Originally Posted by TheLAWNOOB
I'm pretty sure a single big core is as efficient as many small cores.
Idle power is 100W. Similar power per core. 7870 1280 core, 7970 2048 core.
Big cores might be more expensive due to more failure rates though.
Well, my school bus is cancelled which means I have more time to work on my essay and chem. Yay -_-
Not really, you need more smaller yet faster/more efficient small cores for optimal performance at memory hard mining algos, specially with the way lookup gap implementations work.
That's why, mainly a software thing.
Originally Posted by DiGiCiDAL
Sorry to disagree, but I'd love to hear the reasoning behind that statement.
Actually a single complex core is almost never
as efficient as many smaller cores. This holds for nearly any
compute job. The only exception to this is in dealing with extremely non-optimized code. If the job is logically impossible to calculate out of order or multi-threaded then having the most complete command set residing in a single core running at the fastest possible clock speed will indeed be faster. On the other hand, if the code is simple and easily optimized to enable out of order executions on multiple threads... then you will not only get there faster with multiple, slower cores - but you have the added benefit of being able to be much more efficient power-wise as you can effectively shut down unused cores as the job allows.
In the case of mining - this is exactly
why ASIC miners beat out GPUs by an order of magnitude (or more if you count efficiency per watt)... because you can have in the same watt-footprint 20-40 ASICs all working independently as opposed to a single GPU which is only using a fraction of it's transistors and pathways to perform the same calculations.
Sorry but there was definitely a reason for CISCs over-taking RISCs in PCs and GPUs... because we started requiring them to do way more diverse tasks than simple floating point calculations and address lookups. There is likewise
a reason that multi-core CPUs and soon GPUs as well are no longer restricted to high-end server, but are found even in smartphones, etc.
Originally Posted by arcade9
what are you mining now mate?
Yacoin, Protoshares and Ultracoin, depending on which performs best on each piece of hardware I own or manage.
Originally Posted by TheLAWNOOB
I understand what you are saying, but GPU cores are highly modular. A bigger package just means more CUDA or stream cores in it. Sorry I should of said bigger dies or packages instead of saying bigger cores.
The way memory hard algorithms like scrypt-jane work, it makes it so that you run out of memory for your threads if you run many threads per device with high intensity.
This implies lowering intensity on high end cards, or having to halve memory requirement per thread and trying to run more smaller threads per device.
Ends up causing you to have to rely on intensity fine tuning, something which is not really optimal with standard cgminer forks (except for the one based in kalroth's cgminer which has raw intensity incorporated in the code by Thirtybird, yacminer 3.4.0.; or cudaminer which shines at memory hard algos more than scrypt, mainly because fast scrypt nvidia cards so far consume way more than similarly fast amd cards->say GTX 780ti, can get to 900kh/s but it'll draw like 400w doing so and need water cooling whereas a 290 can do the same thing at 300w)
If scrypt asics come, mining scrypt will be kinda pointless eventually and I wanna be prepared. Nvidia might be able to compete if the higher end Maxwell cards can indeed match or beat AMD at power efficiency mining scrypt, but they will still probably cost a lot and/or suck/don't beat lower end cards by too much when it comes to memory hard algorithms like yacoin's scrypt-jane or Vertcoin's variable N factor Scrypt.