Author: Anthony Cozzie
Date: 18:27:28 02/19/03
Go up one level in this thread
On February 19, 2003 at 20:58:20, Charles Worthington wrote:
>That is like comparing a Yugo to a Ferrari.
OK, as a computer architecht I have to correct this foolery.
1. Deeper pipelining do not necessarily lead to greater performance.
A deeper pipeline decreases the cycle time of the processor, but increases
the branch misprediction penalty and causes all sorts of other nasty forwarding
stalls. Intel believes in superpipelining; I read a paper where they have
simulated a 50 stage pipeline and believe it has higher performance. There are
also people who think this is hogwash.
2. It is very difficult to determine the cycle time of a circuit
The problem is that the TOPOLOGICAL longest path is not always the longest
data correct path. It is very difficult to give a good example for this, but I
suggest you do a search for 'false path' on google and read up a bit. When
intel makes chips, they don't classify them as 2.8GHZ or 3.06GHZ. They just
make chips. When the test them according to heat resistance, etc, then they are
classified. This is why overclocking a PIV 2.2GHZ to 3.0GHZ is safe;
overclocking a 3.0GHZ to 4.0GHZ is much more suspect: no one knows all the real
limits in the chip. The chip may have very very annoying corner cases where it
fails.
3. Processor performance is very dependent on the application.
Anandtech did a review of the Barton version of the Athlon [512KB cache]. In
some benchmarks the PIV beat the Athlon, and in some the reverse.
Unreal 2003 Botmatch: AMD over PIV, 75:70
Rendering Time in in 3DMAX: PIV over AMD, 169:227.
This brings up another point. Intel has a vision of the processor as the
multimedia center of the home. The PIV is designed to excel at multimedia
signal processing applications (like 3DMAX). These applications have few
branches and a great deal of parallelism; the deep pipeline of the PIV does not
matter. A Chess Engine, however, is integer code with lots of branches, which
is why the Athlon usually performs well.
anthony
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.