On 6/20/06, James Richard Tyrer <[EMAIL PROTECTED]> wrote:
> That's why we chose PCI-X. That's because PCI will be around a while, > and we can slap a PCIX-PCIe bridge chip on a board and get PCIe. > That is what nVidia is currently doing, and it can have performance issues. <SNIP>
Oh, I know. But I'm counting on using DMA so heavily that the extra latency won't be noticable.
>> > Can we get the performance we need with 65 nm or do we need >> > to go smaller? >> >> If the design were ready today, could we even get 65 nm? Isn't AMD >> still using 90 nm? > > I don't think it's necessary. 65nm, compared to 90, buys you mostly > power consumption and yield. Performance doesn't go up that much. > Keep in mind that it's wire delays, not transistor switching time, > that dominates chip performance. 65 nm gives you shorter wires. Are they faster?
I'm not sure. I think the capacitance goes up, taking away some of the advantage.
<SNIP> >> What versions of Unix block I/O to do other things? It is normally >> I/O that gets preference. > > I'm not sure what's being discussed here. Usually, you try to overlap > CPU and I/O by using DMA. But the process waiting on the I/O is > blocked. And sometimes, you can't do the I/O via DMA. The issue that I have is when either keyboard or mouse input is blocked by other processes.
Linux has a problem that Solaris resolves. Under Linux, the mouse driver talks to the X server, which talks to the graphics card. If the X server is heavily loaded, the cursor skips around. Under Solaris, the mouse and graphics drivers have a standardized interface that allows the mouse driver to talk directly to the graphics driver, bypassing X.
A secondary issue is when a redraw has to wait for another process. Some of the time, I presume that it has to wait for a swap which can't be fixed except with more main memory.
True, but if it's X11 that's swapped out, at least we can deal with the mouse cursor issue.
> With a graphics card that uses DMA, you offload the I/O overhead from > the CPU to the GPU, so the CPU can do other things. This has the > effect of lowering the load for the X server, so it gets a higher > process priority. With a graphics card that runs the X server, you would offload even more overhead. :-D
True, but is it worth it? _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
