- Can I compare results between Operating Systems (OSes)?
Direct comparison between OSes is strongly discouraged. Due to differences between the graphics systems, the tests performed on one OS may be slightly different from those on another OS. For example, on Windows all composite operations use bitmaps with premultiplied alpha channels, whereas the preferred test is full compositing as per the compositing equations (i.e., with a straight alpha channel). Since these operations are different, comparing them directly would lead to biased (incorrect) conclusions as to relative OS performance. These differences also affect the total scores. Likewise, one OS may not support certain operations, which will also affect the scores.
- Why do tests at lower screen modes have lower scores than tests at higher resolutions?
Graphics cards tend to achieve peak performance when rendering large regions (just look at the performance vs rectangle size graphs). Since large regions do not fit into smaller screen modes, they cannot perform the largest rectangle size tests. As a result their total scores are more highly affected by the smaller region tests, and they receive a lower score.
- So I cannot compare results with different screen modes?
If you wish to make accurate comparisons, then you should try to keep as many variables constant between the results being compared. This includes the screen mode (both resolution and the bit-depth). Otherwise you would not be comparing the same thing.
Comparing different screen modes with the same hardware and system would be a correct comparison.
- Couldn't you compensate for the screen mode differences somehow?
No. Graphics performance is affected by many variables, and any attempts to compensate for one variable could invariably accentuate or bias another. More importantly, other factors could invalidate the compensation. For example, while the scores are weighted by the bit-depth (which affects the VRAM bandwidth required for operations), since the performance is not determined solely by the VRAM bandwidth, it is still a bad idea to compare results obtained with a 16-bit screen mode with those obtained with a 32-bit screen mode.
- So how should I compare graphics cards then?
Try to find results that have as many variables the same as possible. Ideally everything except the graphics card would be the same. It is important that the screen mode's bit depth be identical, and the resolution should be similar if not the same. This will yield the most accurate comparisons.
Rather than trying to hide the differences, GfxBench2D collects all of the specifications that are needed to make sure that comparisons are fair and accurate. As a result, there is a great wealth of information available to those who look for it.
NOTE: The GfxBench server is still a work in progress. I am planning enhancements which will make it easier to compare graphics cards. Even once those are ready, nothing beats making sure that you are comparing "apples with apples, and oranges with oranges."
- Why is there no version for <<insert OS of choice>>?
Simply because no version has been written yet. It is hoped to support as many OSes as possible. However, time and hardware availability limits what is possible. If you are a developer and would be interested in helping create a version for your favourite OS, send me an email, or post your interest on the GfxBench2D forum.
Windows Specific Questions
- Why is the frame-rate so low on Windows Vista and Windows 7?
Don't worry, the results are completely accurate regardless of the frame-rate. With Windows Vista and 7, GDI has been deprecated in favour of Direct2D. Unlike GDI, Direct2D does not allow rendering direct to the screen's front buffer. As a result, all of the rendering is done to a back buffer which is not visible. Since swapping the buffers takes time, this is only done after each test is finished. On the up side this makes the display much less flickery; on the down side, it can make the test seem to take longer due to lack of visual feedback.
- Why are the Copy to VRAM and Copy from VRAM results 0?
Since Windows does not support locking bitmaps in VRAM for CPU access, these tests cannot be performed.
- Why are the CompositeSrcMask results 0 on Windows XP?
Windows XP does not support Direct2D, and GDI cannot do masked compositing (alpha blending).
- I'm having trouble uploading results.
This is not really a question, but is still important. It seems that very occasionally firewalls (e.g., Comodo) can interfere with the uploading, without showing the usual "Do you wish to allow this application to use the internet?" prompt. If this happens, you may need to look at your firewall settings. I cannot give any guidance at this time, since the reports received have been rather sketchy, and nothing conclusive is known.
- Where do the test names come from? Some of the test names seem a little unusual.
GfxBench2D started out as a personal testing tool for my AmigaOS RadeonHD driver project. The terminology is based on what is used there.
Benchmark » GfxBench2D » F.A.Q.