Sorry to disappoint you. You can do pretty much whatever you want with your computer and the game. High scores is the only thing that counts.

In my first post that started the thread I stated that no rules applies. High scores was the only thing that counts.
Ehh. One... maybe two rules. A k6 something cpu and a soundcard is required. 😀

Necrodude wrote on 2019-02-02, 11:18:

The rules are simple.Any K6, k6-2 ,k6-3 cpu, Full screen. Any gpu. Any OS, any resolution. So basically no rules except some so […]
Show full quote

The rules are simple.Any K6, k6-2 ,k6-3 cpu, Full screen. Any gpu. Any OS, any resolution.
So basically no rules except some sort of k6 cpu
High scores is all that counts.
Please post your tweaks 😀 I gonna put them to good use 😀

I also talked about my history with the Quake games my drive to squize out as much performance as possible.

Necrodude wrote on 2019-02-02, 11:18:

I am a Quake fan and have spent a lot of time getting out as much performance as possible out of quake2 and quake3.
Overclocking, cpu, gpu, registry hacks, bios settings, cfg files. My scores in quake2 and quake3 is right now as follows.

The following post (the second post in this thread) I posted my heavily tweaked config file.

I hope this does not discourage you from trying. I would find it awesome if someone found some more nice tweaks to speed up the games.

Edit:
I will as as possible post download links to my custom Quake3 Ultra low spec texture and low spec sound pack. A custom low spec mod.

Btw. My Tweaked custom Quake2 is kinda awesome. I have a socket 7 Pentium 233MMX overclocked to 266mhz. It has a nvidia Riva 128 PCI Graphic card, and 64 megs of ram.
With the vanilla unmodifed Quake2 i got 12 frames per second in timedemo 1, map demo1.dm2 timedemo.

With my config file and the low spec texture pack I get 33 frames per second. So it really did a wonder on that low spec computer.

Sorry, but I fail to see the usefulness of this thread. The data provided has no real value for anyone, and worst of all, it’s actually misleading some members of this forum, as it can be seen from this conversation here.

bloodem wrote on 2022-02-14, 20:35:

Sorry, but I fail to see the usefulness of this thread. The data provided has no real value for anyone, and worst of all, it’s actually misleading some members of this forum, as it can be seen from this conversation here.

Sorry but I fail to see the usefulness of your reply. If you arent interested in this thread, please keep it to yourself. I thought it was manditory to actually read the topic and the first post before complaining. If you actually would have read it it would be very clear what this thread is about.

I suppose your thread, your rules. But I’ve always thought there was a unspoken rule that when having a benchmark competition that one always used default settings as to make a fair battlefield.

From there it was up to the user to tweak drivers, overclock, change hardware.

But like I said, nothing wrong with your thread your rules 😀 have at it!

Sphere478 wrote on 2022-02-14, 21:59:

I suppose your thread, your rules. But I’ve always thought there was a unspoken rule that when having a benchmark competition that one always used default settings as to make a fair battlefield.

From there it was up to the user to tweak drivers, overclock, change hardware.

But like I said, nothing wrong with your thread your rules 😀 have at it!

Thank you!

I totally agree with you with your point that the default standard settings should be used by everyone when comparing systems. Tweaked cfg:s and other stuff cant be mixed into it. Thats plain dumb.
But the sole purpose of this thread is to see how fast we could go with Quake2 and Quake3 on a k6 systems. It is fun to tweak stuff to its maximum. 😀

Necrodude wrote on 2022-02-14, 22:06:

The texture mod does its job best on slower computers. Computers slower then k6-3:s. But it will give a few frames per second.

Dropping the JPEG quality doesn't reduce the amount of texture memory required nor does it grant more compression to textures to fit into memory. The image data will load up to the same amount as if it were unaltered and still mean nothing for slower CPUs. Same applies for downsampling sounds, as the engine will resample them to the mixing rate at load.

The big 'optimization' from the forced settings snuck into autoexec.cfg is just that r_vertexLight option, which usually save CPU by dropping all the shader stages, skip all the dynamic lighting calculations, and is generally only best useful for single TMU video cards/3d cards as a last ditch option however is more popularly used to cheat by disabling shadows in maps and to gain physics-exploiting framerates id software didn't foresee.

leileilol wrote on 2022-02-15, 04:31:

Necrodude wrote on 2022-02-14, 22:06:

The texture mod does its job best on slower computers. Computers slower then k6-3:s. But it will give a few frames per second.

Dropping the JPEG quality doesn't reduce the amount of texture memory required nor does it grant more compression to textures to fit into memory. The image data will load up to the same amount as if it were unaltered and still mean nothing for slower CPUs. Same applies for downsampling sounds, as the engine will resample them to the mixing rate at load.

The big 'optimization' from the forced settings snuck into autoexec.cfg is just that r_vertexLight option, which usually save CPU by dropping all the shader stages, skip all the dynamic lighting calculations, and is generally only best useful for single TMU video cards/3d cards as a last ditch option however is more popularly used to cheat by disabling shadows in maps and to gain physics-exploiting framerates id software didn't foresee.

First of all all kinds of downsampling uses cpu cycles. If it already is downsampled it doesnt.
The config files are set to a low khz rating.
Less data in files means less data to be processed by the cpu. Simple as that.

On my k6-3 @616mhz the Quake3 texture and sound mods does not do so much. It gives 2-3 extra frames per second. How could that be?? 2-3 extra frames per second. On slower systems it reduces stuttering, and drive swapping, it also reduces load times.

As mentioned in a prevoius post the quake2 texture mod did miracles on a Riva 128 PCI card. That card is slow and it has a low amount of memory. Its because of that the mod worked so great.

The greatest performance giver by a large margain is the config files. The Quake3 cfg doubles the fps. The Quake2 cfg gives around a 30% fps boost.

Rated second is the recompiled exe files. They can give a 0-10% performance boost. Usually 2-5%.

Rated last is the texture and sound packs. They usually dont do as much. They give a tiny boost.
They give the biggest boost on really slow systems. On my 233mmx @266 with a riva 128 the cfg, exe, and texture pack trippled the fps in quake2.

Believe me. I have done some serius testing. Try it yourself will ya.

Btw I will post links to the faster exe files this evening.

Necrodude wrote on 2022-02-15, 09:12:

Less data in files means less data to be processed by the cpu. Simple as that.

Yet your config enables compressed textures which in q3, slow down load times greatly as it tries to call the driver to convert them to DXT1 after the JPEGs get decoded to full RGB888 data... (Q3 has no direct compressed texture format loading at all)

Necrodude wrote on 2022-02-15, 09:12:

As mentioned in a prevoius post the quake2 texture mod did miracles on a Riva 128 PCI card. That card is slow and it has a low amount of memory. Its because of that the mod worked so great.

The secret there is "gl_ext_palettedtexture".and "gl_picmip 2". Any percieved speed gain from converting paletted WAL data to 1% quality JPEGs and back to paletted WAL data is BS. It doesn't take a mod to bump the texture slider down a couple notches and turn 8-bit textures on.

Necrodude wrote on 2022-02-15, 09:12:

Believe me. I have done some serius testing. Try it yourself will ya.

If I had a nickel...

leileilol wrote on 2022-02-15, 09:27:

Yet your config enables compressed textures which in q3, slow down load times greatly as it tries to call the driver to convert […]
Show full quote

Necrodude wrote on 2022-02-15, 09:12:

Less data in files means less data to be processed by the cpu. Simple as that.

Yet your config enables compressed textures which in q3, slow down load times greatly as it tries to call the driver to convert them to DXT1 after the JPEGs get decoded to full RGB888 data... (Q3 has no direct compressed texture format loading at all)

Necrodude wrote on 2022-02-15, 09:12:

As mentioned in a prevoius post the quake2 texture mod did miracles on a Riva 128 PCI card. That card is slow and it has a low amount of memory. Its because of that the mod worked so great.

The secret there is "gl_ext_palettedtexture".and "gl_picmip 2". Any percieved speed gain from converting paletted WAL data to 1% quality JPEGs and back to paletted WAL data is BS. It doesn't take a mod to bump the texture slider down a couple notches and turn 8-bit textures on.

Necrodude wrote on 2022-02-15, 09:12:

Believe me. I have done some serius testing. Try it yourself will ya.

If I had a nickel...

Seriously. You come here and complain. You dont want to try it out. And you say it does not work. Wtf is your problem?

I have runned litterly hundreds of benchmark runs. Tweaking and testing. Tweaking and testing. Trying this, trying that.

You come here and calling bullshit. Seriously if you arent gonna proof me wrong. Leave this thread alone.

Necrodude wrote on 2022-02-15, 10:35:

You come here and calling bullshit. Seriously if you arent gonna proof me wrong. Leave this thread alone.

Also your "patch" halves the texture resolution which you failed to mention so that's effectively bumping up the picmip value by one and skews results and texel parity.

The q3 engine already has a very fast way of loading textures smaller: it's r_simpleMipMaps. It just skips every few pixels to load a fraction of an image than resampling.

(If you wanted to SERIOUSLY optimize Q3 for slow CPUs, you'd start at programming a better QVM recompiler and a faster sound mixer (Q3 1.27+ will always do 22khz 16-bit mixing regardless of what sounds there are)... not this copyright infringing data modification snake oil)

leileilol wrote on 2022-02-15, 10:54:

Also your "patch" halves the texture resolution which you failed to mention so that's effectively bumping up the picmip value b […]
Show full quote

Necrodude wrote on 2022-02-15, 10:35:

You come here and calling bullshit. Seriously if you arent gonna proof me wrong. Leave this thread alone.

Also your "patch" halves the texture resolution which you failed to mention so that's effectively bumping up the picmip value by one and skews results and texel parity.

The q3 engine already has a very fast way of loading textures smaller: it's r_simpleMipMaps. It just skips every few pixels to load a fraction of an image than resampling.

(If you wanted to SERIOUSLY optimize Q3 for slow CPUs, you'd start at programming a better QVM recompiler and a faster sound mixer (Q3 1.27+ will always do 22khz 16-bit

Sorry dont know anything about programing. I was hoping someone here might be up for it.

I never claimed the Quake3 texture pack would do any miracles. The config file does the good stuff. So snake oil..... Naaah.
I claimed the texpack gave me, on my k6-3+ @616mhz a couple of extra frames per second. The many hours of work put into it didnt give much. But it did give something. It made it possible to exceed 100 frames per second on a socket 7 machine.

You can argue how much as you want that it is BS. See for yourself will ya. Try it on a similiar computer. If you get 1 extra frame per second compaired to the vanilla stuff, then it works regardless what you think and say.

However the q2 texpack made miracles on a slow machine.

1-2 fps is well within standard deviation, essentially the results don’t mean anything.

TrashPanda wrote on 2022-02-15, 12:14:

1-2 fps is well within standard deviation, essentially the results don’t mean anything.

No. If you run the timedemo again and again and again and again and again and so on, and the results are consistent and repeatable time after time. Then its not.

leileilol wrote on 2022-02-15, 10:54:

The q3 engine already has a very fast way of loading textures smaller: it's r_simpleMipMaps. It just skips every few pixels to load a fraction of an image than resampling.

If you have any tips and pointers to the cfg that increases the fps, im eager to hear.

Or any other, oc, driver whatever. Im interested

Necrodude wrote on 2022-02-15, 12:18:

TrashPanda wrote on 2022-02-15, 12:14:

1-2 fps is well within standard deviation, essentially the results don’t mean anything.

No. If you run the timedemo again and again and again and again and again and so on, and the results are consistent and repeatable time after time. Then its not.

That's not how it works, if you could repeat it and give consistent results outside of standard deviation then yes you could say your results can be proven, the results would also have to be proven on more than just your machines and with standardised methods that do not change. eg same settings , hacks and mods for every machine running the test. 100s of tests per machine to build a good data set.

1-2 FPS means nothing here, come back when you can produce results outside of standard deviation on a repeatable basis.

There are people here who know exactly how the quake engines work to a level of detail that you dont seem to grasp and even they are not moved by your 1-2FPS gains because its within standard deviation and means nothing.

I'm actually surprised no one here has modded quake to simply output the entire time demo in wireframe mode, bet it would happily give you more than 100FPS and still be within your "No Rules"

TrashPanda wrote on 2022-02-15, 12:14:

1-2 FPS means nothing here, come back when you can produce results outside of standard deviation on a repeatable basis.

Pfffh!! Oh really....... Exactly who do you think you are??? You know what? This isnt Harwad University. This is not a scientific study. There is no proof burden here. It is a thread about getting the max out of Quake 3 and Quake 2. You simply have to take my word for it. I have runned a atleast a couple of hundred benchmark runs to find the maximum speed. I run every run 5 times in a row. To see what happens. It is the same thing every time. The second test is the fastest. The first has stutters to loading stuff. The third, fourth, and fifth just gets slower. When you do run the same thing over and over again, you start to see patterns. Every testrun of five runs follows the same pattern. I have compared the texture mod VS the vanilla textures a gazilliion times. It is faster. Believe me!. Try it yourself. Prove me wrong.
But hey!! I bet you even dont own a K6-3 machine, and you will never even try to prove me wrong.

TrashPanda wrote on 2022-02-15, 12:14:

There are people here who know exactly how the quake engines work to a level of detail that you dont seem to grasp and even they are not moved by your 1-2FPS gains because its within standard deviation and means nothing.

Its obvious that you are only here to mock me! You know what? I dont know much about the Quake 3 engine. And I really dont care either. I have never never ever claimed to be a Quake3 engine Technician. I do everything by trial and error. If it gets faster, then it is faster.

So if you are not going to contribute anything to this thread. Dont post in it! Mocking me and calling me a liar.
Thats so rude. I dont know how I offended you and the other guy. Must have touched your holy grail or something.

The sole purpose of this thread is to get as many FPS:s as possible out of Quake2 and Quake3 on a K6 system. By Any means.
So instead of harrasing me and calling me a liar. Stay on topic! And come up with some great new way to improve the performance of Quake2 and Quake3

Edit:
Btw. Today I did some testing today on my modern machine since the K6-3+ is severely cpu limitied and acts like a huge bottleneck to the graphics card.

TrashPanda wrote on 2022-02-15, 12:14:

1-2 FPS means nothing here, come back when you can produce results outside of standard deviation on a repeatable basis.

I just did. I dont think that you can argue that a 52 frames per second difference is inside the standard deviation margain.

Here is the config file and the custom recompiled exe used in testing above.
Place the autoexec.cfg in the baseq3 folder

Edit: At 4.4ghz the game reached its maxium frame rate. So it was impossible to see what really happend after you reached 1000 frames per second mark.
So I ran the same series of test again. This time at 3.0 ghz. This time it was a 200 fps difference between vanilla Q3 with the performance config and Texture/sound mod Q3 with the performance config.

Yepp! A difference of 200 frames per second. Someone earlier talked about snake oil...
It really depends on your system and what combination of config files, dll:s, exe:s you are using. Anyway. We wont see performance gains like that on K6-3 and similar systems, due to cpu bottlenecking.

Yesterday I did some new benchmark on my:
K6-3+ 400@616mhz (112x5.5)
128mb PC133 SDRAM Cl2
Geforce 256 SDR stock speeds
Wpcred Tweaks + CTU loading with autostart

Lol wtf is going on on that white case hahah

Sphere478 wrote on 2022-03-22, 05:29:

Lol wtf is going on on that white case hahah

Ha ha ha.

Some friends thought it was funny to pimp it with a sticker. So did I

Edit:
Anyway

Here is a comparasion of vanilla Q3 and the tweaked one

Edit2:
Did some more WPCREDIT tweaks. Got another five FPS:s.
Edit3:
Changed a BIOS setting. PCI latency timer. Changed from 32 to 64. Got another five Fps:s