2D, Acceleration, And Windows: Aren't All Graphics Cards Equal?Tom2D Conclusions, Preview Of Part 2Based on our own hands-on analysis of the current situation, we have to observe that the ATI's Radeon HD 5000-series cards are really struggling with 2D graphics. It’s also somewhat embarrassing that an older on-board graphics chipset is not only faster in a several areas (against both ATI- and Nvidia-based discrete cards), but also that there’s no real workaround for dealing with vector-based programs, either. This is not just a deficiency being measured in our testing; it’s also readily discernable to those who work with 2D graphics on a daily basis. Frankly, it’s quite difficult to imagine how an older Radeon HD 4870 can come so close to matching or beating the newer card in so many tests.
While 2D acceleration (including 2.5D layering) functions well, ATI has not yet managed to implement a number of pure GDI functions in its Radeon HD 5000-series cards. With a number of driver revisions behind us since Windows 7's initial launch, this situation is difficult to comprehend for the folks spending hundreds of dollars on next-gen hardware and running into trouble in 2D apps. We must also point out that this applies not only to our synthetic 2D benchmark, but also to various other real-world programs we used for our testing, including AutoCAD, as well as Corel Draw, Adobe Illustrator, Photoshop CS3/CS4, Microsoft Publisher, PowerPoint, and more. This calls for urgent and dramatic improvements, especially because our test results for Vista demonstrate significantly better scores than those in Windows 7 (we'll get into more depth on this in Part 2).
In preparation for and during testing, we participated in several calls with Antal Tungler, technical PR manager at AMD, to discuss the 2D performance of the Radeon HD 5000s under Windows 7. After we proved to ourselves that these performance issues stemmed purely from GDI problems, which nearly every program must handle in creating windows for the desktop, this pronouncement no longer seemed as appropriate, especially for small offices and home users. This is doubly true in light of our discovery that a competitor’s on-board GPU could handle 2D graphics more deftly.
We can only speculate (and hope) that the Catalyst drivers are the where the issue and probable solution lies. If this is true, ATI should be able to remedy things relatively easily. Given that its latest budget and consumer grade graphics cards were recently introduced, general 2D slowdowns across the whole product line seem probable. What troubles us most, given the results of our testing, is that rectangles receive genuine acceleration, while all the other geometric primitives (especially lines and curves) do not. By contrast, an extreme performance fall-off when GPU acceleration is used suggests that nothing good is going on here. Nvidia also falls off when rendering ellipses and polygons, and we’d like to figure out why this is happening as well.
For the time being, we recommend that users deactivate Aero when working with 2D graphics programs if they’re using one of the latest Radeon HD 5000-series cards in their PC. The resulting performance improvement of up to 300% that follows should make up for the lack of eye candy in those windows. That’s why we implore our readers to deactivate Aero for affected programs, so as to avoid turning off Aero altogether on their machines.
How-to: Deactivate Aero for a Specific Program
Right-click the program icon, then select the Properties item from the resulting pop-up menu. Click the Compatibility tab in the properties window, then select the setting checkbox that reads "Disable visual themes" (or similar language).
DWM usage gets deactivated when visual themes are disabledDWM usage gets deactivated when visual themes are disabled
What’s NextIn Part 2, we’ll test the reflexes and capabilities of our ATI cards thoroughly once again and compare the results with direct and indirect competitors from Nvidia. In the meantime, we’ll also circle back with ATI and Nvidia regarding the results of our initial testing, and wait in rapt suspense for their responses.
After intense conversations with our peers and colleagues (and raging rants on our forums) we want to exercise this benchmark more thoroughly and also put it through its paces on various hardware running XP, Vista, and Windows 7. Our aim is to develop a more comprehensive understanding of the 2D graphics capabilities for all of these systems. We plan to look at older S3 cards, Voodoo, numerous older GeForce-based models, and a full range of ATI cards. In short, we plan to leave nothing out, including even integrated graphics chipsets from our large collection of motherboards.
We already know we will encounter one or more moments of illumination during this process, along with our share of disappointments. A more precise description of the benchmarks, ratings for test results, and a download link for the Tom2D benchmark software will also be included in Part 2, as well. We’re pretty sure that you’ll find all this material pretty interesting, especially as we discover which consumer-grade graphics cards deliver the best 2D graphics performance for specific Windows versions. Let us surprise you: stay tuned!
Update (1/26/2010): With preliminary research into our 2D performance analysis, AMD reports back with the following:
* Tom’s Hardware has tripped over a workload area (2D lines, etc.) that we have not optimized yet.
* Until this new benchmark, we have not seen any other applications that are bottlenecked by this path, and hence have not focused on it until now.
* Our initial analysis has shown that we have no hardware limitations in this area.
* We now have our driver team engaged to optimize this path and will release a new driver to address this workload as soon as possible.
* We have already found an easy way of increasing our performance greatly, and are now going to try and schedule this in a future Catalyst (need to code in production, validate, ensure it doesn’t break anything else, etc.).
http://www.tomshardware.co.uk/2d-window ... 95-11.html
quick, 
