A Sandy Bridge to Haswell upgrade won't blow your hair back

Credit: Image credit: Wikimedia Commons

I tore my main computer apart just to replace a Sandy Bridge chip with a Haswell. Here's what I got for the effort.

Several years ago, Intel introduced the tick/tock model, where the "tick" would be a shrink of the existing architecture and "tock" would be a whole new architecture using the manufacturing process from "tick." This was done to spread out the risk. Trying to do a new architecture and manufacturing process shrink was too much risk to be done in one fell swoop.

It's been my personal policy to upgrade my computer on Intel's "tocks" -- when it comes out with a whole new architecture. There is no performance gain to be had on a "tick" because it's just a drop in the size of the transistors they use, such as from 45nm to 32nm.

This summer came "Haswell," the tock that featured a new architecture. So, thanks to a courtesy chip from Intel (a Core i7-4770K, a 3.8Ghz quad-core processor), I went shopping. My existing case would house the current PC, a Sandy Bridge chip with the model number Core i7-2600k, a 3.4Ghz quad-core processor. I needed a new motherboard (obviously), case, and power supply. The memory (16GB, 1600Mhz DDR3), storage and GPU (Nvidia GTX670) would all be transferred over.

I had hoped that I could get away with using my existing OS install, but no dice. It crashed immediately on boot. A reinstall consumed a day of my time as one thing after another came up (I need Photoshop, where's the CD? Oh yeah, in a storage unit 12 miles away).

When all was said and done, I noticed something: the computer was no different than before. I mean, usually you see some kind of change. Starting up a new PC up always gave you that hair blowback experience where you say "wow, that sucker is fast." That didn't happen.

It came down to benchmarks, and when you need benchmarks to determine a difference, well, you didn't gain much. I benchmarked my existing SB system against the Haswell. Both can be overclocked, but I chose not to. The boot disk is an OCZ Vertex 4 512GB drive, memory is 16GB of DDR3 1600MHz memory and the GPU is an Nvidia GTX6700.

After benchmarking the SB system, the storage, memory and GPU went into the new Haswell machine. So the only thing new is CPU and motherboard. In both cases I used Windows 7-64. Nothing was running, just basic Windows.

To add a little diversity, I dug out an old Core i7-860, a Nehalem-generation chip from 2009. However, it could not, or would not, run certain tests.

So here are the stats:

Well that's not very impressive. The Core i7-860 was left out because it's an 8GB machine booting a regular hard drive, so that comparison would be by no means fair.

On to the next test. I used HandBrake, a popular media conversion tool. It would convert "Jason Becker: Not Dead Yet," a 1 hour, 26-minute file currently in XviD compression. The file was converted to MP4 format, without using any GPU acceleration.

Now we're seeing results. The Sandy Bridge to Haswell leap is not great, but from Nehalem to Haswell you get a near doubling of frames per second processed by HandBrake.

Test #3: MediaConverter 8, converting a XviD-encoded rip of the live action adaptation of "Space Battleship Yamato," running 2 hours, 18 minutes. By the way, all you "Superman" fanboys crying about Zack Snyder's movie not following the series canon? Your cries fall on deaf ears. Talk about a disappointment...

Anyway, MediaConverter 8 was to convert the film to QuickTime format, no GPU acceleration. MC8 will not let you use more than 20% of the total CPU for the conversion, which is odd. And it absolutely would not load on the Nehalem system despite two hours of fighting with codecs.

Ok, shaved five minutes off a 138-minute movie conversion. Not bad, but not great either. The final test is PCMark 7, from Futuremark.

Here again we see the real jump is from the Nehalem generation, while the Sandy Bridge to Haswell jump is not as pronounced.

Conclusion:

It seems abundantly clear that there is not a whole lot of CPU performance gain to be had in going from the previous architecture to the new one, but if you have anything older -- Westmere, which was the Nehalem shrink, on down -- you will see appreciable gains. You can argue that I only did a few tests, but Tom's Hardware Guide did their usual batch of OCD testing and came to the same conclusion.

If you have anything older than a Sandy Bridge/Ivy Bridge, which means a PC a year or two in age, you won't gain much by upgrading. Owners of older systems will notice an appreciable gain, however.

Andy Patrizio covers all the chip news you can use, from benchmark leaking to dropping DRAM prices, and everything in between on ITworld's Chip Shots blog. Follow Andy on Twitter (@apatrizio) and Google+. For the latest IT news, analysis and how-tos, follow ITworld on Twitter and Facebook.

From CIO: 8 Free Online Courses to Grow Your Tech Skills
Join the discussion
Be the first to comment on this article. Our Commenting Policies