LLVMpipe vs. OpenSWR Software Rendering On A 40 Core / 80 Thread Tyan Server

[darkbasic]

Please add Ryzen, Epyc, Threadripper and also Intel i7 to the comparison.

[devius]

Ah, good old times when we had Quake 1 running on a 486 in 320x200. It was still playable. And pure SW rendering.

Well, if you consider 10fps playable. And that is on a AMD 5x86 133MHz, the top of the line 486. On a more common DX2 66MHz or DX4 100MHz the framerate is even lower. Quake was really meant as a Pentium game, and even on a Pentium 133MHz it only averages about 27fps (which was considered playable back then). I'm not making up these numbers BTW. I tested Quake on these systems a few months ago.

As for the article I think the only thing missing is some kind of low-end GPU to compare these numbers to. Intel HD graphics would be fine.

[blacknova]

Quake was really meant as a Pentium game, and even on a Pentium 133MHz it only averages about 27fps.

True, though Quake was (relatively) playable even on 486 dx2-66, to run it smoothly at least at 320x200 Pentium 60 were needed.
WRT article, I think testing llvmpipe or swr with games have no or little meaning. Both renderes are too general to be really optimized for realtime graphics.

[wdb974]

Wasn't something like Half Life 1 the last game that was offering pure SW rendering (okay, maybe also some implementations of DooM source ports like zdoom)?

I remember changing the settings to full software rendering in UT '99 because of an OpenGL issue. I think the culprit was the ATi Xpert @Play (or something approaching), even though the game should've worked on the Voodoo 2.

It totally worked in software mode, despite not being as smooth and looking like pixel porridge.

[sdack]

Worst, most interesting benchmark ever.

[eydee]

This server should be used for hosting Phoronix, so it doesn't crash every time AMD releases something.

[DMJC]

What's the raytracing performance like on this sort of hardware? I mean it's an old technique, surely the performance gains since the early 1990s have been vast.

[devius]

True, though Quake was (relatively) playable even on 486 dx2-66,

Right... if you consider 6fps playable.

...to run it smoothly at least at 320x200 Pentium 60 were needed.

Better, but you're still looking at about 14fps only.

[arjan_intel]

What's the raytracing performance like on this sort of hardware? I mean it's an old technique, surely the performance gains since the early 1990s have been vast.

the number of pixels (kind of square with resolution) has grown a lot too in that time....

[coder]

Quake was really meant as a Pentium game, and even on a Pentium 133MHz it only averages about 27fps (which was considered playable back then). I'm not making up these numbers BTW. I tested Quake on these systems a few months ago.

At what resolution? I had a Pentium 75 and I'm sure my average framerate was somewhere in the 20's. I'd have been using something around 320x200 (or 320x240, since it was co-developed by Michael Abrash, who discovered/popularized that mode).

Anyway, all this talk of Quake is irrelevant, given how much more sophisticated graphics APIs and game engines have become. But it was impressive what they managed to achieve, back in the day.