Benchmarks: SiSoftware Sandra
Clearly moving to an on-die memory controller has paid dividends as the Core i7 920 squeezed out a memory bandwidth of 17GB/s in SiSoftware Sandra using nothing more than DDR3 memory running at 1066MHz. Running three of these modules in triple-channel mode has huge performance benefits, as we have just witnessed. The slightly higher clocked Core i7 940 provided another gigabyte per second of memory bandwidth, matching the Core i7 965 Extreme Edition processor.
Overclocking the Core i7 965 EE along with the DDR3 memory allowed for a memory bandwidth of 21GB/s, which is very impressive. When compared to the Core 2 Quad Q9650 running dual-channel DDR3-1333 memory, the Core i7 920 delivered well over twice the memory bandwidth. Although the Phenom X4 9950 also features an on-die memory controller, it is limited by the front side bus and therefore maxes out at around 10GB/s.
Looking at the Sandra processor arithmetic performance we see that again the Core i7 920 is easily able to waste the Core 2 Quad Q9650. Here the Core i7 920 delivered 60% more performance in the whetstone test, and 24% more performance in the dhrystone test. After this point the Core i7 processors just continued to become faster as the clock speed was increased.
In the Sandra multi-media test we see the Core i7 processors offer significant performance advantages over both the Core 2 and Phenom X4 processors. The Core i7 920 was 17% faster than the Phenom X4 9950 in the multi-media float test, and 21% faster in the multi-media int test. Again the performance increased steadily with the processor's frequency, and overclocking the Core i7 965 Extreme Edition had a huge impact on performance. The 3.88GHz overclocked Extreme Edition processor was 38% faster than the Core i7 920.
This benchmark is specifically designed to measure the efficiency of the different multi-core processors with their different architectures. This benchmark does not test how fast the cores of the processors are, but rather how fast the connection between them is. Not all multi-core processors are made equally as their architectures differ greatly. When looking at multi-core efficiency it is clear that the Core i7 is going to bring multi-core processing to a whole new level.
According to SiSoftware the Phenom X4 9950 has extremely poor multi-core efficiency, providing just 3GB/s of bandwidth between the cores. The Core 2 Quad Q9650 sees a more impressive 18GB/s of bandwidth between the cores. Well, this would seem impressive if the Core i7 920 did not deliver 82% more bandwidth at 32.6GB/s, which is again amazing. The Core i7 940 is even faster and so is the Core i7 965 Extreme Edition.
However, the most impressive result is seen when overclocking the Core i7 965 Extreme Edition processor to 3.88GHz, where the bandwidth between the cores reaches an insane 45GB/s! At this point scalability would seem to be a rather tangible improvement of the Nehalem architecture.