AMD confirms Crysis AF bug in Catalyst Hotfix
During the tests for our Geforce GTX 285 review, we notice an irregularity related to AMD's Catalyst Hotfix ("Beta 9.1”). The driver doesn't display anisotropic filtering in Crysis Warhead.
There is a reason why we try to use the latest official WHQL drivers for our graphics cards reviews. But for brand-new cards this is impossible because there usually aren't final drivers available for the tests. The official drivers are fully tested and thus are usually free of bugs. Beta versions on the other hand have not been validated and thus they can contain failures - this also applies to Hotfix versions, even if they are officially available for download.
Catalyst Hotfix: The problem
At the launch of the Geforce GTX 295 AMD delivered a fresh Catalyst Hotfix. We, as we had just tested all graphics cards with the Catalyst 8.12 WHQL, released the review with the results of the final Catalyst 8.12 (version 8.561.0). But in order to be up-to-date, we also delivered results with the Geforce 181.20 and the Catalyst Hotfix (version 8.561.3) at that next day. At the end of the brief comparison we - like most reviewers on the web - announced an impressive performance benefit in Direct3D10 games. Crysis Warhead for example wondrously ran up to 80 percent faster.
Only recently we noticed that the Catalyst Hotfix doesn't apply anisotropic filtering to Warhead, although the related option in the Catalyst Control Center is set to 16x. Therefore we ran more tests and were able to identify the core facts. First of all: With the Catalyst 8.12 WHQL everything worked fine (AF is active). Second: The problem of the Hotfix only affects Direct3D10 - the discipline which was optimized and revealed the biggest benefits. Our tests revealed that the Hotfix displays AF correctly at Direct3D9. We validated our observations on a second test system with Windows Vista x64 SP1.
Geforce-Karten stellen mit dem 181.20 sowohl unter D3D9 als auch D3D10 AF dar.
The Hotfix increases the Direct3D10 performance in some games, but given the results in Crysis Warhead, one notices that it was assembled in a hurry. Far Cry 2 and Stalker: Clear Sky on the other hand don't seem to be affected.
AMD confirms the bug
AMD confirmed that they were able to reproduce the bug and that a fix will be delivered as soon as possible. The problem seems to be limited to Direct3D10 only. In the meantime posts in diverse forums claimed that the driver (8.561.3) also runs on Widows 7 and reveals the same bug there.
When and in what form, a solution will be available, cannot be said up to now. We estimate that there might be a "Hotfix for the Hotfix” or that AMD solves the problem with the Catalyst 9.2. The 9.1 (first release in 2009) is expected by next week already. So long: Install the Catalyst 8.12 WHQL if you can't do without AF.
At this point we have to blame ourselves. To be up-to-date, we delivered benchmark results with the new drivers as fast as possible, but didn't notice this quality faux pas in the heat of the moment. Usually we take a real close look at what happens (and we are often critiqued for that).
For our defense we have to say that our Warhead test sequence runs on tracks, which are flanked by Parallax Occlusion Mapping (POM) at Enthusiast details. This is not affected by anisotropic filtering and thus the textures are washy even at 16x AF. Below you can see two mouseover pictures that visualize the problem.
Grafikkarten-Rangliste 2017: 30 Radeon- und Geforce-GPUs im Benchmarkvergleich [März]