New analysis report ready for BladeBit CUDA performance from Science of Mining #379
Replies: 4 comments 2 replies
-
you can achieve lower GPU wattage by forcing lower core/mem clock My old Quadro M4400 used 50W (267 TiB C5 plots) underclocked 18W Lookup times same. bladebit simulation is weird - my GPU supposed to not even farm such a big farm...yet it easily does ~0.5s lookups. In case of CPU, simulation may work. AMD Ryzen 5 4500 had lookup times ~10s with same farm size modern RTX GPUs may be even more efficient than old Quadro GPUs. |
Beta Was this translation helpful? Give feedback.
-
Interesting idea to reduce the clocks on the GPU to reduce consumption, makes sense since there's still ample headroom to meet the response times. A deeper dive into this area could be useful. |
Beta Was this translation helpful? Give feedback.
-
GPU is efficient but adds too much overhead. On small farms <300TiB problem. So far my GPU can keep up at lowest clock. No idea what will happen once I replot all, and farm size hits 1+ PiB So far |
Beta Was this translation helpful? Give feedback.
-
There is an issue with setting core, mem clock on consumer GPUs. I can set core clock, and force power limit only on |
Beta Was this translation helpful? Give feedback.
-
Hello BladeBit CUDA fans, we have produced a new performance report on the performance and power efficiency of BladeBit CUDA. The report is available at our tech blog: https://scienceofmining.com/20230828-BladeBit/BladeBit%20CUDA%20Performance%20Analysis-1.html
Feel free to check it out and let us know your feedback. We surveyed the Xeon W5-2455X, Core i9 12900k, Core i9 11900k, and Core i9 9900k for farming efficiency and also the RTX 3070 + RTX 3090.
Sept 2, 2023 update: We released a how-to for getting started with BladeBit CUDA benchmarking: https://scienceofmining.com/bb_guide/guide.html
-- The Science of Mining team
Beta Was this translation helpful? Give feedback.
All reactions