Hardware Freaks with:

This space for rent!

• blah
  • blah

This chart shows the results for various systems and their L2 latency. The program used to collect the data is called Latency2 (pretty cool name, isn't it?), and was written by roelof, who is also the author of SetiSpy. For those interested in comparing their system to the chart, feel free to download the program here.  However, please note that I don't need any more times for the table, unless you have a unique processor that isn't included already (such as a Cyrix or the 256K Athlon (when it arrives)).

The chart has the Data Size running along the bottom, which shows the size, in K, of the data chunk being processed.  The vertical axis shows the time in ticks that it takes to finish the operation. Lower is better all the way across, although we're concerned with performance up to the 1024K mark, as the Seti client works with data in 1M chunks.

Looking at the results of the chart shows some very interesting information. Notice the line for the P90 at the bottom- it's almost straight all the way across!  My guess is it's due to the low (1.5) CPU multiplier, so the CPU isn't constantly waiting on the memory to catch up. Looking at the results for the 2 Xeons with 1M of L2 cache shows why they are such excellent performers, as the memory latency stays very low all the way up to 1024K. Look at a PIII E or Celeron for comparison, both jump up right after they reach their L2 cache size, which is smaller than the working set that Seti uses. The Pentium Pro also demonstates similar performance here, staying very level until the L2 size is exceeded. However- look at the results from the Athlon. While it's latency is very low up to the 64K mark (the size of the L1 cache), it steadily increases after that, performing very erratically. Notice how it actually speeds up from 164 to 256K, and then begins a long ascent all the way up to the 2M mark, where it finally levels out. I think this is one of the reasons that the Athlon does poorly at Seti.