Thinking of upgrading the CPU/mobo in my desktop. Machine is mostly used for work* and gaming. GPU is pretty decent and it has an SSD for the system drive, so weakpoint is the CPU.

Right now I'm looking at an i5 4670K, but was wondering should I wait for the upcoming haswell refresh? Even if the new chips aren't great, it might mean a price drop in the 4670K.

Any advice, thoughts, opinions?

* I use this machine to allow me work from home when it suits me, but work aren't going to pay for it. Work involves C# and C++ development and I frequently have to build a legacy C++ app. I'd love more cores, but CPUs are so expensive in NZ, I really cant justify dropping NZ$1500+ on a CPU for an i7 extreme.

If you're talking the straight-up haswell refresh (Devil's Canyon), they aren't expected to be a great improvement. You'll get about a 100Mhz clock increase under boost, but that's basically it. No new instructions no new features. I might run a little cooler or with lower power. They're using a non-shitty thermal interface material inside the package, so that might buy you a little more thermal headroom for longer boosts.

If I were you and you're in no rush, I'd sit awhile and just keep an eye on a good price for the mobo/CPU combo you're looking at. Strike when there's a good sale on either the current or refreshed CPU.

If you're talking about Broadwell, which I think is supposed to be out in the fall or possibly later, its a tougher sell. If I were building a new PC I'd wait, but since you're upgrading I assume you already have DDR3 and don't want to drop money on that too. Plus, its a new platform which means limited options and high prices for the honeymoon phase, on top of waiting all that time.

* I use this machine to allow me work from home when it suits me, but work aren't going to pay for it. Work involves C# and C++ development and I frequently have to build a legacy C++ app. I'd love more cores, but CPUs are so expensive in NZ, I really cant justify dropping NZ$1500+ on a CPU for an i7 extreme.

CPUs priced over $240 offer rapidly diminishing returns when it comes to gaming performance. As such, we have a hard time recommending anything more expensive than the Core i5-4670K, especially since this multiplier-unlocked processor is easy to tune up to 4.3 GHz or so with the right cooler. Even at stock clocks, though, it matches or beats the old $1000 [? Traditional cell phone cost per year] Gulftown-based Core i7-990X Extreme Edition in our benchmarks.

We have seen a small handful of titles benefit from Hyper-Threaded Core i7 processors, though. Because we believe this is a trend that will continue as developers optimize their software, we're including the Core i7-4770K as an honorable mention, now selling for $340. In a vast majority of games, the Core i7 won't demonstrate much advantage over the Core i5. But if you're a serious enthusiast who wants some future-proofing and values highly-threaded application performance, this processor may be worth the extra money.

I don't know where Mr. Ravyne got his information

It all makes sense if you comprehend the question. Chaos specifically implied that he didn't want to spend a ton of money on a new CPU, and the model he says he's looking at is a quad core i5 that runs $200 USD. The Haswell refresh is Devil's canyon -- simply an improved Haswell, not Haswell-E. Haswell-E or Xeon are worthwhile to content-creators or very heavy multi-taskers, but more than 4 cores is of little benefit to a programming and compiling, unless your job involves writing very parallelizable code.

I tend towards consumer-grade i7s myself. Hyperthreading is good for a 15% boost or so in overall performance and that's about as good of a cost/benefit ratio as you're going to find on the higher end of the market. But admittedly, I'm also the type to build the computer I want up front and drive it for 4 years or so; I'll usually do a GPU upgrade after 2-3 years, but that's the only component I replace, save defect. Hyperthreading is basically all the processor state for an extra thread per core that can be switched to quickly when its alternate stalls awaiting IO, or even when there are simply untapped execution resources when there isn't enough instruction-level parallelism in one thread to keep the CPU fed. The OS could do this in software, but having it built into the CPU allows it to jump in more quickly, and also to be transparent to the OS -- this is why they call hyperthreads 'logical cores' (although, really, its more accurate to say that in a hyperthreaded implementation that there are two logical cores sharing one physical core, Sparc's Niagra and friends actually have 4 or more logical cores sharing a single physical core).

Thanks for all the replies. Electricity in NZ is about US$0.17/ KWh (varies with night time usage, etc). It's not really a huge concern.

I don't know where Mr. Ravyne got his information, but he's quite wrong; the Haswell-E processors will feature 6 and 8 cores, for up to twice the theoretical performance.

Where did you see that? As far as I've read, the only CPU that will get a core count increase is the i7 extreme.

A xeon definitely isn't worth the price increase.

I tend to view the i7 Extreme thing as a rip-off at the best of times. I quote this month's CPU price/performance breakdown from Tom's Hardware:

Yeah, I'd read that article myself. Which was why I was planning to stick with the 4670K. Still... I can dream :D

but more than 4 cores is of little benefit to a programming and compiling, unless your job involves writing very parallelizable code.

It appears you seem to be implying there is a link between the degree of parallelizability of the execution and compiling of a piece of code; but there is none that I know of. Afaik, msvs makes great use of additional cores during compilation (although you may have to enable it in the options for C++). Compilation involves a lot of steps which are trivially parallelized.

Yep, and that's the main reason I'd like a hexa or even octo core. One of my main work projects involves some legacy C++ code. The thing's a bloody nightmare, coupling is through the roof and the whole thing is in one giant lib that needs to be recompiled every time I change something in the C# it relies on. I'm slowly refactoring it to be more sane, but in the meantime, more cores = faster compilation. When I started here, I enabled parallel computation and the compile time quartered from 40 mins to 10. Yes, it's really that bad. :(

I've decided I'll wait until next week and see what the pricing does. Even if the new haswell refresh doesn't offer much, at the very least we should see a small price drop in existing models.

Cheers all.