dshr@SUN.COM (David Rosenthal) (04/21/89)
> Perhaps someone at Sun can comment on the projected > performance of X on the recently announced color > Sparcstation I. > > Is the processor fast enough to do a good job with > an X server, or is the GX accelerator really needed. > > My own opinion is that even with the lastest Purdue > speedups, the X11R3 server on a 3/60C is not within a > factor of 2 of being "good". > > A related question for Sun: Is the GX upgrade for a > 3/60C a big win for X. The port of the MIT sample server to Sun hardware is not designed or implemented to provide optimal performance. It was intended to provide a timely, correct and highly portable basis for other ports. As part of this, it uses all Sun framebuffers as if they were dumb (memory-only) even though they may have graphics hardware (e.g. the CG2) capable of providing better performance. I have not tried the MIT sample server on any of the new Sun machines, but there should be no reason why it would not work in the same way, again treating all framebuffers as if they were dumb. The GX hardware is designed to provide high performance for both X and NeWS imaging operations, but the MIT server will not use it. It will treat the GX hardware as if it were a dumb framebuffer. Sun does not support the MIT server and will not be providing any way for the MIT server to exploit the GX hardware, just as we did not provide any way for it to exploit the CG2 hardware. There is no commitment from Sun that the MIT server will be able to run on future display hardware; we might, for example, decide to build a display that could not be accessed as a dumb framebuffer. The X/NeWS server (currently in Beta) provides much better performance (especially in color) than the MIT server on all platforms we have compared them on. It runs on all the new Sun hardware platforms, but as far as I know we have not compared the two servers on them (I don't work on the window system any more, so I might not be up-to-date). It exploits the GX hardware if it is available. I have used the X/NeWS server on a SPARCstation 1 with the GX option, and the X performance is impressive. So the answer to your question is that the GX is a big win for X, but not for the MIT sample server. David.