Do New Drivers Really Boost Performance?

[No1sFanboy]

Overall I liked the article. It tones down some of the hyperbole you see in the forums where driver improvements are overstated. I recall when the 2900xt came out some forum members were preaching; wait for better drivers. The fact is ATI/AMD is glad to see that card behind them.
Still there are cases where drivers can have a big impact and it would have been nice to see this covered as well. The article paints a very decisive conclusion with a very small driver sample. The example Ape gave or something like that that follows a card from release to a year later would give a more accurate idea of how drivers impact performance.

I also liked the inclusion of CPU scaling. Some of the most demanding games combined with the top GPU setups make this more relevant than in the past. I'd welcome reading more material on this topic with as suggested quads vs. duals and even higher clocks.

Still this was a good piece that in my mind makes amends for some of the weaker stuff we've seen here. It doesn't excuse releasing new material on old drivers but it does speak to the relevance. Thanks for giving me a reason to stop at the front page on the way to the forums.

 

[Relayer]

I have a 3870. The last stable driver version for me was 8.7. 8.8 crashed Gmax (Virtually the same as 3D Max 4. I use it as a file converter for game models.) 8.9 crashed Cinema 4D (my 3D app. of choice) when I use it's "enhanced open GL" feature. Nice to see that I'm probably not losing any performance. In fairness to ATI, I submitted a bug report for Gmax and Cat 8.9 doesn't crash Gmax anymore.

 

[Codesmith]

What I got out of the article is that an "if it ain't broke don't fix it" attitude toward graphics drivers updates will save a lot of time and effort without costing you much in terms of performance.

 

[Guest]

anyone with software development experience will be able to tell you guys that software performance don't magically improve with each software version. it takes effort to fine tune software performance for large complex software and even more large effort to produce significant performance improvement by redesigning performance critical codes.
the most that this article could claim is only for a certain set of drivers. unless you thoroughly analyze release notes of each version and know what's the difference between each software version; this article doesn't do much except debunking the myth that newer versions will automagically improve performance.

 

[TeraMedia]

What if instead of testing drivers across a 3-release span, THG tested the current drivers (at least as current as is feasible given time-to-test, time-to-write, and time-to-publish) for an array of cards against the drivers released with the cards? Obviously the 46xx and 48xx series are just out, and the GTX series hasn't been out much longer. But compare e.g. the current drivers for an X1950 vs. the original, or current for the HD2900 vs. the original, and any performance enhancements should be much more apparent. By choosing drivers that are so close together in time, the article limits the amount of time that AMD and NV were able to spend on driver performance enhancements. The only problem with this approach is when the earliest or current drivers aren't stable - in which case the oldest and/or newest stable drivers could be employed instead.

I think the article does a good job of demonstrating that there can be a performance difference between driver releases. But I also think that the results are peppered with bug-fix noise and minimalized by the lack of development work that is incorporated into the driver enhancements.

 

[neiroatopelcc]

[citation][nom]TeraMedia[/nom]What if instead of testing drivers across a 3-release span, THG tested the current drivers (at least as current as is feasible given time-to-test, time-to-write, and time-to-publish) for an array of cards against the drivers released with the cards? Obviously the 46xx and 48xx series are just out, and the GTX series hasn't been out much longer. But compare e.g. the current drivers for an X1950 vs. the original, or current for the HD2900 vs. the original, and any performance enhancements should be much more apparent. By choosing drivers that are so close together in time, the article limits the amount of time that AMD and NV were able to spend on driver performance enhancements. The only problem with this approach is when the earliest or current drivers aren't stable - in which case the oldest and/or newest stable drivers could be employed instead.I think the article does a good job of demonstrating that there can be a performance difference between driver releases. But I also think that the results are peppered with bug-fix noise and minimalized by the lack of development work that is incorporated into the driver enhancements.[/citation]
They'd have to compare with the first catalyst featuring the card, and one a few releases later, and then the latest one I believe. The first release will never really be the fastest, so the newest is bound to be faster in almost any conceivable situation, making benchmarking it a bit pointless when you know the outcome.

 

[KrisInFla]

Great article guys! For me it provided the added benefit of helping to properly match CPU performance to GPU utilization for the build I'm configuring. It was a subject for which I actively scoured the net looking for data, usually drawing blanks or finding only garbled junk. Thanks tons!