GPUs are ubiquitous. When you have optimizations of hashing algorithms, like for scrypt and sha1 before it, you end up with specialized hardware often coming out of a few hardware companies dominating the entire mining scene.
That means a stark concentration of mining power, because it means "average joe shmoe" can't just start running cgminer on their integrated openCL hardware and get coins at reasonable rates for the money invested (in power).
I think it is too late for litecoin, and would rather see them implement this in a less entrenched altcoin like peercoin, because there is too much inertia and scale now to change it in the litecoin space.
I understand and agree with, to some extent, the argument against centralisation. However, I don't see how this fork solves it - it's switching from one form of centralisation (ASICs) to another (huge clusters of mining rigs).
GPUs may be ubiquitous but it's not viable to run a mining rig with just one, which means that all those GPU miners are basically running "specialised hardware" too. Joe schmoe does not just happen to have an 8-GPU tower computer lying around.
I run a 2-GPU tower and make a couple bucks a day with zero upkeep effort on my part.
It's not totally vanilla hardware like the 15-watt laptop I'm typing this on, but it is common enough I happened to have it already.
With ASICs the barrier to entry is five figures. (Most of the cheaper ASICs have terrible payoff timelines) With GPUs the barrier to entry is far, far lower. You won't make it rich on a $150 GFX card, but you can make some revenue.
My Cuckoo Cycle PoW (https://github.com/tromp/cuckoo)
configured to require for instance 7GB of memory, is
all of the above. And is very power friendly, since
only 5% of the runtime is computation and 95% is waiting
for main memory random access latency...
Sure, if you can come up with a good proof-of-work algorithm that runs well only on CPUs.
It could be possible. The straightforward way would be, I believe, to use an algorithm that has weak parallelism and depends strongly on branching. CPUs are king of branching, and both GPUs and ASICs get ahead by taking advantage of parallelism.
Although, one nice thing about GPUs being the standard-bearer, is it makes most general purpose server hardware useless for mining. That's good in that it makes hijacking a webserver to mine bitcoins, a largely pointless affair.
That means a stark concentration of mining power, because it means "average joe shmoe" can't just start running cgminer on their integrated openCL hardware and get coins at reasonable rates for the money invested (in power).
I think it is too late for litecoin, and would rather see them implement this in a less entrenched altcoin like peercoin, because there is too much inertia and scale now to change it in the litecoin space.