Win32 commit c874975: fixed compute 2.0 and added compute 5.2
CUDA 6.5 build, compute 2.0, 3.0, 3.5, 5.0 and 5.2. No idea if compute 5.2 helps the new 900 series cards in any way, good luck. You'll need fairly recent drivers due to CUDA 6.5.
Win32 binary for commit ddb355a
Compiled with CUDA 6.0 for compute 2.0/3.0/3.5/5.0, cuda runtime included in the zip. Minor update, cleaned up the compute 2.0 version a bit and included Wolf's keccak implementation.
Win32 binary for commit 96b2ced
Should give a 15% or so boost for at least some compute 3.0+ cards, Maxwell and 2.0 cards unaffected.
Win32 binary for the latest commit
Major improvement to interactivity, see README.txt on --bfactor and --bsleep. Should also further alleviate TDR issues due to the major time hog of the algorithm being broken down into smaller launches.
Experimental release, needs testing on non-Maxwell compute 3.0+ hardware
Results from testing decide whether I release the latest changes or roll back.
Windows binary, commit 44bc11e
With the other algorithms removed and my algo not requiring the Kepler shuffle (at least not yet) I've enabled compute 2.0 builds. This release only adds compute 2.0 support in addition to 3.0 and 3.5, no need to update if you're running the previous release already. Compute 2.0 performance is a mystery, no idea what to expect.
Windows binary, commit 72f35da
Compute 3.0/3.5, +18% on a 750 Ti compared to previous release.
Windows binary, commit 78a196f
32-bit Windows, compute 3.0 and 3.5. Major overhaul under the hood, doesn't show much on the outside. Biggest change would be splitting the cryptonight kernel into smaller pieces. This should help with the Windows TDR problems by releasing the GPU for a short while between the steps through. Also gained an unintentional 3-4% boost to hashrate, woop...
Windows binary for commit dd83f59
32-bit Windows binary with compute 3.0 and 3.5 support, see commit for details on what's new.