20140422

Intel Core i5-4670 Haswell Review

Released last summer, the latest processor from Intel is known by the codename Haswell.  Following on the heels of Sandy Bridge, and then Ivy Bridge, the Haswell family of chips gets used across the product line in both desktop and mobile parts.

Spec wise, it is an 84 watt part.  It is a quad core part, but no hyperthreaded (the hyperthreaded quad core chips are the i7's).  It has 6 megs of cache, with a base clock speed of 3.4 Ghz, and a turbo boost to 3.8 GHz.  The graphics are Intel HD 4600.  It is based on the 22 nm manufacturing process.  I posted the CPU-Z below to confirm the processor used.





Most previously reported reviews indicate that Haswell is only a little faster than the chips it replaces from the Ivy Bridge family.  Curious to validate this, I ran it against a Sandy Bridge 2500k, which is also a Core i5 with a base clock speed of 3.4 Ghz.


SuperPi is a mathematical calculation of Pi to 1 million digits.  It is a reasonable test of single core performance.  In general, the Intel parts usually perform better than their AMD counterparts on this test.  The 2500k "Sandy Bridge" can complete this task in 14 seconds, and the 4670k "Haswell" got the job done in 10 seconds.  While one could argue that this is a 28% improvement, and points to improvements in single threaded performance, it is only a few seconds faster.

Passmark is a commercially available benchmarking suite that tests a variety of functions of the processor.  I pulled the numbers of their site, and the 2500k gets scored a 7337, and the 4670k is credited a 7794.  This represents a 6.2% improvement, which is disappointing considering that the Haswell is not one, but two families ahead (with "Ivy Bridge" in between).  Still, these are available numbers, but I did not run this benchmark personally, and I wanted to see on another benchmark what the difference really was.


I ran the Fritz Chess Benchmark.  I find this is a good test of the multicore performance, and benchmarked the 2500k against the 4670.  I was quite disappointed to find that the two processors yielded almost identical results.  For the relative speed, the 2500k part gives us a 21.09, and the 4670 gives us a 21.13.  In the kilo nodes per second, the 2500k is at 10124, and the 4670 clocks in at 10144.  These differences are quite insignificant, and represent no significant increase in processing power.

On the one hand, the Core i5-4670 is a formidable processor, with excellent performance from its 4 physical cores.  It turned in the fastest SuperPi score I have ever seen on a system I have benchmarked.  However, for those looking for significant gains on complex calculations from an existing Sandy Bridge or Ivy Bridge system, I would recommend to sit tight, and not upgrade this year.  The Fritz Chess Benchmark confirms that the performance gains have been quite insignificant.

-Jonas


No comments: