ATIâ€™s beta motherboard gave us some real problems. We could not get dual channel memory to work at all and eventually the motherboard just stopped working entirely. However, we did have it running long enough to do a complete run of Spode Mark 3D. Unfortunately we couldnâ€™t run SYSMark, so I have no idea if the new South Bridge is an improvement over the last. Even when running in single channel mode, there was still an obvious improvement in performance, with a few exceptions that Iâ€™m going to strike down to the fact that itâ€™s a beta motherboard. These results are included at the end of the article, as they always are.
The bulk of my testing was instead done using nForce 4 and nForce 5 as a Socket 939/AM2 comparison, as was done in my previous nForce 5 article. This time we used SYSMark 2004 SE and PC Mark 2005 as a way of gauging performance difference. As the core itself has not changed significantly, it doesnâ€™t warrant much more testing than this, that is â€“ not until we have a Conroe machine to compare to that wasnâ€™t built by Intel.
SYSMark 2004 SE runs a whole suite of common tasks including web design, image editing, 3D manipulation, general office work and video encoding. This gives us a pretty good impression of overall system performance. Using 800MHz DDR with 4-4-4-12 timings, we saw a four per cent improvement in performance, with most of this being in â€œOffice Productivityâ€ which increased by six per cent. With the faster 940MHz memory, but higher latency 5-5-5-15 timings, we noticed a six per cent increase in system performance, once again mainly in Office Productivity which saw a ten per cent increase in performance.
These sort of performance increases are similar to those we saw in our gaming testing last time. It was interesting to see how the Asus RD580 motherboard we used for testing was a better performer than the nForce platform, more than likely due to its use of a ULi South Bridge.
PCMark showed a similar trend, with the one exception of the hard drive portion of the testing, which was significantly faster on nForce 4 than any other platform. This was a repeatable result, but I find it hard to believe that nForce 5 would suffer from poorer SATA performance than nForce 4. When I return from Computex, I intend on investigating this phenomenon further.
AM2 is faster in every situation when compared to Socket 939. However, this speed increase is minimal and I would advise anyone with a fairly up to date system to wait out a little longer. Newer 65nm processors with Level 3 cache should offer considerable speed increases. Socket 939 will still be available for some time, as there are far too many Opteron users to justify halting production just yet.
Then of course there is Conroe, which is looking to be an excellent performer. I have my own doubts about Conroeâ€™s multiple-core performance â€“ an area where AMD is traditionally quite strong, but only time will tell where your money should be spent. For many people, a high end Socket 939 system may currently offer better value while prices are cheap. It will still be PCI Express and you can reuse that DDR400 memory thatâ€™s in your current system. Just donâ€™t go investing in any expensive cooling solutions as you may find it useless when you go to upgrade.