AMD decided to surprise us this summer with new generation of dual core CPU. Goal is, obviously, to take key positions in mainstream segment and to pull the rug out from under Intel’s feet. One thing is for sure: K10 cores are starting to dominate in every segment of AMD product lineup. This is very good news for all users that own an AMD Socket AM2 motherboards because new CPUs present themselves as easy and cheap upgrade solution compared to older Brisbane and Windsor models. It is obvious that there is a lot of hidden potentials in new AMD 45nm manufacturing process. As illustration of this statement we will tell you that new dual core CPUs can be easily overclocked up to 4GHz while stability won’t be jeopardized in any way.

AMD Athlon II 250 - Regor core

Regor core got its name by one of the most brilliant starts at the night sky: Gamma Velorum. This is native dual-core CPU, without L3 cache, very much alike Athlon 64 X2. Athlon II X2 250 has 2x1MB L2 cache where each core has its own 1MB of L2 cache. This architecture was used on older 90nm K8 X2 CPUs, based on Windsor F2, F3 and Toledo cores. Lack of L3 cache on Athlon II is compensated by larger low latency L2 cache. This combination has proved itself as good combination for gaming platforms. Regor is made in advanced 45nm manufacturing process and it wasn’t made by cutting defecting Daneb cores. This means lower electron “leakage”, lower thermal dissipation and smaller core surface. Initial frequency is set high at 3GHz, while thermal dissipation is set at 65W. As far as speed and efficiency goes, this core doesn’t fall far behind Callisto core that has 2x512MB of L2 cache and 6MB of L3 cache. If compared clock-per-clock with older Kuma X2 65nm K10 core is provides similar or even better results, but as contrast to Kuma, has lower thermal dissipation levels and better overclock potentials.
AMD Athlon II 250 can easily be as good, or even better, as Core 2 Duo CPUs with 2MB of L2 cache on default clocks and when overclocked, which wasn’t the case before in this price range.

This diagrams show how clearly how pre-fetch and cache algorithms are improved on new AMD CPUS and 64-byte stride clearly shows difference in memory and L2 access that is almost twice larger while 256-byte stride is better in case of older F3 Windsor core because of two things. First, DDR2 memory at 1000MHz and tight timings has lower latency then DDR3 memory, and secondly memory controller on K19 CPUs operates at only 2GHz compared to Windsor F3’s that operates at CPU frequency. Because of this K8 core performes better WinRAR benchmark.
Nevertheless, K10 core memory controller has wider bandwidth than K8 and internal path between two cores is twice as large, which applies also to L1 and L2 cache. Double 1280bit load is privilege of new K10 architecture. Science Mark 2 results show clearly that operations with L1 and L2 cache are twice as fast as on older K8 core.
Because of better pre-fetch mechanism and new (and improved) memory controller, K10 microarchitecture is optimized for high latency and bandwidth DDR3 memory. K10 core gains 5% in performance levels if DDR3 memory is used while old K8 architecture showed almost no performance gains when transition from DDR1 to DDR2 memory was made.