New PC theory crafting

@zx0 Didn't know CPU determined the number of lanes! Whole different type of sport compared to programming: hardware knowledge

If I am not mistaken number of lanes used to depend on chipset, but since 2011 and 1155 it depends on CPU.

Just to put a counter argument on the GPU side, @LennardF1989 are you considering any GPGPU type work (even just for experimenting)? If you are then I *highly* recommend looking at an AMD GCN based gpu rather than anything based on either Kepler or especially Maxwell (GTX 900 series). The 900 series is very efficient *if* your sole purpose is to game, however to *achieve* that efficiency Nvidia stripped almost all the GPGPU related logic out of the core (strange given they championed it in the first place). A GTX 580 or anything GCN (even a lowly R9 270 which costs nothing) can mop the floor with Maxwell for compute. Also worth noting is the 900 series has a very narrow memory bus width compared to other modern cards. I'm really not convinced it's as a good a buy as most people make out- it's performance lead vanishes as resolution goes up....

If your looking at AMD, the best cards for GPGPU (without going FirePRO) are the R9 280X / 280 (based on the older Tahiti uArch), although the more modern Hawaii R9 290X / R9 290 aren't too bad (they have roughly 1/2 the DP rate of Tahiti but are otherwise faster in SP GPGPU and general gaming they're about 30%+ faster). The main argument against say a 290X vis a vis the GTX 980 is the 980 is a bit faster at 'standard' resolutions (e.g. 1080p, higher it's less definitive) and uses quite a bit less power (250W for the 290X vs 190W) *however* in it's favor you do have a 512 bit memory interface on the 290X (compared to 256 bit width on the 980, albeit with faster memory it still has significantly less bandwidth to play with) and a significant lead in SP and DP gpu calc work. I guess it just depends if you see yourself experimenting with using GPU accelerators at all? Also I personally rate the 290X as a better option for very high resolution displays (i.e. 4k).

Others have mentioned there are new GPU architectures on the way, whilst this is true I think the 'wait something new is coming' argument is a bad one for PC hardware- i mean there is *something new* from each manufacturer every few months, so you can just wind up waiting forever following that philosophy.

Requesting source on Maxwell having worse compute. AFAIK, only DP was cut down further. http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/20

Nvidia's Maxwell (GTX 970 & GTX 980) is much, much slower at double precision and is memory bandwidth constrained compared to AMD Hawaii (R9 290 & R9 290x), so yes if you're doing compute that only needs single precision and you aren't doing massive image or data processing heavy tasks Maxwell will be faster. It's worth noting that Nvidia's own new compute specific hardware, the Nvidia Tesla K80, is still Kepler based (770, 780, Titan).

It's also worth noting that the R9 290 and R9 290x are way cheaper than even the GTX 970 when it comes to euro prices... The price disparity in the US isn't as extreme (in part because the R9 prices are still kind of jacked up from Bitcoin farmers).

Also, don't do dual GPU. If you have interest in VR, regardless of what Nvidia or AMD marketing says about dual GPU being good, all VR stuff forcibly disables multi-GPU because it adds extra frames of latency which are bad, bad, bad.

Ah you are correct, it's good at sp (better than I remembered) although I'm not a fan of them strangling dp so much. I mean it's a good card, though I think it's a hard sell against 290x that costs less. I mean the lower power consumption is the selling point I suppose but since when were gamers so hung up on efficiency? Now when a fully leaded 250w 'big' Maxwell gpu is launched then I think it will be interesting.