Paper numbers are the numbers you'd see in a "normal" review. 25FPS at XXX by XXX resolution for example. Strange was talking about the way Hard reviews, where they talk about the experience. Sometimes they will mention that although X card was on paper faster, there is no real way to tell the difference between the two unless you could look in the case, or see the benchmark results.
My problem with synthetic benchies is that they don't model real world usage. Take the previously mentioned 2900XT. 3DMark said it should compete really well, but in actual games it could barely take on the 8800GS. Even the 8800GT was faster. Synthetic harddrive programs have the same issue. They model what the drive is theoretically capable of, but they fail to take into account real world usage. (not all clusters needing to be read are next to each other for example.) I've seen to many times where a synthetic benchmark says X should be faster, but real world programs say Y really is.
The only time when I use them is to stress test X device. If an overclocked video card can handle many loops of 3DMark (or a new to me used video card.) then I can say the overclock is stable.
Again, saying/guessing that X will be Y% faster then Z gives us an idea as to Xs relative performance. Reading Rescawen's post gives me an idea as to how fast s/he thinks it will be. I do not understand why Strange says its "shite"
Synthetics are fine as long as you are using the same hardware every where else in the rig and are just comparing the performance of one component.
My problem with synthetic benchies is that they don't model real world usage. Take the previously mentioned 2900XT. 3DMark said it should compete really well, but in actual games it could barely take on the 8800GS. Even the 8800GT was faster. Synthetic harddrive programs have the same issue. They model what the drive is theoretically capable of, but they fail to take into account real world usage. (not all clusters needing to be read are next to each other for example.) I've seen to many times where a synthetic benchmark says X should be faster, but real world programs say Y really is.
The only time when I use them is to stress test X device. If an overclocked video card can handle many loops of 3DMark (or a new to me used video card.) then I can say the overclock is stable.
Again, saying/guessing that X will be Y% faster then Z gives us an idea as to Xs relative performance. Reading Rescawen's post gives me an idea as to how fast s/he thinks it will be. I do not understand why Strange says its "shite"