- Monday , November 14, 2011
- Kyle Bennett
Intel Core i7-3960X - Sandy Bridge E Processor Review
Intel debuts its $1000+ Extreme Edition 3960X processor parroting how great it is for the gamer and enthusiast. With 6 cores and 12 threads, a new motherboard and chipset platform, and quad channel DDR3, Intel has done the impossible, given us everything we don't want, and nothing we do want.
Clock per Clock - Thread per Thread
In talking with our readers many of you voiced wanting to see IPC (Instruction Per Clock) demonstrated. Needless to say, when you are working with 4 core, 6 core, and 8 core processors, some with HyperThreading enabling computing up to 12 simultaneous threads at a time, and all of those core clocks scaling dynamically, the IPC perspective is easy to lose in translation. Honestly, given all the real world variables involved, IPC is somewhat less important than it used to be in the days of single threaded applications and single cores. Still we understand the enthusiast wanting to get a handle on IPC though, and personally so do we.
What we have devised below are three simple test scenarios that are single thread / single core aware. All of these tests were run with the processors locked at 4GHz CPU clocks with 1333MHz memory clocks. The Core i7-920 is our only multiplier locked processor so it is using an accelerated processor bus speed as well as having a memory speed advantage. (20*200MHz=4GHz / 1603MHz) The rest of our processors are unlocked and allow easy scaling of the processor multiplier in order to reach the targeted 4GHz.
Hiper Pi, is simply a GUI front end for Super Pi. Super Pi is a utility that allows you to calculate between 16 thousand and 32 million decimal places of Pi. In the benchmark below, we have timed how long it takes each processor to calculate out to 1,000,000 decimal places of Pi.
What we see here is not really surprising. We see our 3960X trailing both the 2600K and the 2500K in terms of IPC. Adding the two cores to the 3960X surely introduces some latency even given the huge 15MB of shared cache. Thankfully though we see the 3960X trump the LGA 1366 990X and model 920 processors in terms of IPC.
Cinebench 11.5 is much more like a real world content creation workload. Cinebench renders a 3D scene. It also has the option to run the benchmark in single core mode. This is one that would be easy to use on your own system and see where it stacks up.
Again we see much the same scaling as we do above in this very different workload. The 3960X does scale much more in line with the other two Sandy Bridge processors and shows a lot of promise when looking at the scaling against the older LGA 1366 parts.
Finally we are using a real world application that is very reliable in giving repeatable results, and that is LAME's MP3 encoder. Below we are simply timing how long it takes to encode a 1 hour long song in .wav format to .mp3 format. We use LAME's default settings.
And again we see much the same scaling in place where our new Sandy Bridge E part is a bit slower than the much less expensive 2600K and 2500K enthusiast parts.