Introduction:

The AMD Athlon64 Socket 939 platform was launched on June 1, 2004 to much fanfare. AMD enthusiasts were eager to get their hands on dual channel Atlnon64 motherboards and CPU's that didn't need registered ram (like the Socket 940 versions). The importance of using regular unbuffered ram was twofold. One, buffered or registered ram suffers a very slight performance penalty compared to the unbuffered ram. Two, Unbuffered rams are cheaper with much better selection on high performance modules. Which means for many, the opportunity to use low latency rams with arguably the fastest platforms that were coming onto market, the best of both worlds so to speak.

Total Hyper transport bus speed was increased from 800mhz to 1000mhz and estimates were that Socket 939 Athlon FX53 would enjoy about 5-10% performance advantage over FX53 chips on Socket 940 platform. Additionally, lower priced CPU’s like the 3500+ @ $520 and the 3800+ @ $720 were to be included in the Socket 939 lineup, to join the top of the line FX series CPUs who were priced between $850-900 each, bringing dual channel Athlon64 into the price range of some who could not have otherwise afforded the platform upgrade.

Reality though reared its ugly head, while the 939 platform was fast and indeed faster than Socket 940 systems at stock settings, there were relative few choices in mainboards, and most if not all had their quirks and bugs that inhibited someone from hitting high bus overclocks to bring forth the much higher levels of performance we desired.

---

To clarify, many boards while claiming to be flagship/enthusiasts boards offered very poor voltage tweaking options. Features that were supposed to be functioning at launch were either not implemented correctly, or due to design, just absent or broken. I refer mainly to the VIA’s KT800Pro chipset and the AGP/PCI lock that was touted to be one of the main features of this chipset, along with 1000mhz Hyper transport support.

NVIDIA’s NF3 Ultra chipset for S939 was curiously absent from the market until mid July, and was then initially only available on the very very expensive Gigabyte K8NSNXP-939 board retailing for $238+. This sought to undermine the platform to an extent and it was about two months before we found Socket 939 setups starting to make their way up the benchmarking ranks for highly overclocked PC's. Socket 940 may have been slightly slower clock for clock, but it was well established and top benchmarking wizards could work their magic with relative ease on the older platforms, but could not crack the nut on the 939 setups right away. Boards were the main obstacle, more specifically, the inability to scale high HTT bus along with high memory speeds. It seems evident though benchmark scores that the Athlon64 most enjoys very high HTT speeds, so it's crucial that high HTT was achieved at a preferable 1:1 memory interface with tight latencies. This was painfully apparent as highly overclocked Athlon64 socket 754 CPU’s, with their single channel memory interface, enjoyed many a benchmarking triumph over Socket 939 platforms, especially in the immediate months after 939's launch. I have been one to ride the bleeding edge of technology so I jumped in with two left feet and purchased an Asus A8V and an Abit AV8 in July 2004 after I had saved enough $$$ for the motherboards and CPU's to go along with them.

The results; I was not thrilled to say the least. My Asus died after about 2 days of tinkering. While under warranty, this was still a hassle to RMA and wait for the return. The Abit functioned well at stock speeds and overclocked somewhat, with a few quirks. The Abit would not maintain 1:1 ram speeds past 215HTT/Mhz (despite ram known to fly past 240mhz with ease) and would not run any 3D benchmark or game over 245HTT. While the AGP/PCI locks appeared to function at lower bus speeds, there seemed to be a "breaking point" at which they became non-functional and the board became useless past 245HTT. Certainly not what this enthusiast was looking for.

My replacement Asus A8V (ver 1.02) made amends for the first board. With the flash 1005.021 Beta bios, available in "Big-Toe's bios bin (registration required)"; I had functioning AGP/PCI locks, I could run 1:1 ram up to the limits of my 2x512mb OCZ PC3500EB ram which was about 240mhz stable. One concern I continued to have though was the HTT bus could only be run to about 250-260HTT with ram running the PC2700/ 5:3 divider before the board became very flakey and unstable.

Was it chipset limitation, or the AGP/PCI lock malfunctioning as in the Abit AV8's case? I wish I could tell you, I'm just not sure. I was lucky in some respects, as I believe most if not all of the non-USA sold boards would not function with the AGP/PCI locks enabled. There is hope though folks! The proud people at Asus are not ones to rest on their laurels and they have come out with a Revision 2.0 of the venerable Asus A8V, supposedly with properly functioning AGP/PCI locks without the need of a beta bios. I purchased one of these boards shortly after it becoming available into retail channels to see what improvements have been made, and what we consumers can expect from our Socket 939 setups.

---

The purpose of this article is to explore the limits of the Hyper transport bus(HTT) on this retail board. The HTT is the speed at which the cpu communicates with the motherboard, while different than a traditional Front Side Bus (FSB) that many of use as a common term, a higher HTT speed will increase overall performance much the same way and is the reason that people have sought to overclock their P3/P4, Athlon, and Athlon XP CPU’s by raising their FSB to highest max stable.