Tests
Our test method is briefly described here. The scores on diagrams are relative to that of our reference testbed that always scores 100 points. As of 2012, it's based on the AMD Athlon II X4 620 CPU, 8GB of RAM and Palit's NVIDIA GeForce GTX 570 1280MB. Detailed (absolute) results are traditionally provided in this summary.
3D Modeling
In this poorly threaded group of tests the two new Core i7s are almost equal. Core i7-3770K is even a bit faster due to its improved microarchitecture. The remaining processors yield noticeably. These benchmarks are not intended for multicore processors though.
Final 3D Rendering
On the contrary, this group of tests can utilize any number of cores. As the quantity precedes the quality, Core i7-3770K yields to the older 990X. Obviously, there has been nothing new in the top segment since Core i7-980X, because Intel doesn't want to pull CPUs with 8 and more cores down to the single socket segment. Moving from Gulftown to Sandy Bridge-E gives 10-15% more productivity, and the coming Ivy Bridge-E is unlikely to show better progress (if 6 cores remain the most it can have).
AMD's progress is significant. The latest microarchitecture will probably be improved with the software optimization (AVX and especially FMA4 should make up for the decreased number of vector blocks compared with the high-range Phenom II X6). Still, the results are even worse than those of the top LGA1156.
Data Compression/Decompression
Under integer load the small number of vector units doesn't lower the scores. The disadvantages of the new microarchitecture here are the slow cache and bad single-threading performance (the latter becomes critical in decompression tests). As for Core i7-3970X, it leads the way with its 4 GHz and 15 MB of full-speed cache. Yet the cheaper CPUs are not that far behind.
Audio Encoding
Both hexacore Core i7s are the best here, but the 990X old-timer excels the i7-3770K and FX-8350 duo just a little.
Compiling
Compiling tests are thread and cache-demanding at the same time. They require a large and fast L3 cache (large especially), so the pair of hexacore Core i7s is the best again with only 10% difference between them (which is rather small for 2.5 years of development). However, when you have no competitors, there is no demand for a breakthrough—true story for high-end CPUs.
Mathematical and Engineering Computations
Performance in this single and dual-threading group of benchmarks can only be improved by increasing clock rate as yet, so Core i7-3970X brings the first prize back to Sandy Bridge-E (which it has formally lost to Ivy Bridge).
Write a comment below. No registration needed!
