Difference between revisions of "Porting/RandomX"
Jump to navigation
Jump to search
JeremyRand (talk | contribs) (Add Comparison to CryptoNight) |
JeremyRand (talk | contribs) (Add note on NX-AES) |
||
| Line 11: | Line 11: | ||
In [https://github.com/madscientist159/xmrig/#power-performance-discussion XMRig POWER Performance Discussion] , [[User:TimothyPearson|TimothyPearson]] says that CryptoNight v8 damaged POWER's competitiveness against x86 by emphasizing integer square root and integer division. Based on looking at [https://github.com/tevador/RandomX/blob/master/doc/design.md#24-integer-operations RandomX Integer Operations] , [[User:JeremyRand|JeremyRand]]'s impression is that neither of those operations is performance-critical in RandomX (which replaced CryptoNight v8 in Monero in the hardfork of Nov 30, 2019). | In [https://github.com/madscientist159/xmrig/#power-performance-discussion XMRig POWER Performance Discussion] , [[User:TimothyPearson|TimothyPearson]] says that CryptoNight v8 damaged POWER's competitiveness against x86 by emphasizing integer square root and integer division. Based on looking at [https://github.com/tevador/RandomX/blob/master/doc/design.md#24-integer-operations RandomX Integer Operations] , [[User:JeremyRand|JeremyRand]]'s impression is that neither of those operations is performance-critical in RandomX (which replaced CryptoNight v8 in Monero in the hardfork of Nov 30, 2019). | ||
So POWER is likely to be competitive against x86 for Monero mining again once the RandomX JIT is ported to POWER. | So POWER is likely to be competitive against x86 for Monero mining again once the RandomX JIT is ported to POWER. | ||
| + | |||
| + | == AES Acceleration == | ||
| + | |||
| + | Unfortunately, RandomX cannot use POWER9's NX-AES accelerator, because [https://www.kernel.org/doc/html/latest/powerpc/vas-api.html#introduction NX-AES is not currently exposed to userspace in Linux]. | ||
[[Category:Ports]] | [[Category:Ports]] | ||
Latest revision as of 21:07, 2 February 2021
Finished
In progress
Comparison to CryptoNight
In XMRig POWER Performance Discussion , TimothyPearson says that CryptoNight v8 damaged POWER's competitiveness against x86 by emphasizing integer square root and integer division. Based on looking at RandomX Integer Operations , JeremyRand's impression is that neither of those operations is performance-critical in RandomX (which replaced CryptoNight v8 in Monero in the hardfork of Nov 30, 2019). So POWER is likely to be competitive against x86 for Monero mining again once the RandomX JIT is ported to POWER.
AES Acceleration
Unfortunately, RandomX cannot use POWER9's NX-AES accelerator, because NX-AES is not currently exposed to userspace in Linux.
