A couple of months back we took a good hard look at the AMD Athlon X4 845 processor, one of the only chips around that uses AMD’s newer Excavator cores. Excavator succeeds the previous Steamroller architecture and despite being the fourth-generation of AMD’s Bulldozer derived processors, it showed enormous promise. We found that it offered way better single core performance in benchmarks such as Super Pi 32M, doing it with lower clocks, a lower NB frequencies with un-optimized, slower memory configurations. Today however we arise to a few new scores on HWBOT that indicate that AMD could actually be improving with each production run.
Enter Aussie overclocker, Newlife. AMD aficionado, retro benching expert and enthusiastic hardware modder – who better placed to examine AMD’s progress with Excavator. Having previously had issues getting hold an Athlon X4 845 chip in his native Australia, Newlife finally managed to get his hands on one. In fact it seems he got hold of more recent batch than had previously been pushed on HWBOT.
Armed with an X4 845 from a batch manufactured in Week 20 of this year (mid-May) Newlife took no time at all pushing the chip to its limits using his ice chilled AiO water cooler. His initial run pushed the CPU core to 4,569MHz (+30.54%) and managed a score of just 10mins 11sec 765ms, a new sub 5GHz record for AMD. The score was made with the CPU using a hefty 1.65v, NB at 1.4v and memory at 1.9v. One key aspect of the configuration was getting the base clock to run at 120MHz, a feat that would only complete a Super Pi run on the 1st core.
A few hours later Newlife had managed to squeeze a touch more out of his new Excavator cores – hitting a clock speed of 4,679.2MHz (+33.69) to complete a run in just 9min 58sec 875ms. This configuration required a somewhat high CPU core voltage of 1.7v, but it reaped the rewards of actually scoring below 10 minutes – a truly landmark result for an ambient cooled AMD processor.
The previous best SuperPi 32M score for the Athlon X4 845 was 10min 12sec 78ms. Posted by l0ud_sil3nc3 back in May of this year, the chip showed nowhere near the same headroom – despite being cooled on LN2 it managed only 4,597.57MHz (+31.36%).
The upshot as far as we can see is that AMD a) has significantly improved its single core, per thread performance and b), it appears to be getting better as time goes by as AMD’s 28nm manufacturing process matures and improves. This is all good news for the forthcoming Bristol Ridge platform, especially for a company that so desperately needs to get back in the high-performance race with Intel. Good news for Overclocking too IMHO.