AMD Athlon 64 3200+ and the Asus K8V Deluxe

by Josh Walrath

     I distinctly remember the release of the original Athlon, and it was a smashing time to be an AMD fan. I had several test machines in my lab, one of which was a well-loved K6-III 450 running on a Tyan Super 7 motherboard. I thought the combination was pretty quick at the time, but it was not the worlds greatest gaming machine. In fact, it was more than a little slow at quite a few games. I also had a P!!! 500 running in another machine, and it on the other hand was a fine gaming machine, but oddly enough seemed a bit slower than the K6-III in many desktop situations. Then AMD released the K7. Delivered to my doorstep was a K7 550 MHz processor and an FIC SD-11 motherboard (the first real mass produced K7 motherboard). I couldn’t wait to test this out. It replaced the K6-III machine, and upon installation and testing, the new combo totally blew me away. AMD had finally arrived and had a very competitive product to go up against Intel.

     September 22, 2003 was another important date for AMD, as they released their desktop K8 products. These new products are divided into two categories, the high end Athlon 64 FX 51 and the consumer level Athlon 64 3200+. AMD apparently only sent out FX 51 samples to many sites, and very few A64 3200+’s, so many of the reviews put up initially only featured the FX 51. I was actually much more interested in the standard A64 3200+, so I went out and bought one. I was curious how much of an improvement the A64 3200+ had over the Athlon XP series, since we all know that the FX 51 is arguably the fastest available desktop processor in the world.

AMD keeps the package clear so the user can see what they are getting into!  The supplied heatsink is more than adequate for this particular speedgrade of Athlon 64.

     There are three things I want to accomplish with this review. The first is a basic introduction to the K8 architecture and the Athlon 64 3200+. The next is a solid review of one of the first available K8 motherboards, the Asus K8V Deluxe. The final thing I want to do is compare the Athlon 64 3200+ to one of the older Athlon XP products, the 3000+ running on the 166 MHz FSB nForce 2. Most other reviews have pitted the Athlon 64 products against the Athlon XP 3200+ and the Pentium 4 3.2 GHz. I was curious how the Athlon 64 stood up against a slightly older, but still fast processor. The XP 3000+ fit the bill nicely, and the performance differences are actually quite impressive.

     One other side note is that of overclocking the Athlon 64. There are some very significant roadblocks with this architecture to adequately perform extreme overclocking. While Intel has the 865 and 875 chipsets that regularly clock up above 250 MHz, the new architecture of the Athlon 64 does not allow such overclocking, and I will explain why, as well as a possible work around.

     This little number is based on the new K8 core, which has some very significant improvements over the previous generation of AMD chips. The first, and probably the most important improvement, is the integrated memory controller. While previous generations of CPU’s have relied on memory controllers in the northbridge of their supporting chipsets, the Athlon 64 takes over this job. This cuts overall data access latencies in half, if not more. When most data accesses are single or double data words, the decreased latency involved in such accesses can significantly improve the overall performance of a processor. Intel has really put an emphasis on streaming data at high speeds, AMD is doing just the opposite with the Athlon 64 3200+. The 3200+ features a 64 bit DDR path to the main memory, while the higher end FX 51 has a 128 bit path (and requires registered DDR at this time). The Pentium 4 features a quad pumped 200 MHz CPU to chipset bus, while the memory controller in the chipset also features a dual channel DDR setup.

     While the FX 51 and Pentium 4 have very similar overall bandwidth numbers, the little Athlon 64 3200+ has only half that number (3.2 GB/sec vs. 6.4 GB/sec). At first glance it would seem that the Athlon 64 3200+ would be at a significant disadvantage, but benchmarks have shown that the Athlon 64 3200+ is at least equal to the Pentium 4 3.2 GHz processor, and only slightly behind the FX 51. Again, this is mostly due to the nature of memory accesses. Streaming data is comparatively uncommon as compared to single or multiple d-words, which is what most applications utilize. Only when a user gets into really heavy duty video editing work, or other such streaming applications, will the Athlon 64 3200+ show its disadvantage. Of course, people who typically do that work have the money to buy the FX 51, which gives the dual advantage of 6.4 GB/sec bandwidth and low latency data accesses.

     The next major improvement is that of the inclusion of a 1 MB L2 cache. If you remember, the original Athlon XP had 256 KB of L2, while the Barton core XP had 512 KB of L2. AMD has shown that its current Athlon architecture is very well balanced, but the pipelines are actually underfed. The first iteration of the Athlon had roughly 33% utilization at any one time of the core, but further improvements to both the memory architecture (DDR) as well as cache, TLB, and other advancements, have put that number up around 50% for the current Barton core Athlon XP. AMD hopes to improve that number again by increasing the L2 caches. Having larger caches allows the processor to access that data at full clock speed, and not have to go to the main memory, which even with the integrated memory controller is around 4X to 6X slower. The already large 64 KB L1 data and 64 KB L1 instruction cache are further assisted by the large 1 MB L2 cache.

     The Athlon 64 pipeline has increased from a 10 stage architecture, to a 12 stage. Many are claiming that these two extra stages are there to help the Athlon 64 clock higher than the older Athlon XP. This may or may not be true. It certainly won’t hurt the overall clock ceiling of the Athlon 64, but there is more to making faster transistors that increasing the number of stages in the pipeline. From the documentation released from AMD, it appears that those two extra stages are mainly for the SSE2 implementation that AMD has designed into the Athlon 64. In other words, AMD had to rearrange the stages of the Athlon 64 to include the SSE2 functionality. Again, those extra stages may help in the headroom department, but they were not specifically placed into the design to improve clockspeed.

Next: More Athlon 64 3200+

Copyright 1999-2003 PenStar Systems, LLC.