Sign In Register Categories Recent Discussions Activity Unanswered Best Of... Categories All Categories Mining Pool Discussion General Project Discussion (non-technical) Education Protocol and Client discussion web3-js Whisper Swarm 2 RLP IoT & Hardware Smart Contracts and Dapps Serpent Solidity Projects Reference clients code and builds Eth & AlethZero- Cpp Implementation Geth - Go Implementation Mist Node.js Implementation Python Implementation Mix Other Implementations Meetups Other Events Jobs & Skills Press and Articles Audio/Video Ether Sale Other Languages Chinese German Italian French 2 Hebrew 6 Japanese Portugese Romanian Russian Spanish Turkish Watercooler GPU Mining is out, come and let us know of your bench scores!StephanTual 1,282 May 2015 in Mining **This thread is for benchmark results only - post questions in other threads**Geth GPU support is coming soon, in the meantime, the C++ miner is out.The purpose of this thread is to post GPU results, not CPU.

/ethereum/cpp-ethereum/wiki/Installing-clientsSecond, run eth -M -GOf course, you'll need a GPU for this to work.Use --opencl-device 0 if Eth is not identifying your OCL device properly Third, post your results!The format is:OS / OS version / GPU / Eth Versionmin / mean / max: xx / xx / xxInner mean: xx.xxxx MHPlease post everything in MH.Please do not post questions etc as they will be removed.Update:14/05/15 - corrected typo on opencl, clarified thread rules May 20156 123456743»Comments StephanTual 1,282 May 2015 23.2143 MH 4 betax 1 May 2015 4 Wil611 77 May 2015 Ubuntu / 14.04 /Radeon R9 280x 3GB / 0.9.20min / mean / max: 22.3 / 22.4 / 22.422.4570 MH 0 don 7 May 2015 0 LarsPensjo 35 May 2015 1 dagide 15 May 2015 0 n12o 6 May 2015 0 literarydevice 13 May 2015 1 0xff 22 May 2015 0 paul_bxd 1 May 2015 2 subtly 7 May 2015 1 Maren85 41 May 2015 **This thread is for benchmark results only - post questions in other threads** 1 luc4 7 May 2015 0 PascalVerst 6 May 2015 1 PascalVerst 6 May 2015 PascalVerst said:Windows 7, NVIDIA GeForce GTX 97025000 H/s Another PC:Windows 7, NVIDIA GeForce GTX 550 Ti14000H/s 1 urubu 19 May 2015 1 mcmike313 27 May 2015 1 dont12 60 May 2015 0 Michael_A 61 May 2015 Mac OS GeForce GTX 775M (OpenCL 1.2 )min/mean/max:2443419/2513231/2531527 H/sinner mean: 2530684 H/s2.5MH/s this GPU is not performing at all 0 Brips 52 May 2015 0 MatekDK 3 May 2015 0 etherninja 24 May 2015 1 ConradJohnson 130 May 2015 1 ijohanne 3 May 2015 0 Dude_x 40 May 2015 1 acros 63 May 2015 1 acros 63 May 2015 1 jzen 49 May 2015 @acros I observed similar odd behavior using recent versions of ethminer, where min/mean/max report rates matching those seen while actually mining, but the inner mean shows either zero or some other undervalued rate.create litecoin farm

However, the version of ethminer referenced by Stephen in his Mining FAQ appears to accurately report the inner mean.I've been meaning to submit a github ticket about this issue but I've been super busy with other matters.1 acros 63 May 2015 Thanks @jzen , at least the min/mean/max is a fair approximation!More benchmarks:Ubuntu / 15.04 / Tonga PRO Radeon R9 285 2GB / v0.9.25-70867904min/mean/max: 14767445/14959684/15204352 H/sinner mean: 10019726 H/s(actual rate closer to 15.5MH/s)Ubuntu / 15.04 / Tahiti PRO Radeon HD 7950/8950 OEM / R9 280 3GB / v0.9.25-70867904min/mean/max: 17126741/17126741/17126741 H/sinner mean: 0 H/s(actual rate closer to 15.5MH/s as well) 1 Brillopad12 17 May 2015 2 123456743» Sign In or Register to comment.ethereum chain downloadThe point of this benchmark is to give a controlled time trial for processing blocks including all of the guff that goes with block processing including PoW verification, transaction signature checking, EVM code execution, receipt verification, uncle validation and database population.fig bitcoin

In order to mitigate biases one way or another for each of those aspects in the benchmark, we use the first 1,000,000 blocks of the Frontier mainnet chain.There are probably several ways of effecting this benchmark for each client (e.g.cpp-ethereum has the --import feature), but for now we're settling on a simple "sync from localhost" method.ethereum cad price chartThis involves setting up a client synced up to block 1,000,000 of the Frontier network (we used eth, specifically, but in principle it shouldn't matter which client is used).ethereum nxtFollowing this, the client to be benchmarked is executed on the same machine with a clean database in the full block processing mode.ethereum cad price chartIt is initialised with the former client's address and allowed to synchronise.

The process is timed until it eventually imports block 1,000,000.We did all tests on a standard Digital Ocean 4GB droplet running Ubuntu 14.04.3 x64.To automate the method described above, the following library has been created: To benchmark a client, first, a "master" node (e.g.geth) has to be installed and synced up from the network up to a certain point (e.g.Then, the master client should be restarted in "no-discovery" mode, e.g.: Next, the client to be benchmarked has to be installed and configured, if necessary.The benchmarking script has to be passed the "master" node's enode address along with other optional arguments (run bin/run-bench -h for more info).Note: since the main purpose of this utility is to benchmark full blockchain processing times, certain optimisation options like geth --fast have been intentionally turned off.The point of this is to give a controlled test between clients that reflects typical on-chain situations.We do this by defining a common dataset of key/value pairs for insertion into the trie and root-calculation.

There are two modes to reflect the two use cases of the trie in Ethereum: For this benchmark, the dataset is inserted into the trie with root hashes being computed at specific intervals ("era_size").This number represents the number of update operations per root calculation.For this benchmark, the important thing is to take all data and determine the root; no roots need be computed along the way and there is no assertion that any state used when determining the root be required.In this sense era_size is equivalent to the dataset size.Sensible implementations may precompute the dataset to avoid the additional burden of SHA3 computation at benchmark time.The final trie root after 1000 rounds should be 36f6...93a3 for SYMMETRIC = True and da8a...0ca4 for False.Test ID is given as pair_count-era_size-key_size-value_type, where valid value_types are ran (SYMMETRIC = False) and mir (SYMMETRIC = True).The standard secure-trie test is 1k-9-32-ran.ran is used as value_type as it better reflects the kinds of values that are typically written in the blockchain.