Some performance question about DirectX
Author
Hello,
I use some 3d performance testing program to test my video card, and it told me the video card can draw properly 13.7 million polygon per second. However, in directX, I can never achieve such rate. I know that adding texture affect the frame rate. But can someone suggest me that with using texture, how many frame rate I should expected if I am using a Geforce2 MX video card? Or where can I get these kinds of information. Most of the video card had been tested in Tom''s hardware is rated by number of polygon it can draw per second, it this mean that I can write a directX program and then achieve such result?
Thanks!
Nachi
There is no real answer to your question, Nachi. No matter what the base hardware or PC spec. the framerate can vary massively. Through bad programming a game can run poorly on the most powerful and up-to-date PC's whilst, with a little bit of good coding practice and optimisation, a game can maintain a decent framerate on a more modestly specced PC. There are far too many other things to list that can influence a game's framerate (such as how geometry is rendered by the game engine).
As far as benchmarking and performance figures are concerned always treat them with a little scepticism. The staggeringly high figures issued by graphics hardware manufacturers are often in no way related to 'real-life' or 'in-game' situations. Of course you have benchmarks along the lines of 3DMark (which aim to simulate in-game situations via a range of performance tests), but even the credibility of such software is thrown into doubt with the recent accusations of cheating by a certain hardware manufacturer and their drivers...
Regards,
Sharky
---
#define _WEBSITE
Sharky's Coding Corner
#endif
---
[edited by - Sharky on July 3, 2003 3:40:58 AM]
As far as benchmarking and performance figures are concerned always treat them with a little scepticism. The staggeringly high figures issued by graphics hardware manufacturers are often in no way related to 'real-life' or 'in-game' situations. Of course you have benchmarks along the lines of 3DMark (which aim to simulate in-game situations via a range of performance tests), but even the credibility of such software is thrown into doubt with the recent accusations of cheating by a certain hardware manufacturer and their drivers...
Regards,
Sharky
---
#define _WEBSITE
Sharky's Coding Corner
#endif
---
[edited by - Sharky on July 3, 2003 3:40:58 AM]
Yep, what Sharky said. 
The figures stated on the sides of graphics card boxes are often the maximums for the raw chip, not even connected to a real PC.
The "real world" maximums stated by manufacturers come from programs like BenMark5 (available to download from the nVidia website) which supply perfect data to the card: long, untextured, indexed static triangle strips so they still really only demonstrate the practical maximum the graphics chip is capable of without showing the effect of game style textures, uploads of data over the AGP bus, your application doing other things (such as AI, sound, physics) etc.
A reasonable rule of thumb to use is you''ll only achieve about half of the stated "real world" maximum in a reasonably well written REAL real world app. If you''re lucky you''ll get up to 70%, but I''d never expect more.
--
Simon O''Connor
ex -Creative Asylum
Programmer &
Microsoft MVP
The figures stated on the sides of graphics card boxes are often the maximums for the raw chip, not even connected to a real PC.
The "real world" maximums stated by manufacturers come from programs like BenMark5 (available to download from the nVidia website) which supply perfect data to the card: long, untextured, indexed static triangle strips so they still really only demonstrate the practical maximum the graphics chip is capable of without showing the effect of game style textures, uploads of data over the AGP bus, your application doing other things (such as AI, sound, physics) etc.
A reasonable rule of thumb to use is you''ll only achieve about half of the stated "real world" maximum in a reasonably well written REAL real world app. If you''re lucky you''ll get up to 70%, but I''d never expect more.
--
Simon O''Connor
ex -Creative Asylum
Programmer &
Microsoft MVP
>> The figures stated on the sides of graphics card boxes
>> are often the maximums for the raw chip, not even
>> connected to a real PC.
I''d go far in my criticism of these peak performance numbers, but I wouldn''t go that far. At least not today (5 years ago was another story, and at least not with the major graphics cards.
The peak performance numbers on the side of the box are typically based on rendering very long, very coherent and collocated vertex data stored in static VBs in local video memory, where the triangle size and all the rendering states are optimal. The overall platform configuration, process and thread priority, and what other programs are (not) running, are also factors in these numbers. You also have to take into account whether the numbers are rated in submitted or rendered vertices.
Long story short, these numbers typically have a lot of fine print defining how they were taken, but they are peak achievable to within about 15% or so.
Unfortunately, a normal game has to deal with running its own non-similar workload on many different non-idealized platforms, and there are other apps running as well.
Development Lead, DirectX
Windows Graphics & Gaming Technology
>> are often the maximums for the raw chip, not even
>> connected to a real PC.
I''d go far in my criticism of these peak performance numbers, but I wouldn''t go that far. At least not today (5 years ago was another story
, and at least not with the major graphics cards.The peak performance numbers on the side of the box are typically based on rendering very long, very coherent and collocated vertex data stored in static VBs in local video memory, where the triangle size and all the rendering states are optimal. The overall platform configuration, process and thread priority, and what other programs are (not) running, are also factors in these numbers. You also have to take into account whether the numbers are rated in submitted or rendered vertices.
Long story short, these numbers typically have a lot of fine print defining how they were taken, but they are peak achievable to within about 15% or so.
Unfortunately, a normal game has to deal with running its own non-similar workload on many different non-idealized platforms, and there are other apps running as well.
Development Lead, DirectX
Windows Graphics & Gaming Technology
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
