CudaMiner kernels

As of the 2014.02.02 release

Kernel	Alias	GPU Family	Compute Capability	registers	scrypt	scrypt-jane	Supports lookup-gap?	Notes
f	X	Fermi	>= 1.0	32	no	yes, high N-factor	?
k		Kepler	>= 3.0	32	no	yes, high N-factor	?	was K <2014.02.02 release
t		Titan	>= 3.5	32	no	yes, high N-factor	?	was T <2014.02.02 release
F	L	Fermi	>= 1.0	64	yes	yes, low N-factor	?
K	Y	Kepler	>= 3.0	63	yes	yes, low N-factor	?
T	Z	Titan	>= 3.5	80	yes	yes, low N-factor	?	max is x24

< 2014.02.02 release	>= 2014.02.02	notes
S b x w	removed
L b x w	removed	use F kernel instead
F b x w	F b x w	no change
K b x w	k 4b x w	must quadruple block parameter
T b x w	t 4b x w	must quadruple block parameter
X b x w	f b x w
Y b x w	K b x w
Z b x w	T b x w

See this helpful post with more info

autotune

By default autotune will use the lower case kernels for scrypt-jane with high N-factor. However the threshold may not be chosen optimally. So please experiment and override the autotune to find which one is actually better.

f|F - Fermi

for compute 2.0/2.1? devices

k|K - Kepler

for compute 3.0 devices

t|T - Titan

for compute 3.5 devices good for higher N-factor scrypt-jane like YACoin

f - Fermi - ported from David Andersen's code

for compute 2.0/2.1? devices

K (was Y) - Nvidia contributed

high-register count kernel for low N-factor scrypt-jane coins

T (was Z) - Nvidia contributed

requires compute 3.5

high-register count kernel for low N-factor scrypt-jane coins

x24 is the limit for this kernel

Strictly speaking the Z kernel shouldn't need support for any -C flags, as a Compute 3.5 device is capable of running any read access through the texture cache...

The Z kernel is good for low N factor only. (scrypt-jane)

I used

cudaminer.exe -q -H 2 --algo=scrypt-jane:MRC -d gtx780 -l Z24x24 -i 0 -o stratum+tcp://mrc.easy-mining.net:3361 -O youwish:youknew

I think -H 0 or -H 1 was screwing you over. Keccak on CPU is too slow for low N factor coins like MRC.

(cbuchner1)

This is due to the new Titan kernel that the Nvidia engineer submitted. S/He stated that the best config for it will be SMX x 24. This is why this setting is preferred. 780 -l T12x24 Titan -l T14x24 780Ti -l T15x24

Please try all -H -C and -m settings before you write the changes off. This is KEY with this release.

(from Morgahl)

TODO

Post from cbuchner1 w/info on Keccak