|
Title: Can some developer tell us what this Vega results mean Post by: Branko on August 16, 2017, 01:09:51 PM How relevant is this test to mining algorithms?
And are these results good, considering its compared to gtx1080, not 1080Ti https://i.imgur.com/nh7Dz3q.png Title: Re: Can some developer tell us what this Vega results mean Post by: QuintLeo on August 16, 2017, 10:03:22 PM A very FEW of those algos are used by cryptocoins - but those (like SHA256 and Scrypt) are algos that have ASIC available, so NOT viable to try to mine them on GPUs.
For comparison, current SHA256 ASIC are measured in GIGAhash/sec, not MEGAhash - and all Litecoin ASIC miners except the early Gridseed Orbs are measured in MEGAhash/sec not Kilohash/sec The LARGE MAJORITY of those algos have nothing to do with cryptocoin mining at all (MD anything, DES, SHA other than 256, etc). I'm NOT sure on Whirlpool offhand, I think that might be one of the components of X11? Title: Re: Can some developer tell us what this Vega results mean Post by: CascadiaCC on August 17, 2017, 12:36:09 AM A very FEW of those algos are used by cryptocoins - but those (like SHA256 and Scrypt) are algos that have ASIC available, so NOT viable to try to mine them on GPUs. For comparison, current SHA256 ASIC are measured in GIGAhash/sec, not MEGAhash - and all Litecoin ASIC miners except the early Gridseed Orbs are measured in MEGAhash/sec not Kilohash/sec The LARGE MAJORITY of those algos have nothing to do with cryptocoin mining at all (MD anything, DES, SHA other than 256, etc). I'm NOT sure on Whirlpool offhand, I think that might be one of the components of X11? SHA-3 is used to mine Nexus (NXS). Title: Re: Can some developer tell us what this Vega results mean Post by: NobodyIsHome on August 17, 2017, 12:48:27 AM Personally, I haven't tried to optimize any OpenCL code since Litecoin back in 2014. I'm sure things are a bit different today.
I'm not a crypto dev, per se, but I have experimented with some things. Algorithms such as SHA-256 and the like demostrate serial execution which is not easily optimized by concurrency or simplification. Benchmarks may represent (to a degree) computational brute force, however, many ASIC-resistant algorithms purposely use techniques which reward high utilization of random-access memory or other CISC-ish behavior. Therefore, SHA-256 is only telling us a piece of the puzzle. There are plenty of comparisons being made between GDDR5(X) and HBM (latency and coherence) which is probably more representative of actual performance with modern coins such as ETH, XMR and ZEC. |
