Intel Core i7 980X Extreme Six-Core CPU Review

CPU Comparison

The 980X is clearly intended to be a high-end component, as it has everything cranked to the max (and even allows for open-ended tweaking of Turbo settings for overclocking, something that is exclusive to the “Extreme” series from Intel). As such, we are limiting the number of CPUs in this review to just a few of the top-end ones. We included the Core i7 975 Extreme, which is Intel’s other $1000 processor this one is essentially replacing. It has identical core, ‘Uncore’, and QPI speeds so will tell us exactly what we get by increasing thread count by 50%.

Also included are Intel’s top non-extreme Core i7 CPUs. This will tell us whether it’s worth spending $1000 on a CPU where others can be  had for a mere $600. The Core i7 960 is based on Bloomfield (the first Nehalem Desktop process), and is almost identical to the 975 Extreme except for lower core, Uncore, and QPI speeds. The Core i7 870 is the top Lynnfield part, sitting on a P55 dual-channel memory IOH. It has some advantages and disadvantages when contrasted to the 9xx chips, which you can read about in our original Lynnfield review. And finally, we included AMD’s top-end desktop CPU, the Phenom II X4 965 Black Edition. At just $200 it is less than half the price of the other chips, and a quarter of the price of the 980X. But does it have a quarter of the performance? We shall see…

Here’s how today’s lineup looks:

AMD clearly has some catching up to do, and we know that they are game. They are currently working on their 6-core CPU, which should be out very soon. Unlike the 980X, it won’t cost over $1000, so it should make things interesting.

Test Setup

Some test notes that need mentioning:

• Both the 960 and 975 are emulated via a Core i7 965 CPU. For the 960, all that was required were some timing changes – core, uncore, and QPI settings were all lowered to 960 speeds. On the 975, we made use of turbo manipulation to replicate the CPU. The base clock is that of a 965, but the turbo speeds are all identical to the 975. BIOS settings were changed so Turbo mode would be used more readily. Since we are only testing the CPU under full load, all scores should be identical to a retail 975. The only ones that may be different are the SYSMark scores, which we’ll make note of in that section of the review.
• For some unknown reason, our Intel motherboard forces the base clock to run at 135 MHz instead of the normal 133 MHz with the latest BIOS installed. This BIOS is required to run the 980X CPU, so all results using that CPU are increased by 2 MHz times whatever multiplier is used for core, uncore, QPI, memory, etc. As you can guess, the performance difference should be very minor, and we’ll retest this CPU when we visit it again in the future.

Let’s get it on! Starting, of course, with SiSoft Sandra:

SiSoft Sandra 2010

As always, we start out with a look at theoretical performance with SiSoft Sandra. Yeah, I said we would mostly stick to real-world benchmarks, but this program has its use, and remains our one fully synthetic benchmark. It’s interesting to see where performance gains are found, exactly how each feature affects performance, and so on. It is especially intriguing when huge changes are made, such as more cores or special instruction sets.

The latest version is handy in that it aggregates the scores, easily giving us overall benchmarks to compare.

Since the i7 980X is the same as the 975 in every way but 50% more threads, performance on these “pure CPU’ tests should be about 50% higher. And in this case, it is (52.1% to be exact).

And the same thing happened here – 51.22% improvement in performance. The question is, will we see this in real world applications?

Cryptography

This relatively new module measures each processor’s performance in cryptography, specifically their performance in encryption, decryption, and hashing tasks. AMD processors typically do very well in this test. This test supports Intel’s new AES-NI instructions, so the few CPUs that support them will see vastly improved performance:

Well how’s that for a lop-sided chart! The i7 980X takes the new instructions, multiplies them by 12 threads, and simply runs away in this test. It looks like real-time encryption will soon be a reality.

So the synthetic tests tell us a couple things – when nothing more than threads and speed matters, the 980X yields a 50% improvement over the 975. 50% more threads, 50% more performance (0% price increase! *wink*). Also, when AES-NI instructions are used, encryption is stupidly fast.

On the next page, we’ll begin looking at the performance of actual Windows applications, using PCMark Vantage.