With a largely shared driver code-base across platforms, the binary graphics drivers offered by AMD and NVIDIA perform at roughly the same speed for OpenGL between Linux and Windows; that's traditionally been the case and what Phoronix benchmarks in prior years have shown for NVIDIA and AMD. However, the OpenGL performance difference between operating systems is beginning to widen due to compositing window managers and other factors now affecting the results to a greater extent. In this article are benchmarks of the proprietary NVIDIA graphics driver from Microsoft Windows 7 and then development snapshots of Ubuntu 12.10 with Unity and KDE desktops.
Recently published were the Intel OpenGL performance comparison of Apple OS X, Microsoft Windows, and Linux. The Intel OpenGL results were interesting as well, but the open-source Intel Linux graphics driver code-base is completely different from the Intel Windows driver with their development efforts not really being shared on the driver stack across platforms. I also have more Sandy Bridge and Ivy Bridge comparative numbers from more recent drivers/code to be published in the coming days. In addition to today's NVIDIA Windows vs. Linux benchmarks, there's similar testing that was done for the AMD Catalyst driver on both operating systems too.
For this cross-platform NVIDIA driver benchmarking, the NVIDIA GeForce GT 220, GeForce GTX 460, and GeForce GTX 680 graphics cards were used. These graphics cards represent NVIDIA GPUs of recent generations as well as being on the low and high-end of their graphics card spectrum. The latest NVIDIA graphics drivers were used on each platform: on Windows this was 301.42 WHQL while under Ubuntu Linux the latest release at test time was the 304.43 binary.
The same system was used throughout testing on both Windows and Linux, which was an Intel Core i7 3770K "Ivy Bridge" system with an ECS Z77 motherboard, 8GB of RAM, OCZ Vertex SSD, and was running Ubuntu 12.10 with its Linux 3.5.0-13-generic x86_64 kernel with an X.Org Server 1.13 pre-release, GCC 4.7, and an EXT4 file-system. Due to the major performance differences between Linux desktops and with the Unity/Compiz performance struggling for the default Ubuntu desktop, tests were done from this Linux software stack with the default Unity 6.4.0 + Compiz and then with KDE 4.9.0. With the KDE testing, suspending of desktop effects on full-screen windows was enabled to ensure maximum performance while all other settings were at their defaults. Microsoft Windows 7 Professional x64 had all available service pack / security updates available as of the time of testing.