to because when you bring the video card into the equation for DOS tests it becomes tricky to get results that aren't skewed by video performance.

Also, Quake was basically written entirely around the Pentium architecture so I really wonder if results for it are valid for anything other than Quake itself. Although that's not to say that Quake results wouldn't be interesting, but I think we can predict how that will go just by remembering the olden days of people bitching about how badly their K5/K6/6x86/486 ran the game compared to Pentium/PPro. 😀

Interesting and true i bet a lot of benchmarks and reviews on hardware these days are a bit skewed because of the gfx cars but, they always do say what they setup is that was used for such benchmarks. it seems to me the mmx233 and the pro 200 have a strong point and one the alu and the other the fpu i don't know what either of those are but, i can tell one be better an one task than the other?

Nice thread, I like it 😀

Which version of SiSoft Sandra are you using?

But no Pentium Pro bench results? 😳 😁

I just set up my ASUS P5A and ran all these different CPUs thru it. Was pretty simple to do. PPro doesn't work very well in Socket 7. 😉

Sandra 2002 is the one I used.

I noticed that I interchanged the Pentium MMX 166 and 233s in the graphs above. I actually meant to keep the reference CPUs identical. I didn't notice that I had gone with 233 for one and 166 for the other.

swaaye wrote:

I just set up my ASUS P5A and ran all these different CPUs thru it. Was pretty simple to do. PPro doesn't work very well in Socket 7. 😉 Sandra 2002 has results for a PPro 200 anyway.

Sandra 2002 is the one I used.

hehe, well...if you set your mind up it'll work 😉

I also discovered that the AMD K5 has the same write allocate feature as K6 and that the board wasn't enabling it. I found a utility that will enable it in Windows (made by Powerleap). It only gained about 20 points in the CPU arithmetic integer test in Sandra. In DOS, I tried the AMD enwa.exe util which said it recognized and enabled the WA feature, but the speedsys results didn't change at all. Curious.

Added Pentium Pro results to first post.

An interesting thing I noticed was that this CPU really drops its power usage with HLT active. I had my Killawatt power meter hooked up and saw system power usage go from 80W booting to ~35W in Windows!

Also, I doubt these tests really show you what that huge L2 cache does for this CPU. Most synthetic tests aren't going to be designed to show the benefits of a cache this big. It certainly doesn't seem to be hurting performance though. I was curious about that because I've been wondering if the cache latency might be worse with this compared to the other PPros.

Amigaz wrote:

But no Pentium Pro bench results? 😳 😁
...
hehe, well...if you set your mind up it'll work 😉

You've won the prize, Amigaz! 😉

I am very disappointed in all of you.

I was hoping all of this testing nonsense would bring about an amazing roundtable discussion of mid-90s CPUs!!!!!! What happened huh? .... Oh I know... I missed the Cyrix tests that would be the catalyst for a flame war. Sorry Cyrix sux too much for anything besides what they designed their CPUs for (MS Office). And lets not forget the occasional incompatibilities that 6x86 had. 😉

Stop just stop. You guys can't tempt me into buying a Cyrix MII. It's a lost cause. Seriously.

Hey wait, I have a Cyrix 5x86. Too bad it's got nothing on either the Am5x86 @ 160 or a PODP @ 100. Oh wellz. Maybe if they hadn't released it partially broken with some of the fancy features disabled (cuz they were borked). 😵

(awaits argument)

I almost bought a 6x86 when they first came out. I'm pretty glad I didn't. I guess at that time the lousy FPU wouldn't have been such a big deal, but I seem to recall they had serious heat problems (corrected with the 6x86L), and used non standard FSBs that ran the PCI bus way out of spec. Using the 75 and 83MHz FSB was a pretty sleazy method of increasing performance (at the cost of stability). I thought the 6x86 logo was pretty neat though.

swaaye wrote:

I am very disappointed in all of you. {snip!} […]
Show full quote

I am very disappointed in all of you. {snip!}

Stop just stop. You guys can't tempt me into buying a Cyrix MII. It's a lost cause. Seriously.

Hey wait, I have a Cyrix 5x86. Too bad it's got nothing on either the Am5x86 @ 160 or a PODP @ 100. Oh wellz. Maybe if they hadn't released it partially broken with some of the fancy features disabled (cuz they were borked). 😵

(awaits argument)

Actually, I'll GIVE you an MII if it will be used for something! I once bid on a mix of various Socket 7 processors that no one else seemed to want, and got the lot for very little, including the MII.

Amazingly enough, compared to far worse luck thereafter on others for that socket, none of those in that batch had bent pins!

I thought I'd throw four more (either rare or laughable) in since swaaye decided to lay down the Gauntlet. 😉

The Socket-3's were benchmarked in a FIC-PIO-3 motherboard. The two Socket-7's were benchmarked in a FIC PA-2013 motherboard.

I've thrown in a:
Cyrix 5x86-120 (40MHz x 3) (Write-back enabled)
AMD-X5 (P75) 5x86-133 (33MHz * 4) (Write-back enabled)
IBM 6x86 PR-166+ 133MHz (66MHz * 2)
Cyrix MII-300GP 233MHz (66MHz * 3.5)

On the Socket-3 side: The Cyrix 5x86 edges out the AMD-X5. But it's clear an X5 at 160 would best the Cyrix. The Cyrix 5x86 isn't very overclockable and I doubt would catch up to the AMD-X5. That said, stock vs. stock, the Cyrix is the winner.

On the Socket-7 side: Both of the Cyrix designed CPU's are pretty uninteresting and their results speak for themselves. I had a heck of a time getting the MII-300 to not lock up in speedsys. It finally completed a full test after disabling the Cyrix CPUID feature in the bios. Unfortunately, disabling the CPUID made speedsys not realize the MII had MMX registers. So the MII graphs don't show any MMX memory benchmarks. 😠 Damn buggy cyrix's. Anyway, enjoy.

heh. The Winchip 240 is beating out that MII 300. That's embarassing. 😀 It's been said around the web that the Cyrix 6x86 may have simply been designed to win the popular benchmarks of the time. Apparently its FPU is even worse than the K5's.

As for the Cx5x86, I've too found that it will outperform a AMD 5x86 @ 133. Its FPU is actually rather significantly better according to Sandra. But when you put the AMD chip at 160 then the AMD chip gets the advantage in integer performance. Cyrix still has a small lead for FPU though (PODP FPU is better though). And yeah the Cyrix 120 wouldn't overclock at all for me.

Also, some of the Cyrix 5x86 chips can have extra features enabled too. There are DOS utilities out there to do this. The problem is that most of the CPUs released are bugged with regards to their fancy branch prediction and will crash Windows. They disabled the features partly cuz of hardware bugs and partly because BIOS support was needed for them. It might be worth the time to check those out if you haven't. You can eek a little more from the CPU, depending on what works.

Oh btw, you should try out the PNG image format. For screenshots like these it'll compress much better than JPG and it's lossless. I noticed that the colors are washed out in the JPGs.

Yeah that AM5x86 score's a bit low. Fastest I was able to get it was 160MHz, very low memory wait states (1?) which leads to the SST result of 59.79.

swaaye wrote:

some of the Cyrix 5x86 chips can have extra features enabled too. There are DOS utilities out there to do this. ... You can eek a little more from the CPU, depending on what works.

Yeah, I have played with the performance registers on the Cyrix 5x86. In the end, I allowed the BIOS defaults to stand for the benchmarks. When I enabled branch prediction, it gained just a single point on the CPU scale, so I didn't bother with it. It wouldn't have been that relevant for speedsys anyway.

Added Pentium II "Klamath" 233 MHz.

I accidentally saved the wrong Sandra Multimedia test so I only have the non-MMX result here. That's kinda ok though because it's a better comparison to the PPro 233 this way.

Swaaye, I noticed in your O.P. that you wanted to throw in a P55C @ 300 MHz for your tests, but no longer had a suitable one (eg 233). I have just tested a 233 OC'd to 300, and DL'd Sandra 2002, and performed the CPU and Multimedia tests on it. Now, please bear with me because I wanted to do this fairly quickly, and my "internet PC" hasn't got a working floppy drive ATM, so I couldn't get the screenies uploaded. So I simply wrote them down instead, and here they are -

Speedsys overall score = 227.31 (Very similar to 5u3's score of 225.13)
Speedsys memory scores (the first 3 listed) = 553.95, 278.13, 192.00 (Almost identical to 5u3's scores)

Sandra CPU score = 638, 343 (These values seem quite good)
Sandra Multimedia score = 906, 238 (Yowsers! If you want Integer ALU stuff done quickly, look no further!)

(5u3's speedsys pic can be seen in this thread here)