Reserved for Clock per Clock tests!

You have PII and PIII, isn't it also more fair to also throw in the 500MHz K6-2 as well?

leileilol wrote:

You have PII and PIII, isn't it also more fair to also throw in the 500MHz K6-2 as well?

I will run K6-2 tests as well, but it will take a few days for me to find some more time. I also plan to test the K6-III @ 500Mhz with the Diamond Monster Fusion.

Could You possibly add SuperPI 1M results, I uploaded a AMD XP Barton running at 500mhz over in the SuperPI thread.

AMD looks really good. I wish you could include game tests besides Quake 1/2 and 3DMark here, though, from '99 or so. As I understand it, AMD doesn't do as well across the board.

Imperious wrote:

Could You possibly add SuperPI 1M results, I uploaded a AMD XP Barton running at 500mhz over in the SuperPI thread.

I tried running superPI 1.5 and it crashes at Loop2 (this program has performed an illegal operation). SuperPi 1.1 works - it completes 1MB in 5 mins 27 secs.

firage wrote:

AMD looks really good. I wish you could include game tests besides Quake 1/2 and 3DMark here, though, from '99 or so. As I understand it, AMD doesn't do as well across the board.

It does great in everything I play - unreal, nfs porsche, dungeon keeper 2, homeworld, etc. If you have any game benchmark suggestions, I'm willing to do the work.

I've tested the same systems (almost) some time ago. 3dmark is great for 3dnow! and so is quake. try running any other real game and the p2/p3 is MUCH faster. I love my k6-3+ with p5a but it just doesn't hold a candle to the pentiums in games.

For example: i did the benchmark of the racing game breakneck 2 (aka n.i.c.e 2) and there my k6-3+@600 could not even fully keep up with a p2-350,a 450 smoked it bigtime.
I did the test with every voodoo (1,2,banshee,3, 4 and 5) and tnt2 to geforce4ti. the outcome was always the same, pentium was much faster.

With some later games my k6-3+@600 struggled to get playable frames with a voodoo3 where a p3-600 simply shined with the same voodoo3. Only with a TNL capable geforce these games would run fine on the k6-3+.

When you run every test of everest you'll find that in the 2 most important fpu tests the k6-3+ will loose big to a p2. Fpu julia is great but not so important for games.

However, in integer tests it is pretty strong and for the time it wasn't a bad cpu. Price/performance was great for the platform. If 3dnow! was better supported (as in quake) it would have been a whole different story for games. I like it anyway and i have a lot of retro systems but the k6-3 is my favorite.

I did not use 3DNOW! enhancements for GL_Quake or Quake 2 - the above results are w/o any optimizations. The only benchmark that I ran using 3DNOW! optimizations is 3dmark99 (it does so by default). GL_Quake does not support 3dnow(as far as I know), and Quake 2 needs a patch for it witch isn't voodoo banshee friendly (or very glide friendly for that matter).

I'll add NICE2 to the benchmark list, as well as Unreal Gold. If you remember other games you remember where the K6-III performs poorly, please post them and I'll add them to the benchmark list.

Overall I disagree with you. I won't believe there is a game where the K6-III lags behind that much until I see it with my own eyes. Even Homeword or Dungeon Keeper II witch are more demanding then Quake 2 or NICE2 run flawlessly on the K6-III, and there isn't much of a performance difference between it and a similarly clocked P3. As far as I've tested the only K6 class CPU that truly has a very slow FPU and performs that bad in games is the original K6 (I have it in 200 and 266Mhz flavours). The're beaten by the pentium mmx 233, the Cyrix MII PR300 and the K6-2 at 266Mhz quite badly.

One thing is surely true - above 500MHz the K6-2 / K6-III don't see much of a performance increase, while the P3 does. Frankly the difference is so small there's no point in stressing my CPU by running it at that speed.

I never was much of a quake fan, so i don't know. Maybe quake doesn't rely as strong on fpu then?

And i didn't say it can't run games nicely because it can and a whole bunch of them as well, p2 is simply quite a bit faster in most of them.

I will be surprised if you can find a game apart from quake where a k6-3 comes even close to the p2/p3 (at the same clockspeed). It is well known that an overclocked k6-3+ is as good in games as a p2-333/ p2-350 on average.

BTW since both 3dnow! and SSE are heavily used in 3dmark99 it is normal for your p2 to score much lower there.

Maybe you can use fraps in some games? It is not a great tool but it gives you at least a way to measure fps in games like NFS etc. It will be lower than normal but that's also true for the pentium.

Make sure you use 640x480 to keep the focus on the cpu.

kanecvr wrote:

It does great in everything I play - unreal, nfs porsche, dungeon keeper 2, homeworld, etc. If you have any game benchmark suggestions, I'm willing to do the work.

There absolutely is a reason Q2 tests were all people talked about back then. A big bunch of cool 3D games to pick from, but benchmarking functionality is pretty rare. Those ones are interesting, if the testing methodology can be figured out; there's also good stuff like MechWarrior 3, Rogue Spear, Aliens vs. Predator, SWAT 3, Descent 3, FreeSpace 2, Flight Unlimited III, etc.

GLquake results look suspiciously like vsync is enabled.

3DMark99 CPU marks overstate the value of SIMD to say the least.

swaaye wrote:

GLquake results look suspiciously like vsync is enabled.

3DMark99 CPU marks overstate the value of SIMD to say the least.

I made sure vsync is disabled from the driver's control panel. When I first saw the results I tough vsync was enabled as well, but "Synchronize screen update to refresh rate" was unchecked in the driver's openGL, glide and direct3d tabs. There might be something wrong tough - I remember getting over 60 fps with a Pentium MMX 233 + voodoo 2... Then again the V2 is faster in GL_Quake and Quake 2 due to much faster multi-texturing.

firage wrote:

kanecvr wrote:

It does great in everything I play - unreal, nfs porsche, dungeon keeper 2, homeworld, etc. If you have any game benchmark suggestions, I'm willing to do the work.

There absolutely is a reason Q2 tests were all people talked about back then. A big bunch of cool 3D games to pick from, but benchmarking functionality is pretty rare. Those ones are interesting, if the testing methodology can be figured out; there's also good stuff like MechWarrior 3, Rogue Spear, Aliens vs. Predator, SWAT 3, Descent 3, FreeSpace 2, Flight Unlimited III, etc.

I can't remember - do any of those have built in benchmarks? If they do, I'll add them to the list.

meljor wrote:

I never was much of a quake fan, so i don't know. Maybe quake doesn't rely as strong on fpu then?

And i didn't say it can't run games nicely because it can and a whole bunch of them as well, p2 is simply quite a bit faster in most of them.

I will be surprised if you can find a game apart from quake where a k6-3 comes even close to the p2/p3 (at the same clockspeed). It is well known that an overclocked k6-3+ is as good in games as a p2-333/ p2-350 on average.

Not the games I play. Not to mention it seems lots of games on my list have both SSE and 3DNOW enhancements (Homeworld and Descent 3 to name a couple) witch explains why the K6 performs so well. And it's not about FPU speed either since the K6-3 beats down the pII in most synthetic FPU tests I've ran. In some the k6 scores almost twice what the PII manages.

Some games were written for the Pentium's pipeline - for example GL_Quake and software quake will run faster on a 233MHz pentium MMX then they will on a 233Mhz pentium II - you can google it (P55C's advantage in these games caused quite some noise when the P2 launched too) - so older games should perform better on a P1 and PII due to optimizations - but not so much on a P3 for some reason. The more the CPU's architecture deviates from the P54 / P55C's design, the bigger the performance hit. Quake 1 is just one example. Another would be DOOM witch doesn't even use a FPU - and as far as I recall it also runs faster on a Pentium 1 then it runs on a Pentium II with the same clock speed.

meljor wrote:

BTW since both 3dnow! and SSE are heavily used in 3dmark99 it is normal for your p2 to score much lower there.

That explains the P2's poor performance... 🙁

meljor wrote:

Maybe you can use fraps in some games? It is not a great tool but it gives you at least a way to measure fps in games like NFS etc. It will be lower than normal but that's also true for the pentium.

I'll give it a go, but CPU overhead for fraps is pretty high.

meljor wrote:

Make sure you use 640x480 to keep the focus on the cpu.

- yup - I ran all game tests at 640x480.

BreakNeck racing's built-in benchmark refuses to run - I get "Strecke nicht geladen !" in red letters on a white background and the game hangs. I ran the game in arcade mode and it's extremely smooth at 1024x768x16 with display correction, triple buffering and force lock enabled. I get way over 60fps - like the game is running too fast in parts. I'll post a video for you if I can't get the benchmark utility to run. I do like the game tough 😜

Yes, breackneck is a nice game. It is not that well known but i really like it as for the time it looked great and is pretty fast (and has a built-in bench).

I hope you get the benchmark working, would be nice to see how your systems perform with that one.

If you have them, try a banshee pci or v3 pci. I had problems with that benchmark on a msi ali V motherboard with agp cards ,voodoo's worked by selecting v2 support in the benchmark i believe but nvidia and matrox wouldn't run while the game itself worked fine?

I had no problems on a p5a which has the same chipset and i used the same drivers.... didn't test it on my AX59pro.

Comparision of CPUs should be done in software modes without 3D acceleration and not with different video cards. For example, Unreal has software mode.
As Pentium 3 coppermine would be interesting to use also as more common P3.

huh?
Running games in software rendering is kinda pointless - that's the reason 3D accelerators were invented.

I agree with havli - there's no point in testing in software mode since we all have 3d accelerators. Also I believe in using SSE and 3dnow! if the game / benchmark has it on by default (like 3dmark) - since the functionality is there, I find it hard to believe someone would deliberately turn it off.

I would have liked to use a coppermine in my tests but the slowest one I have is 700MHz (7x100), socket 370, and it's multiplier locked. In any case, I'm sure a 500MHz coppermine (if one could force it to run at that speed) would whip everything in my tests, including the katmai. I'll add a celeron 466 to the test suite as soon as I find more time for benchmarking. Coppermines are quite common these days, but I don't have one in slot 1 format and the're not easy to mind in these parts - most common slot 1 CPU's here are 450 to 600MHz katmai and 266-350MHz deschutes. In fact I've personally never encountered a slot 1 coppermine in the wild....

meljor wrote:

I hope you get the benchmark working, would be nice to see how your systems perform with that one.

I'll get it working eventually.

Thank you for posting this! Very good read. I appreciate the time that went into it.

kanecvr wrote:

I agree with havli - there's no point in testing in software mode since we all have 3d accelerators.

Then you compare not CPUs but subsystem CPU+3D card. In this case it's better to insert faster 3D card as slow card being a bottleneck may make the difference between CPUs less. Geforce 3 or 4 seem good for P3.

most common slot 1 CPU's here are 450 to 600MHz katmai

slotket and 370 CPU is an option

Tertz is right. Banshee is a bottleneck. I'd use a Voodoo5. Geforce cards are unlikely to work well on Super 7. Test everything at 640x480.