SANDRA - Multi-Core Efficiency
Inter-core efficiency is a very important part of execution. Sandy Bridge doesn't always need to store data on the RAM. That's a big reason why Intel Quick Sync is so fast. Here, we see just how efficient the cores communicate with the L2 cache features.
There's a slight latency increase on the 3960X yet it enjoys about 9GB/s more inter-core bandwidth. That will come in handy for things like encryption, compiling and transcoding. The shocker here is the 8150's disappointing performance. I wasn't expecting it to be this low.
SANDRA - Memory Bandwidth
Remember, X58/Gulftown systems utilize three memory modules at once. X68/Sandy Bridge utilizes two memory modules at once. Let's see how X79/Sandy Bridge-E does with its four modules at 1333MHz and 1600MHz frequencies.
The 2600K continues to show off its great integrated memory controller getting close to the matching the 980X at stock settings. The FX-8150 does quite well here but what good is all that bandwidth if it can't efficiently use it as we saw above. The 3960X cranks out some serious bandwidth at 1333MHz and even more at the CPU's supported 1600MHz.
Cinebench will put all the previous synthetic CPU features to the test. The more efficient the CPU, the better the scores will be. Keep in mind that the CPU is doing the bulk of the work.
It's interesting to see the 2600K come out on top in the Single test. That's because the CPU always seems to be a more aggressive given there's less architecture to throw in the mix. However, cores will win out every time in the Multi-Core test where the 3960X breaks the 10.25 barrier without being overclocked. Nice!
Now, Sandy Bridge-E is suppose to have a more aggressive Turbo frequency when using 1 or 2 cores where it reaches 3.9GHz. No matter the settings or BIOS file, I could not get the X79 test system to reach and sustain that frequency at will. And, when manipulating the Turbo BIOS options, the board would double boot indicating that something isn't quite right or ready for prime time.
Like Sysmark, PCMark7 runs a much shorter series of tests that utilize each test system's entire resources. It basically gives us a better real world test result compared to running the Windows 7 own test. Let's look at the results.
I wasn't really expecting the 3960X to walk all over the 980X and it didn't. In this benchmark, the number of CPU cores and test components directly influence the results. Since both are hex-cores, the results makes sense. The poor FX-8150 just falls flat compared to the Core i7 series.
Media Espresso 6 - 450MB 1080i Transcoding
I wanted to see what ME6 can do with the extra memory bandwidth and revised cores. Here, a 450MB 1080i video is transcoded to fit the iPhone 4 with hardware acceleration turned on and off. The time is listed in seconds.
Naturally, the FX-8150 keeps within range of the 2600K. There's a slight performance gain in that the 3960X shaves off a few seconds compared to the 980X. The revised silicone in the new hex-core seem to be handling the task a bit more efficiently.
Again, the FX-8150 and 2600K trade spots at stock frequencies. The revised Sandy Bridge-E 3960X is a bit more efficient at this task versus its 980x cousin. Given that their Turbo frequencies are so close, a near 2,000 point gain is a worthy revision.