The HD5850 is based on the Cypress RV870 core, packing an impressive 2.15 billion transistors on a 334mm2 die size thanks to the 40nm process it is based on. This is only two thirds of the 3.0 billion transistors the Nvidia GF100 Fermi core sports but as proven with the last generation of cards, bigger is not always better.
To differentiate the two GPUs using the Cypress cores, ATI have disabled a few stream processors and texture units. The full core, as found on the HD5870, features 1600 stream processors, 80 texture units, 32 ROPs and a clock speed of 850/4800MHz. The HD5850 on the other hand, has one cluster disabled thus featuring 1440 stream processors and 72 texture units, representing a 10% reduction. The number of ROPs has stayed intact but the clock speeds have been lowered to 725/4000MHz for increased yields. The table below summarizes the key specifications for the reference designed boards.
The resultant memory bandwidth, fill rate and shader operations are certainly very impressive. The HD5850 should be provide about 80% of the HD5870 based on those figures alone but the benefit is a 16% drop in maximum power draw and a lower price.
The ASUS card tries to alleviate some of the deficit with a mild 40 MHz overclock on the core and 500 MHz on the memory, a 5.5% and 12.5% increase respectively. Of course, this isn't enough to threaten the HD5870 but we will see how it overclocks later on.
ASUS Specifications