I've always wondered what the required computational power for generating vanity addresses is. Could anyone point me in the right direction for this info? Thanks.
=snip=
Prefixes are also very fast to search, and a list of thousands of prefixes may be specified with little or no reduction in key search rate.
Regular expressions are programmable pattern filters. They are very powerful, and can be used to match prefixes, suffixes, varying-length sequences, etc.. For a quick tutorial, see
http://www.regular-expressions.info/quickstart.html. To enable regular expressions, use the
-r command line option. Unfortunately, regular expressions are very slow, and will have a significant impact on key search rate. Because of this, regular expressions should only be used if their expressive power is needed.
Oclvanitygen is only effective at searching for prefixes. Regular expressions will not work effectively with oclvanitygen, as oclvanitygen is currently unable to execute the regular expression on the GPU.
What key search rate can I expect from hardware X?Detailed list forthcoming. Some ballpark estimates are listed below.
Dual-core desktop CPUs, 32-bit mode: 100-250 Kkey/s.
Dual-core desktop CPUs, 64-bit mode: 150-450 Kkey/s.
Quad-core desktop CPUs, 32-bit mode: 200-400 Kkey/s.
Quad-core desktop CPUs, 64-bit mode: 300-750 Kkey/s.
As vanitygen performs a lot of large integer arithmetic, running it in 64-bit mode makes a huge difference in key search rate, easily a 50% improvement over 32-bit mode. If you are using a 64-bit edition of Windows, and not using a GPU, be sure to use vanitygen64.exe.
In custom builds, CPU performance will be less than expected if the OpenSSL library is an older version (<1.0.0d) or is not built with the appropriate optimizations enabled.
General formulas for expected performance on GPUsNVIDIA GeForce 96xx, 98xx, GT 1xx, GT 2xx, GTX 2xx (G90/GT200):
Key/s = (CUDA Cores) x (Shader MHz) x 17
AMD Radeon 58xx, 59xx, 67xx, 68xx (VLIW5):
Key/s = (Stream Processors) x (Core MHz) x 20.1
AMD Radeon 69xx (VLIW4):
Key/s = (Stream Processors) x (Core MHz) x 13.6
AMD GCN, NVIDIA Fermi/Kepler: Please contribute some numbers!
Unfortunately, AMD VLIW4 does not perform as well as VLIW5 with the same number of cores/clocks. Oclvanitygen is sensitive to integer multiply throughput, and VLIW5 can multiply concurrently with other operations, whereas multiply consumes all four ALUs in VLIW4. At similar clocks, a hobbled Radeon 5830 will outperform a Radeon 6970.
