Tracy R Reed wrote:
Does anyone with more knowledge of hardware than I consider this significant? How much RAM could they fit on the chip if instead of making it quad-core they just made it dual-core and used the rest of the chip real estate for RAM? Since we don't use much of our CPU power as it
Not very much. Even one stick of DRAM chips (just count the silicon area) is probably as large as a quad-core CPU.
is and are almost always IO bound investing in RAM instead of cores could be a really great way to improve performance:
Actually, that's not really true anymore. We have *lots* of RAM and bandwidth in a system. Moving up the speed of your front side bus and memory buys you *very* little for most workloads.
The reason is that DRAM *latency* is still on the order of 50ns (20 MHz) while the rest of the system is now running on the order of 500ps (2 GHz).
Yes, I know that the numbers are probably sub-20ns, but the processors are now probably sub-200ps at the high end, and it doesn't change the fact that DRAM latency is 2 *orders of magnitude* slower than the processor. This hasn't changed in over 10 years--you might be able to make the case that it hasn't changed in 20 years.
DRAM latency has *NOT* followed a moore's law curve even remotely. In fact, DRAM latency is now so bad relative to DRAM bandwidth that it looks more like a sequential storage device than a random-access one.
Anybody have some operating system optimizations for drum or bubble memory we could recycle?
-a -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
