There are plenty of raw system testing and game tech benchmark results, but not much in the way of "This way, with this game" stuff. There might be some old HL write-ups about MMX vs 3DNow usage, but I honestly never really saw either one getting any real chatter in game magazines by the time Windows XP hit.

My understanding is that it was more latent potential than actual use because software needs to be crafted to target the instruction set.

Hobbyist Quake III Arena mods have since demonstrated how 3DNow! could have been used to deliver >100% FPS increases. It's a small step from Q3 to all Q3-based games, and AFAIK even that small step has not been taken so the latent potential is confirmed sat idle.

But, in the "Wintel" era, why would a software company that relies on sales intentionally alienate Intel and owners of Intel CPUs?

MattRocks wrote on 2026-04-14, 05:53:

My understanding is that it was more latent potential than actual use because software needs to be crafted to target the instruction set.

Hobbyist Quake III Arena mods have since demonstrated how 3DNow! could have been used to deliver >100% FPS increases. It's a small step from Q3 to all Q3-based games, and AFAIK even that small step has not been taken so the latent potential is confirmed sat idle.

But, in the "Wintel" era, why would a software company that relies on sales intentionally alienate Intel and owners of Intel CPUs?

Yep, if 3dnow! had been used properly then there could have been much higher impact on execution speed. In ideal cases well optimized 3dnow! code can run ~5/6x faster than well optimized FPU code on K6-x.
3 comparisons from here:
https://falcosoft.hu/mandelx_benchmark_results.php

K6-III 500 MHz -> FPU: 5.47 pixels/msec; -> 3dnow!: 30.81 pixels/msec ( 5.63x faster)
Pentium III 500 MHz -> FPU: 10.75 pixels/msec -> SSE: 29.95 pixels/msec (2.79x faster)
Athlon 500 MHz -> FPU: 15.87 pixels/msec; -> 3dnow!: 28.66 pixels/msec (1.80x faster)

In case of the K6-x the 3dnow! performance is roughly the same as the SSE performance in case of the Pentium III. But the Pentium III has/had a faster pipelined FPU (roughly 2x faster compared to K6-x FPU).
The 3dnow! and MMX execution units of K6-x were 'overly' optimized. Clock for clock the 3dnow! performance of the K6-x was even faster than later Athlon/Athlon64/Phenom CPUs from AMD but the Athlons/Phenoms had a much faster FPU (roughly 3x faster compared to K6-x FPU).
So for K6-x using well optimized 3dnow! code instead of optimized FPU code would have been much more important.

Quake2 3DNow with Non 3Dfx card and how to get small performance boost with your 3Dnow CPU!

1. Use VGA driver that has 3Dnow and OpenGL support (probably many DX6.0 drivers and newer ones)
2.Extract the AMD Quake 2 3DNow! patch to your Quake 2.
3.Start Quake 2. Under the video options menu, choose 3DNow! OpenGL as your rendering device (not OpenGL or 3Dnow! 3dfxGL).

Is there any non Quake2 based OpenGL games that use 3DNow and OpenGL?

I know there was Quake1 build with 3DNow support, that is lost by now?
Re: MMX/3DNow!/SSE usage

There is a lot of test of K6-2 and K6-3, but always with 3Dnow! on, so only other types of realm games results, would be something like make some magic constant by finding what is ratio to similar cpu.
We have very few games results, we can maybe assume that performance boost for 3Dnow! Quake engine games is huge, but not much else.

Otherwise if no tests result, someone would have to revisit them and make research which games are some possible options to 3Dnow! off, or have old unpatched version.. or im not sure if possible make some cpu driver, which could make cant cpu as is does not support of these instructions.. other tool which would be nice to have, is some tool, which would confirm that these instruction were / werent called, some *.exe analyzer, which would check if these instruction are included in the *.exe.

Grok claims this but it could be completely wrong, i no time / energy to backcheck it:

• Unreal launch parameter -nok6 , for disable 3Dnow.
  These should be have 3DNow! patches, but original release without 3Dnow:
• Half unOfficial 3Dnow patch
• Descent Free Space -
• NFL Blitz - has patch
• SiN - 1.0.5 patch
• Speed Boat Attack
• Subculture
• Tomb Raider III
• Uprising 2

Yeah there could be other performance improvements in patches, but probably difference would be small.

Falcosoft wrote on 2026-04-14, 07:37:

Yep, if 3dnow! had been used properly then there could have been much higher impact on execution speed. In ideal cases well opti […]
Show full quote

MattRocks wrote on 2026-04-14, 05:53:

My understanding is that it was more latent potential than actual use because software needs to be crafted to target the instruction set.

Hobbyist Quake III Arena mods have since demonstrated how 3DNow! could have been used to deliver >100% FPS increases. It's a small step from Q3 to all Q3-based games, and AFAIK even that small step has not been taken so the latent potential is confirmed sat idle.

But, in the "Wintel" era, why would a software company that relies on sales intentionally alienate Intel and owners of Intel CPUs?

Yep, if 3dnow! had been used properly then there could have been much higher impact on execution speed. In ideal cases well optimized 3dnow! code can run ~5/6x faster than well optimized FPU code on K6-x.
3 comparisons from here:
https://falcosoft.hu/mandelx_benchmark_results.php

K6-III 500 MHz -> FPU: 5.47 pixels/msec; -> 3dnow!: 30.81 pixels/msec ( 5.63x faster)
Pentium III 500 MHz -> FPU: 10.75 pixels/msec -> SSE: 29.95 pixels/msec (2.79x faster)
Athlon 500 MHz -> FPU: 15.87 pixels/msec; -> 3dnow!: 28.66 pixels/msec (1.80x faster)

In case of the K6-x the 3dnow! performance is roughly the same as the SSE performance in case of the Pentium III. But the Pentium III has/had a faster pipelined FPU (roughly 2x faster compared to K6-x FPU).
The 3dnow! and MMX execution units of K6-x were 'overly' optimized. Clock for clock the 3dnow! performance of the K6-x was even faster than later Athlon/Athlon64/Phenom CPUs from AMD but the Athlons/Phenoms had a much faster FPU (roughly 3x faster compared to K6-x FPU).
So for K6-x using well optimized 3dnow! code instead of optimized FPU code would have been much more important.

I asked the developers of Shogo Mobile Armor Division why 3dnow! wasn't implemented in their game and they claimed they got little to no support form AMD so they gave up... I have no clue if that is a true statement across the board but it surprised me that AMD didn't support more...

SHOGO got 3DNow! after release, but it was never finalized code, as the game was completely abandoned by the devs within a few months of being released, so the 3DNow! option is just experimental and applies only to the D3D renderer. At that time, everyone was trying to cram 3DNow! into the codebase of their games, but it just didn't work well until the XP days, and it was really only landing in non-game software by then.

Carrera wrote on 2026-04-15, 07:58:

...
I asked the developers of Shogo Mobile Armor Division why 3dnow! wasn't implemented in their game and they claimed they got little to no support form AMD so they gave up... I have no clue if that is a true statement across the board but it surprised me that AMD didn't support more...

I can tell you from my own experience that 3dnow! is/was one of the smallest and easiest extensions of the x86 ISA. It consists of only 21 new instructions that compared e.g. to the 57 new instructions of MMX or the 70 new instructions of SSE is rather few.
It is basically MMX for 32-bit floating point data. Since it uses the same register set as MMX it also uses the same move, bitwise, unpack etc. instructions as MMX. The arithmetical instructions also use very similar mnemonics:
padd ->pfadd; psub ->pfsub, pmul ->pfmul etc. After reading the rather short documentation you can start writing 3dnow! code within an hour (about 60 pages including the examples and instruction set tables):
https://www.amd.com/content/dam/amd/en/docume … ences/21928.pdf

So I do not really understand what kind of special support the developers of Shogo Mobile Armor Division expected from AMD.

3DNow! (introduced 1998) was AMD's first SIMD instruction set for floating-point multimedia tasks on the K6-2, sharing registers with MMX. 3DNow!+ (also called Enhanced 3DNow! or Extended 3DNow!) was introduced with the first-generation Athlon and Mobile K6-2+/K6-III+ processors, adding 5 new DSP instructions and 19 new MMX instructions for increased efficiency, often rivaling early SSE.

RetroGamer4Ever wrote on 2026-04-15, 08:26:

SHOGO got 3DNow! after release, but it was never finalized code, as the game was completely abandoned by the devs within a few months of being released, so the 3DNow! option is just experimental and applies only to the D3D renderer. At that time, everyone was trying to cram 3DNow! into the codebase of their games, but it just didn't work well until the XP days, and it was really only landing in non-game software by then.

I have never found a 3DNow! release for Shogo. As far as I can tell it was hyped through some news releases on their website but eventually abandoned.

Interesting thread. I got quite curious about 3DNow! a while back and tried all sorts of patches. I think you should also keep in mind that Hardware Vendors started adding 3DNow! support in their drivers at some point and in the case of 3Dfx it seems to have made a significant difference.

SHOGO never got a 3D Now! release because the experimental D3D renderer that used it, was added in the game updates. All the CD-ROM releases seem to be the originally shipped version of the game, even those released many years after the game's original CD-ROM release.

Okay, so what does this mean exactly? Are you suggesting that the D3D renderer was updated in, say, the final patch released for the game to include 3DNow! optimizations? I have never been able to locate any 3DNow! patch for the game, only some apocryphal mentions on their website.

See more here:
Shogo D3D 3DNow! renderer?

My Gawd... I feel like it's 1998 all over again... 😁

Carrera wrote on 2026-04-16, 07:07:

My Gawd... I feel like it's 1998 all over again... 😁

Did that moment bring back the hope that a fix will be found? 😉

I reckon Intel caught wind of the 3dnow.ren file and intervened (maybe just hired the testers, maybe threatened to pull a different project, ..). Industrial sabotage was routine and probably still is.

Apparently, the experimental renderer shipped on some later CD-ROM prints that shipped with a patched (2.2) build of the game, which was necessary to have it be playable on (then) current systems with the latest gaming APIs that were standardized by that point. I know that I used it on my AMD system, but I don't know if any of the copies that I currently have in my possession have the file. If anyone has a copy of the disc, checking the files will be the only way to find it. Anytime I come across a copy of the game at a thrift store, I grab it, so I have a few to go through. I just have to find where I put them in the pile of stuff I have in storage. On some of the disc prints, the game version is shown on the disc art. The Red and Black disc - original and Interplay - might not be the one you are looking for, while the B&W disc is the 2.2 version and has the numbering on it.