VOGONS


First post, by sheath

User metadata
Rank Newbie
Rank
Newbie

Hello,

I have been reading all of the benchmark and comparison threads searching for some sort of standard solution for benchmarking Windows 98, SGL, 3DFX, Direct X, Open GL games and hardware. I am currently stalled with Rage software's Expendable. I see benchmark charts out there, but can't find the timedemo command line for Expendable. Does anyone care to spell it out for me?

I'd really like to put all of the benchmark hints and tricks into a single thread if it hasn't been done already. I've seen Phil's DOS benchmarking project, which I will definitely be participating in also. I love that his project runs from a simple batch file, can this be replicated for Win98 games? Quake II and III are easy to benchmark with their timedemos. Others don't leave log files so I'm debating how to record thier performance in such a way that I don't miss the outliers or average the frame rates wrong.

This writer references some software "that saves the frame rate for each frame of a demo, so that you can analyze it later." I don't even think FRAPS does this in a reasonable and easy to parse way.
http://www.thg.ru/graphic/19981007/index-01.html

In addition to something like that, which I think I could painstakingly replicate with a random screenshot using FRAPS with the counter visible, I have one more bit of info I'd like to eventually collect for each game. If any game developer has boasted how many polygons per frame their game averaged, or peaked at I'd love to catalog each source and what their claims are. For example, various sites online claim that Quake III Arena is averaging 10,000 polygons per frame, but others claim 15,000 polygons per frame. By contrast the Dreamcast's peak polygons per second was supposedly 3.5 million polygons per second, which would be 116,000 polygons per second at 30FPS. Yes, I know, polygons per second is just marketing performance, but the comparison is very interesting to me and I am looking for as standard that can be applied across all engines that can be recorded in some way.

Another interesting benchmark idea harkens from the 2D generations, in simply counting animated objects on screen and framerate to determine some measure of performance. This would definitely make a distinction between 3D FPS with only one or two enemies on screen, and those with dozens for example. It would break down with destructable environments however. Then there is multi-pass rendering effects, which the PS2 apparently excels at. How would one measure something that is only visible in motion? And so, unless this fine group can help, I will continue lurking on 90s benchmark reviews and prior threads until I run out of ways to search.

Thank you for your time and consideration.

Reply 1 of 8, by Putas

User metadata
Rank Oldbie
Rank
Oldbie

Run Expendable with -timedemo parameter.

I was looking for that THG tool, does not seem like it was ever published.

Quake 3 has a parameter to show triangles per frame, and other statistics, no need to speculate there.
Note that games are diverse instances of engines and do not represent polys/frame performance of engines.

Reply 2 of 8, by collector

User metadata
Rank l33t
Rank
l33t

This forum is for Windows games on modern systems, as in Very Old Games On New Systems. Ask old hardware and old OS questions in Marvin: Marvin, the Paranoid Android

The Sierra Help Pages -- New Sierra Game Installers -- Sierra Game Patches -- New Non-Sierra Game Installers

Reply 4 of 8, by sheath

User metadata
Rank Newbie
Rank
Newbie

THG tool is more info than I had before. I was thinking I could have FRAPS take random screenshots for framerate measurements and I could go by standard polycounts per frame, or root out interviews about each individual game for any stats I could find. Thanks for the tip for expendable, I saw a previous thread that the go.exe is the program to run I just didn't figure I had to run it from a command prompt with the -timedemo parameter. Now I'm getting 70FPS average (45-105 Min-Max) with the Matrox Bump Mapping patch on my Kyro II system, but my Banshee just crashes back to the desktop with the default install. The Banshee doesn't like running the Incoming gameindex file sometimes too, but I've already benchmarked it several times successfully. Now to install Expendable on the Geforce 256 with Dec 1999 drivers to see how it runs. 😉

On the polys per second benchmarking methods. I wonder if it'd even be interesting to take a min-max-avg poly count per object for each engine or game and count the maximum number of objects on screen. I think this model probably falls apart when particles, missile trails, explosions and the like are involved as they cannot easily be counted even in Quake II and definitely hit the performance.

-edit-

Thank you for moving it to the correct forum.

Reply 5 of 8, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

Pretty sure the thg guy is just talking about regular fps, not per-frame timing. If you want more resolution, grab a capture card and some software to analyze the output. Like an overlay that color-codes frames and so on.

Reply 6 of 8, by sheath

User metadata
Rank Newbie
Rank
Newbie
vvbee wrote:

Pretty sure the thg guy is just talking about regular fps, not per-frame timing. If you want more resolution, grab a capture card and some software to analyze the output. Like an overlay that color-codes frames and so on.

I've been using Premiere Pro CS 5.5 for years to edit capture footage and haven't stumbled on this solution. I'll see if there is a unique frame counter. All this time I've just been frame advancing and counting frames. That's no good though when the video's 30FPS though.

Reply 7 of 8, by Putas

User metadata
Rank Oldbie
Rank
Oldbie
sheath wrote:

On the polys per second benchmarking methods. I wonder if it'd even be interesting to take a min-max-avg poly count per object for each engine or game and count the maximum number of objects on screen.

I don't follow. What object? Who cares about maximum number of objects?
I can imagine putting same model or scene into different games and measure they performance, but even that is not so interesting to me.

Reply 8 of 8, by sheath

User metadata
Rank Newbie
Rank
Newbie
Putas wrote:
sheath wrote:

On the polys per second benchmarking methods. I wonder if it'd even be interesting to take a min-max-avg poly count per object for each engine or game and count the maximum number of objects on screen.

I don't follow. What object? Who cares about maximum number of objects?
I can imagine putting same model or scene into different games and measure they performance, but even that is not so interesting to me.

I'm thinking specifically about sourceable or visible facts about the game, or specific levels, characters and objects in one video. Something like what 3DMark help did.
https://web.archive.org/web/20050404173017/ht … tests.htm#game1