CPU cache is designed to minimise the need to access the system memory, so the more there is the less often the CPU has to wait around. Making sure that the right data and instruction sets are ready for use in the cache is one of the best ways of improving performance in the modern era. This priming of the caches is called prefetching. Prefetching is one area in which the Core architecture makes great improvements over the competing Athlon 64 technology. Each core has three prefetches (two for data and one for instructions) plus two further prefetchers for the L2 cache. In order to stop these prefetchers from starving the running programs of bandwidth, Intel has a prefetch monitor that gives priority to the “demand bandwidth”. In addition to an advanced prefetching subsystem, the cache itself is equally impressive. The main 4MB L2-cache (or 2MB for the entry level chips) is shared between the cores and can be accessed in just 14 cycles, with a 256-bit-wide bus allowing the cores a huge amount of bandwidth at very low latency.
Power Efficiency
The new Intel Core 2 Duo consumes considerably less power than the previous generation of Netburst processors, and is also more efficient than the Athlon 64 X2. A typical Core 2 Duo PC will consume around 155-165w at full load, compared to 165-170w for the Athlon 64. At the top end of the scale a system based on the Athlon FX62 will draw some 200W where as the flagship Core 2 “Extreme” that runs at 2.93GHz consumes 15% less at a little over 170W. By contrast the Pentium Extreme Edition 965 drew more than a whopping 220W. Whilst Intel therefore has an overall lead in terms of power efficiency, its performance per watt is even more impressive. In the majority of applications we saw the Core 2 Duo E6600 (2.4GHz) part outpace the FX 62, yet consume 30% less power to do so. Amazingly it is actually the Extreme X6800 that offers the best performance per watt of the entire Core 2 family. Much of this can be attributed to the highly efficient 65nm manufacturing process.
Dual Core, short pipeline
All Core 2 Duo processors feature two independent processor cores in a single physical package. These share a single 4MB cache as well as the same FSB. Whilst the Pentium D was a rather inelegant way of making a dual core CPU – at least in comparison with the Athlon X2 – the Core 2 Duo has been designed from the ground up to be a dual processor chip. Each core has more efficient SSE performance thanks to the fact that all 128-bit SSE instructions will now execute in a single cycle. This means that in programs that make use of SSE/SSE2/SSE3 optimisations you will see a hefty performance increase on Core 2 compared to Pentium D.
Pentium D CPUs have a very long 31-stage pipeline (a pipeline is a chain of data-processing stages where the output of one element is the input of the next) which means they have to do less work per MHz, and therefore can run at a higher clock speed. When the Pentium 4 CPU was first released, MHz was the only meaningful measure of CPU performance with the vast majority of consumers purchasing their PCs based on this highest number on the box. The downside of a long pipeline is that when a program branches (‘if’, ‘then’, or ‘else’ commands rather than ‘add’, ‘multiply’ or other mathematical functions), the entire pipeline must be flushed. Branch predicting helps to alleviate this problem, but in a desktop environment branching happens constantly, greatly reducing the speed gain of pipelining.
Core 2 has reverted to a much shorter pipeline of 14 stages which in basic terms makes the CPU much more efficient per clock. Advanced manufacturing techniques mean that the clock speed of the CPU need not be compromised in order to retain a higher workload per clock, hense performance is increased.
Whilst there are a number of other extremely important advancements to the Core 2 line, including wide dynamic execution and accelerated out-of-order executions thanks to smart memory access, in an effort to keep this article as byte-sized as possible we will forgo looking at those features in more detail until a later article. If you simply must learn more we encourage you to take a look at Intel’s Core 2 Duo white papers.
New chips, the same platform.
With our Core 2 press kit Intel was kind enough to supply a Core 2 Extreme X6800 CPU, a Core 2 Duo E6700 CPU and a 975XBX test motherboard. Intel actually has a new Core 2 ready chipset called the P965 which features an improved, lower latency memory controller and the newer ICH8 Southbridge, but as you may have guessed from its name, the 975XBX uses the slightly older but similarly speedy 975X instead. The 975X also features the benefits of ATi Crossfire support which the 965 does not enjoy. Crossfire has played a crucial part in making sure our gaming tests were not too GPU limited. nVIDIA fans will also be pleased to hear that the nVIDIA 590 Intel Edition chipset is ready to go and has already shown itself to be a stunning performer, bringing SLI to the Core 2 platform.
