On Sat, 23 Nov 2002 19:37:02 +0000 Andrew Suffield <[EMAIL PROTECTED]> wrote: [snip] > Turning on IDE DMA is a performance improvement of around a factor of > 6-10 (2-3Mb/sec -> 25-40Mb/sec), for disk-bound operations. You can > get *more* than a 6-times improvement in performance?
Well, no. But then, I've not had to do anything other then use a 2.4 kernel to get IDE DMA. But most desktop work is letency based and using a kernel with the low latency patch and preempt cuts max latency by about 6 times based on some workloads ( 600ms --> 20ms ). > > > Under > > debian's X/glibc/kde I can't move windows with contents without > > 'tearing.' > > Something is seriously wrong, most likely with your display > driver. You should be able to achieve this on a 486, albeit at a lower > resolution. Yep. I was using the "nv" driver from then woody's X. I blows chunks. The non-free "Nvidia" driver works fine but taints the kernel (and causes random crashes sometimes.) I would bet that debian's "nv" driver has gotten much better over time as newer X versions move into testing. > > X is CPU bound if you can't > > move windows smoothly because it has latency that is too high. > > Moving windows around is a memory-to-framebuffer-throughput-bound > operation, not a CPU-bound one (unless your processor is really slow > and your resolution is absurdly high). Even with DRI, moving a window about over mozilla peged my 1GhZ duron. Well, I guess 1600x1400 is absurdly high, but I like how the fonts look (even the non-true type ones.) Thomas
pgp6uEJiXJetZ.pgp
Description: PGP signature
