Some years ago I did informal benchmarks of 10 mb ISA cards
doing disk copies. I don't recall the exact details, but
at approx 250-300kbytes (certainly well over 2 megabits),
and including file system overhead, cpu utilization was
not tremendous. I want to say 30-40%, but sadly I can't
remember exactly. Anyhow, it was MUCH less than full
capacity. This was on a 50mhz 486, btw. I just repeated
on a P6 200 at around 600kbytes/sec (say 5 mbits), including
file system overhead and IDE disk processing, cpu ran
about 20% (total cpu around 25%, but I run about 5% background
normally).
I read benchmarks of some of the early 100mb PCI cards.
The Intel 10/100 card was able, according to the benchmark,
to maintain a transfer rate of 70 MEGA bits per second at
something on the order of 60-70% cpu utilization of
a 120 or 133 mhz P5. Possibly it was 100mhz, and possibly
it was less cpu than this, but I'm virtually certain
it was less than 70%.
Intel was the least cpu intensive benchmarked
at the time, but I noted later benchmarks in which others
improved rapidly. This has now been at least two and perhaps
3 years ago. Performance has improved across the board.
Of course, I'm still running those 3-year old Intel cards <G>.
In short, I just don't think you're going to see significant
CPU degradation in ANY modern NIC with a PII class processor.
Note that these speeds are orders of magnitude higher than
anything you're likely to get off the internet, even with
cable or DSL. 10 megabits is 10 T1 lines, remember <ggg>.
Sorry, I know nothing about Wintel multi-cpu systems.
Maybe someday.
Spots |
