Moo Wrapper, aka the Distributed.Net RC5-72 project via BOINC, is traditionally one of the strongholds of AMD, though the recent price gouging on the RX series has brought the mid-upper Nvidia cards into direct competition on it at current price points. RX Vega might change that if their price points are low enough, Vega Frontier is THE highest performance GPU on this particular project at this time despite the relatively poor drivers available for it so far.
Hashrate is directly proportional to the product of shader cores times core clock rate, bus bandwidth is tiny, memory speed DOES NOT MATTER to any noticeable degree as the data set + core of the execution code fit into the CACHE of CPUs going back at least to the Pentium/K5 days as well as into the cache on any semi-recent GPU design - it's possibly the ONLY project where an AMD iGPU is competative with some fairly recent discrete cards (abet lower-end ones). Even Intel iGPUs are sorta-decent at it, if you're not pushing the CPU cores on the processor at the same time.
The difference between AMD and Nvidia is pretty close to zero, on cards with the same product of clock rate times shader cores.
The traditional AMD dominance has been due to having more cores at close to same clock rate at a given price point, though Pascal has evened that up quite a bit so far RX Vega looks likely to revert it in favor of AMD again once the consumer Vega cards start shipping.
The top current computer in Moo Wrapper uses a pair of Fury cards (I can't tell if they are Fury, FuryX, or Nano as all 3 identify the same nor is there any way to tell if/how far they are overclocked).
My dual R9 290 machine (with an A10-7860k) is currently the 4'th ranked producer on Moo Wrapper and generates about 9 GigaKeys/sec (bit over 4G for each the R9 cards, bit over 600M for the A10, but I've got that machine set with an app_config.xml that runs 2 instances per card so the reported hashrates are half that per card), for some perspective on what it takes to achieve a high-end Moo Wrapper machine.
A pair of RX 480 or 580 + an A10 should match the keyrate of my dual R9 machine with a MILD overclock, fewer cores but higher clock rate - at quite a bit less power consumption.
As I recall from some testing I did at one time, a stock GTX 1070 will exceed 4 Gigakeys/sec, a stock 1080 will do more like 5.5 Gigakeys/sec.
For comparison a stock FuryX or Nano should exceed 6, a stock 1080 ti should be in the 8-9 GigaKey/sec range, and a Vega Frontier Edition should be right about 10 Gigakeys/sec.
On the low end, my HD 7750 cards (same core count and almost identical core clock to my A10 7860k/7890k iGPUs) also generate a bit over 600Megakeys/sec - despite having DDR5 vs the much slower DDR3 the A10s use.
On the CPU side - pretty much don't bother, the best CPUs manage ballpark 20 Megakeys/sec per core AT MOST. Might be worth messing with if you have a server-type CPU with double-digit cores, though each core is going to be quite a bit less the total might add up to something noticeable.
Feel free to toss questions at me - I've been a D.Net participant for almost 20 years, and a RC5-72 participant almost since it started.