VOGONS

Search

Common searches

Why not to use 3dbench

Topic actions

First post, by feipoa

Posted on 
User metadata
Rank l33t++
Rank
l33t++

Perhaps some of you have run across this website before, but I just ran across it. http://www.complang.tuwien.ac.at/misc/doombench.html

Here is the excerpt.

3dbench is a popular benchmark, because it does something quite similar to the most hardware-hungry applications of many users (games).

However, it has several shortcomings: The most serious one is that it seems to use a low-resolution clock, resulting in quantization errors. For machines based on the 586-120, this error is on the order of +/-6% (i.e., two machines with the same result may differ in performance on this benchmark by 12%). Second, on fast processors it stresses video performance too much (at least for the resolution it uses); i.e., who cares if a game plays at 60 or 70 fps (that's what 3dbench gets for, e.g., a 486-100 with a local bus graphics card). If you have frame rate problems on such a processor, the game is much more CPU-intensive than 3dbench. Third, it just uses filled polygons. Present-day games use texture mapping or Gouraud shading, and a benchmark that uses these techniques would be more realistic than 3dbench. Fourth, it is an 80286 (or even 8086) program; current game utilize the features of the 386 architecture in protected mode.

Therefore I propose using Doom as an alternative. Doom has no quantization errors that I know of, it is more CPU-intensive (although still too video-heavy on today's faster machines), it uses texture-mapping and it is a game for 386 protected mode. On the practical side, it is free of charge and it has a benchmark mode. You may find it interesting that Doom 1.9s is not Pentium-optimized.

Plan your life wisely, you'll be dead before you know it.

Reply 1 of 2, by elianda

Posted on 
User metadata
Rank l33t
Rank
l33t

I do not agree fully with this.

3DBench: It is 8086 code and likely uses the default timer which is not very precise (probably 55 ms). This leads to the decribed high error range at faster systems. Also 3DBench uses only 8 bit accesses to the VGA RAM.

Doom: Yes Doom is newer and uses 386 code. But it uses ModeX, which has different performance charcteristics as regular 320x200. So you bench something else here.

Generally: What do you want to measure? Pure Fillrate? Then you actually don't want a program that spends much time for game calculation. For example with Doom on a slow PC you spend 90% in the 3D calc and only 10% in the graphics draw, while on a fast PC it may be the other way around. So the final fps number is only influenced 10% by the graphics draw speed in the first case and 90% in the latter. This means it is only loose related to the fillrate.

I usually look at the values from vspeed which simply draws random values to the graphics card with 8, 16 and 32 bit accesses. This tells you much more about the pure fillrate and memory interface width than running a game.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 2 of 2, by Scali

Posted on 
User metadata
Rank l33t
Rank
l33t

3DBench should be used for the era of machines it was aimed at, which was 8088-386/486 at most.
Obviously if you want to measure a different kind of scenario, you use a different benchmark.

DOOM can't be used to benchmark anything lower than a 386, so it's not a replacement for 3DBench.
Besides, DOOM is a very specific use-case. It's not a true 3d engine like 3DBench is. It's a 2d raycaster, which was a popular technology in a very small time window (around 1992-1995), until it was replaced by true 3d engines with texturemapping such as Descent and Quake.
Which will give you different kinds of benchmark results again.

The same reason why new videocards are no longer benchmarked with Quake 3 Arena these days. Q3A was a good game and a good benchmark, but hardware and 3d technology have evolved a great deal since, and it is hardly representative for the capabilities of today's systems.

So I don't agree with this at all. I think whoever wrote this lacks basic insight in what benchmarks are, and what they're trying to measure, and how to use them properly.

http://scalibq.wordpress.com/just-keeping-it- … ro-programming/

Go to top of page Go to top of page
Back to General Old Hardware
Close window
Close window