Synthetic Testing Analysis with SiSoft Sandra - Latency
When the Brisbane core was first introduced in February, there was a fair amount of talk about latency issues. This probably would not have gotten as much attention as it did except for the fact that the processor was specifically billed as having a smaller die size and no loss in performance. It turns out, the Brisbane core had higher latency when accessing cache, but the performance impact showed up more so in synthetic testing rather than real world tests. Experience told us we should take a look at several latency tests and overall cache performance to see if anything unusual shows up with the new low power Athlon X2 BE-2350. To start, we ran SANDRA XI's Memory Latency test.
According to SANDRA XI's Memory Latency component, there was a nominal increase in latency recorded between the Brisbane based Athlon 64 X2 4000+ and the Athlon X2 BE-2350, with 6ns leaning in favor of the 4000+
Contrary to the Memory Latency test, the Cache and Memory module reported improved performance compared to the Athlon 64 X2 4000+, with the Athlon X2 BE-2350 posting an additional 234MB/s over all. Ultimately, these results are all very close in the end and we'd rather rely on real world application testing before drawing any major conclusions.