I have an update on a couple of issues that have existed for a while but are now better understood.
1. The definition of AES
The miner's AES optimized code also includes AVX instructions that are not included in the first generation
of AES CPUs, however the miner's CPU capabilities check uses the pure drefinition of AES (without AVX).
This results in the following message being displayed on startup on CPUs with AES but not AVX:
Rebuild with "-march=native" for better performance.
This is not an error, the miner will work, but not at AES performance levels. This message should be ignored
for Nehalem series CPUs and similar AMD. These CPUs can't mine using the miner's AES & AVX code.
I other words the miner will mine at the best rate possible for that CPU in spite of the suggestion otherwise.
The minimum requirement for AES mining will be changed to Sandybridge to conform with the actual limitations
of the AES mining code and the capabilities check will be changed to also check for AVX support in the CPU.
Update: I have changed the terminology to be more precise. It will now be called AES-AVX1 to more clearly
exclude CPUs that have AES but not AVX1. Also the rebuild warning will not be displayed for those CPUs.
Documentation will also be updated.
2. Low difficulty shares mining hodl
Since v3.2 hodl mining has started producing low difficulty shares at a rate of 1 or 2%. Prior to that there were
virtually no rejects. I may have found the problem and am testing a fix. At this point the first 170 shares have
been valid.
Update: I am optimistic the fix works. The first test run produced no low diff shares in 268 accepts
3. Hodl support on Windows for non AES-AVX1 CPUs. This is a more challenging issue and will not be fixed
in the next release.
I will build a new release when these 2 issues are solved.
In the meantime users should ignore the warning to rebuild for faster performance and hodl users with Linux
can use v3.1.18 if the rejects are unacceptible. Hodl miners n Windows will have to live with the rejects until
the fix is released.
Note to Cryptimining Blog: You might want to drop the Westmere build for the next release as it seems to have
no practical use. A core2 build would expand the range of supported CPUs and would also work on Westmere.
Update:It appears both issues are fixed. The fix to hodl was more of a trial-and-error proceess so I want to test longer
to raise the confidence level. I don't know why it broke and I don't know why the fix works.
The AES issue, on the other hand, is now well understood so I am confident in the fix although I don't have the
resources to test it properly.
An interesting contrast in processes.
V3.3.2 coming in a matter of hours.
