cxm717 wrote:

I tried that demo on my K6-3 system and it just crashes the game. I tried with the Geforce card and the TNT2 with driver versions 2.08 and 12.41. I ended up finding an iso of the goty version of the game (I just googled it). I thought maybe there was something wrong with the patch I was using. It's same 1.1.0.8 that he used to record it but it still just crashes. I'll give it a go on a P3 system and see how it goes.

It played for me on my 486, but took what felt like 10 minutes to complete. Then when it finished, I didn't see an average frame rate displayed. Is there a console command I was supposed to type to see the a) instantaneous frame rate and b) average frame rate?

rasz_pl wrote:

auron wrote:

, plus the game is probably leveraging the mmx instructions to speed up a few things

nah, mmx was useless for games

half-life definitely has some sort of mmx detection built in that can also be overridden ("r_mmx"), i'm not sure what it speeds up exactly though. https://github.com/ValveSoftware/halflife/issues/550 seems to claim that the variable is only used for software rendering. what about the software audio dsp stuff in the game?

besides that of course there were a number of games from the day that had mmx settings, for instance jazz jackrabbit 2, and unreal where the readme (indirectly i guess) suggested that it's used for the audio engine. of course there's also stuff like rebel moon rising which required mmx.

but yeah, overall i would think a pentium pro would run games like half-life better than a pentium mmx.

I was able to install Half-Life version 1.0 and run the blowout timedemo at 800x600. My IBM 5x86c-133 system with a Voodoo2 scored 4.3 fps. If I recall, the same system with a Matrox G200 scored around 8 fps in Quake II.

Attached are the results.

anandtech has some benchmarks of the blowout timedemo:

https://www.anandtech.com/show/205/5 (rage 128)
https://www.anandtech.com/show/219/6 (tnt)

results aren't quite comparable between these two tests as for example their k6-200 with rage 128 scored 11ish, while the k6-233 with tnt did just 7.1; same motherboard, but they used a version with 2 instead of 1mb cache on the latter test. they didn't specify the used API for the latter test either.

pretty remarkable how the overclocked 300a offers perfect performance scaling.

I first played Half Life from start to end with a Pentium 133 32MB RAM Nvidia Riva TNT Sound Blaster 16, and it was totally playable.

That's cool but Riva TNT is waaaayyy more powerful card than VooDoo...

Back in the day I played through it on a Pentium 166 MMX, 32MB and though it did have an ATI Rage 3D, rendering was buggy in Direct3D (maybe it was it's short video memory, I think it was 2mb), I played it using software mode, at 512x384. I have that same hardware nowadays, might give it a try.

Riva TNT is still way more powerful than ATI Rage 😀 No compare.

Cyberdyne wrote:

I first played Half Life from start to end with a Pentium 133 32MB RAM Nvidia Riva TNT Sound Blaster 16, and it was totally playable.

What result do you get with the blowout timedemo?

Cyberdyne wrote:

I first played Half Life from start to end with a Pentium 133 32MB RAM Nvidia Riva TNT Sound Blaster 16, and it was totally playable.

totally playable probably meant 15fps at the time, just like I managed both Doom 1 and 2 on 386DX40 at ~10fps and it seemed 'fine' _until_ discovering gaming cafes with Quake and 3dfx cards and 60fps silky smooth Daytona USA Arcade twin cabinet at local burger king.

I tried a few different CPUs in my Asus P5A-B setup to see how they run half life. The setup was the same as before. Asus P5A-B, 128MB PC100@CL2, Rage pro AGP, sound blaster live and 2 12MB Voodoo2 cards in SLI.

I used half life 1.0.0.5, sound was enabled, high quality sound and eax were disabled.

Edit: Something else I thought might be interesting. I tried a few different 3Dfx miniGLs to see how much it changed performance. The one labeled AMD is the miniGL from the quake2 3Dnow patch made by AMD, to use it in half life you need to set gl_flipmatrix 1 at the console.

I played HL back in the day with a Pentium 133 with TNT2 M64 PCI.

When indoors or smaller maps I had no issues playing at 1024x768 (probably >15 fps), but as soon as you were outside or in a larger map the frame rate dropping significantly (~10fps).

I'm not sure how demanding the blowout timedemo is, but from the above bar chart, it looks like the P133 scored about 7.5 fps average. So I guess this is bar minimum for "playable"?