Nehalem is finally here. We’ve been hearing about it for some time, but it has finally arrived and it is indeed impressive. Without a doubt, it’s the fastest x86 CPU available on the market today, and probably the fastest single socket CPU available for any platform today.<\/p>\n
What may surprise many is that the biggest changes in the CPU, besides the physical differences, are the architectural changes, which are quite significant. Besides all the technicalities of the CPU, the most important thing about this CPU is performance. The 965 model operates at 3.2GHz, just like the outgoing
Intel Core 2 Extreme X9770, but the difference in performance is staggering.<\/p>\n
This is particularly true in games, be they multithreaded or not. The i7 965 delivers performance usually 10% to 20% faster than the X9770, which is plenty, considering that the clock speeds are identical. There are many contributing factors to this, but the most obvious contributions are the integrated DDR3 memory controller, the 8MB of L3 cache, wider execution units, better branch predication, the return of Hyper-threading and far better inter-core communications through Quick Path Interlink (QPI).<\/p>\n
The system is similar to what AMD has been using for several years now, actually since the advent of the Athlon 64 CPU. However, QPI operates at a higher frequency and delivers more bandwidth than what is currently available on the AMD platform on the AM2 socket. So how does the 965 perform in synthetic tests?<\/p>\n
Well, very well, actually. This is right across the benchmarks from the old ones to the latest tests, especially in the two Vantage tests that are inherently multi threaded. These particular benchmarks may exaggerate the Core i7’s advantage over the previous generations multi core CPUs; however, the synthetic numbers prove to be more than academic as video encoding proves that hyper threading speeds up video encoding signifi cantly, and together with the integrated controller and other technologies, allow the i7 965 to deliver twice the performance of the 9770 in some instances.<\/p>\n
In single threaded tests like Super Pi and Wprime, the i7 is significantly faster than the outgoing processors, sometimes up to 10% more efficient than the Penryn-based CPUs in these single threaded applications. This is particularly useful because the i7 CPUs, despite what some Intel engineers insist on, do not overclock as well as the last generation CPUs. For the enthusiast or the gamer who has pushed their current CPU quite far, the clock efficiency advantage of the i7 may not be enough to offset the frequency advantage their current CPU has.<\/p>\n
While a 4GHz i7 965 is impressive in any application, a 4.5GHz X9650 is just as impressive, and in some situations may be a little faster. So the value of the i7 at present will depend largely on what system you are currently running.<\/p>\n
Our particular sample was not a fantastic overclocker; in fact it was amongst the poorer samples available. However, it is worth mentioning that was a very early engineering sample which is very sensitive to which motherboard it is installed in.<\/p>\n
Maximum clock speeds using an air cooler were around the 4.1GHz mark, while -75°C temperatures only managed to improve the clock by 400MHz, which isn’t much when you consider that X9770 CPUs improve by anything up to 1.2GHz. With that said we have no doubt that overclocking on the CPUs will improve with better revisions and as the core gets process improvements.<\/p>\n
We would usually not concern ourselves with professional application performance when reviewing a CPU, however, the i7 shines here more than it does in the game benchmarks. This is particularly important for Intel, as the competition was still beating them in specific server space tests, but with the Nehalem architecture, Intel has improved significantly in this space, and as a result, we as the consumers have gained performance in our favourite games and programs.<\/p>\n
Tech Specs<\/strong><\/p>\n Core: Nehalem<\/p>\n Number of Cores: 8 (4 physical + 4 logical)<\/p>\n Process: 45nm High-K<\/p>\n Cache: 4 x 32KB Instruction Cache, 4 x 32 Data Cache, 4 x 256KB L2 Cache, 8MB L3 Cache<\/p>\n Frequency: 3200MHZ<\/p>\n