Athlon XP 2800+, AMD Hit 333!

Author : Wayne Date : 1st October 2002

HotCPU Tester Pro 3 :

HotCPU Pro is a great little piece of software written by Damon Chitsaz of Opusware who have now joined forces with 7Byte Computers. In addition to a diagnostic routine which pumps CPU usage to 100% for a definable number of hours it also features a quick and effective CPU benchmark. I ran the test on the 2800+ downclocked to 2200+ specs, on a genuine 2100+ and on the 2800+ running at full steam.

2800+ @ 2200+

2100+

 

2800+

The results here seem to pan out nicely with nothing unexpected to report. Certainly the 2800+ makes the previous two sets of results look rather ordinary by comparison.

Using a 6 hour burn in over night last night I'm pleased to say that both our test system and the 2800+ emerged victorious!

Burn-in Diagnostic Results

MemTach :

MemTach comes courtesy of CPUReview and again offers a variety of memory benchmarking results. In their words "MemTach 0.93a adds two new experimental tests; RASum Double and RAPSum Double - these tests traverse memory backwards in order to disable automatic prefetch lookahead in Athlon XP processors. This version also adds file consistency checking in order to prevent corrupted test results. MemTach 0.89a produces significantly more accurate test results than v0.80 due to: better compiler, assembly code "Sum" tests, optional "prefetch" code."

I've pasted the results from the genuine 2100+ (in yellow) alongside those of the 2800+ and again things seems right on track.

2800+ Results - White + + 2100+ Results - Yellow

ScienceMark 2.0 :

ScienceMark was written by Alex Goodrich, Sean Stanek and Tim Wilkens. They say that "Science Mark 2.0 is an attempt to put the truth behind benchmarking. In an attempt to model real world demands and performance, SM2 is a suite of high-performance benchmarks that realistically stress system performance without architectural bias.
The benchmarks in wide use today fail to accurately reflect system performance due to the following:

1.Relevance: The benchmarks test only 1 application and don't address a wider array of applications more representative of the user's market.
2.Abstraction: They are entirely comprised of synthetic tasks that don't perform a complex meaningful task.
3.Quality: May be poorly constructed from a C or Fortran perspective and limited in their ability to measure the true potential of a system.
4.Objectivity: The test is developed on or tuned for one architecture resulting in implicit performance bias.

Synthetic benchmarks are useful, and can tell the user valuable performance characteristics about their system's performance, but they should not be used in entirety to measure system performance; this role is reserved in greater part to real applications performing real tasks. Science Mark 2.0 is comprised of 7 benchmarks, each of which measures a different aspect of real world system performance.

For now I'll be running their MemBench module to see what it tells us.

"MemBench" - An unbiased, synthetic benchmark that sweeps through more than 15 different, publicly available, memory copy algorithms to measure the peak memory performance of your CPU’s caches and your memory subsystem. "

2800+

2100+

Okay, let's look at it in a different format :

What we need to remember is that this graph represents memory latency, the amount of time taken to retrieve data in simple terms, therefore the lower the result the better the performance. There's actually not a huge difference between the 133MHz 2100+ and the 166MHz 2800+ until we reach the 256byte data blocks. Considering the 2800+'s increased FSB I would have perhaps expected a greater disparity than this and I'd like to rerun this test later with a Thoroughbred core CPU to verify things.

<<< Back to Review Index | Page 4 - Performance >>>

Home