Benchmarks: AMD's New Athlon XP 1800+, Part 1

By Van Smith

Date: October 9, 2001

Due to time constraints, benchmark results will be presented with little commentary.  We will offer follow-up articles later today and in the days to come.  For commentary on AMD's controversial "Model Number" rating system, please view our recent article.


Test Beds

We used AMD's reference system which consists of an AMD Athlon XP 1800+ with 256 MB of CL2 PC2100 DDR SDRAM.  The motherboard was the GigaByte GA-7DX which uses the AMD 761 north bridge with a VIA 686B south bridge.  The video card was a GeForce 3.

We compared the Athlon XP's performance with the 1.7 GHz Pentium 4 running on VIA P4X266 DDR SDRAM chipset.  It too used 256 MB of CL2 PC2100 DDR SDRAM.  For logistical reasons, we had to use a 32MB ATi Radeon DDR so we have refrained from reporting 3D benchmark results in this article. 

We will follow up with more hardware permutations shortly including results from an i850/RDRAM based P4 configuration as well as scores from mating the Athlon XP with the high performing VIA KT266A chipset.

Both systems used in this review ran Microsoft Windows XP.


SysMark 2001 Handicaps Athlon XP

As shipped and even with the latest patch (patch 3), SysMark 2001 severely handicaps the Athlon XP.  Although the benchmark states that it supports Windows Media Encoder 7 and higher, in fact the program ships with a static version of WME that does not recognize SIMD enhancements native to the Athlon XP.

Fortunately, AMD and Microsoft have worked closely together to provide a patch to perform an end run around this slighting.  Although Windows Media Encoder is a relatively small component of SysMark2001, simply with this one small change, the Athlon XP's performance shoots upwards on this test as shown below.

The Office Productivity portion should not have been directly impacted by this fix, so the small difference was likely due to the innate variability of SysMark 2001.

We have written before about the intimate connection between BAPCo and Intel, but even so it is hard to take SysMark 2001 seriously simply based upon the obviously skewed workload it employs, not to mention this WME oversight that could have been corrected with a patch.  We hope to provide an article detailing SysMark 2001's workload in the days to come and illustrate how its usage model is atypical and contrived.

We have discussed BAPCo and SysMark numerous times with AMD and the chip company is well aware of BAPCo's position and only grudgingly accepts SysMark becuase it is so widely used with many sites appearing dead set set on featuring this benchmarketing product. 


Memory Performance

There are two primary memory subsystem performance characteristics.  Latency is the measure of the delay from when data is requested and until the time it actually arrives.  In other words, latency is literally the lateness of data.

Latency has not always been a strength of the Athlon, but the Athlon XP screams.  This vast improvement is possibly due in part to hardware data prefetching.  The results are total raw execution times, so lower numbers are better.

Keep in mind that the results above are a function of the entire memory subsystem, so the KT133A chipset plays a role in the 1 GHz Athlon score.

We have also developed a benchmark that reflects latency versus varying dataset sizes.  This test demonstrates cache performance as well as main memory / chipset latencies.  Higher numbers are better in this test.

The first graph is from the 1.7 GHz P4, while the second is from the Athlon XP.

The 1.7 GHz Intel P4

The Athlon XP 1800+

The Athlon XP shows between a two and three fold advantage over the Intel Pentium 4 over the entire range of this latency heavy test where values from random elements of a source array are copied to random elements in a target array.  The array sizes are gradually increased throughout the test.

To be continued...

Check back later today because we have many more results to post.

Click here to go to Part 2


Pssst!  Our Shopping Page has been updated.