Rendering, CAD, large database / spreadsheet manipulation and video editing are areas most would agree that memory speed matters..... In gaming it's a point oft argued it doesn't matter but there's enough reviews around which point to the contrary and there's plenty that support the argument.
I have oft posted these two references to when I see posts saying memory speeds and timings don't matter.
22.3 % (SLI) increase in minimum frame rates w/ C6 instead of C8 in Far Cry 2
18% (single card) / 5% (SLI) increase in minimum frame rates w/ C6 instead of C8 in Dawn of War
15% (single card) / 5% (SLI) increase in minimum frame rates w/ C6 instead of C8 in World in Conflict
The argument was then that SB / IB were not so affected because this article showed just 1 fps changes.
Well when it's 22 and 23 fps, that is a 5% increase in min. fps
The answer I usually give is "sometimes it does, sometimes it doesn't"
But Anandtech has just done an article comparing 1333 - 2400 speeds and it shows that .... at least for those for those looking to not invest in discreet GFX cards till later when budget allows, yes, memory speeds do matter, even on SB / IB. Gamess selected for the test ranged from those that were obviously unplayable on IG to those that ran well (Metro 2033, Civ 5, Dirt 3, Portal 2 and Batman AA)
For all but the DDR3-1600, the cost increase is greater than the performance increase..... however, the faster memory makes ya whole system go faster, so system cost should be taken into consideration:
The following lists the ROI (return on investment) on a systems of various costs. In the example in bold text, DDR3-1866 costs ya 26.67% more than DDR-1333 but gets ya a 12.73 % performance increase for an increase in system cost on a $1200 box of just 1.67 % which gives you an ROI of 7.64 to 1.00
Memory ROI $1200 Box
DDR3-1600 19.18 DDR3-1866 7.64
While each step up obviously gets harder to justify, no argument can be made that the increase in system cost isn't worth the return on performance. In every single case, the increase in performance is larger than the increase in cost with the worst case being almost 3 to 1. To my mind, DDR3-1866 looks to be the sweet spot for those in this situation.
The article does makes mention of discreet GFX card performance:
Graphical performance is all about vector calculations - moving data from memory to the compute units for calculation then placing it back out again where required. High end graphics cards do this quite well, with the high end NVIDIA GTX680 video cards achieving a rated bandwidth of ~192 GB/s.
Obviously that bandwidth is dependent on the card's abilities which in turn is based upon card cost. The logical follow up to this article then is redoing the tests with cards of varying performance and seeing just where the break points lie. With the 680, they mentioned, I'd expect small differences ..... with a 640 tho, I'm curious.
Nothing that most of us here recommending RAM to someone didn't already know, really.
I'll still stick with my general way of thought, though. For gaming with a discreet GPU, I'll still recommend no faster than 1600 (the percentages may be better for faster RAM, but they don't sway my own personal opinion that it's still not worth it), and for gaming with an iGPU I'll recommend as fast as the person can afford.
For memory intensive work, it'll be a case by case basis because sometimes, the amount of RAM matters more than pure speed and vice versa.
For someone that doesn't game OR do much memory intensive work, 1333, or even 1066, is enough.