Well, they're not working together. Unless you're not willing to tweak it a little bit. So, out of the box, you won't be able to test brand new Linux CFS scheduler, merged in the 2.6.23-rc1 release, if you drive your Nvidia card with the proprietary driver. I guess that's what we get for running binary drivers.
Fortunately, the driver needs only few simple fixes to compile properly. I'm attaching them in the patch form below. What you need to do is:
% NVIDIA-Linux-x86-100.14.11-pkg1.run -x
Creating directory NVIDIA-Linux-x86-100.14.11-pkg1
Verifying archive integrity... OK
Uncompressing NVIDIA Accelerated Graphics Driver for Linux-x86 100.14.11.....
% cd NVIDIA-Linux-x86-100.14.11-pkg1
% patch -p1 < nvidia-fix.patch
patching file usr/src/nv/nv-linux.h
patching file usr/src/nv/nv.c
In other words, unpack the archive, patch it and only then run the installer. Otherwise you won't be able to compile the kernel module because of two changes in the kernel ABI.
kmem_cache_create() doesn't accept parameter for cache deconstructor anymore, and
unregister_chrdev() now returns void. Those two changes together produce the following error (unless the driver is patched) in
/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.c: In function 'nvidia_init_module':
/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.c:1326: error: too many arguments to function 'kmem_cache_create'
/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.c:1435: error: too many arguments to function 'kmem_cache_create'
/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.c:1569: error: void value not ignored as it ought to be
/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.c: In function 'nvidia_exit_module':
/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.c:1601: error: void value not ignored as it ought to be
make: *** [/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv/nv.o] Error 1
make: *** [_module_/root/NVIDIA-Linux-x86-100.14.11-pkg1/usr/src/nv] Error 2
make: *** [modules] Error 2
NVIDIA: left KBUILD.
nvidia.ko failed to build!
make: *** [module] Error 1
make: *** [module] Error 2
ERROR: Unable to build the NVIDIA kernel module.
Finally, my personal opinion about the new scheduler (still in heavy development) and 3D display performance. Well, one thing's for sure, desktop (compiz) is snappier. Simple tests of cube rotation while the kernel is compiling (on both cores) show that CFS is much better than the old scheduler, cube rotation is much smoother than before.
3D game (tremulous) performance on the other hand really is somewhat suspicious, the game seems a little bit jerky when run on the new kernel, but I think it needs more testing before I can claim that for sure. Whatever the result of testing is, I'm quite sure Ingo Molnar will do everything in his powers to improve the scheduler in the coming months, such that 3D games run better, too (if there are problems now).