Greetings, We are pleased to announce that we have released builds of ccminer (for NVIDIA) and cgminer (for AMD) which may be found on GitHub at the following location: https://github.com/decred/decred-release/releases/tag/v0.1.0_miners Part of the improvements contained in these builds were funded as part of Rfp-5: Mining Protocol Development And Pool Integration and include work by @tpruvot and @Wolf, many thanks to them. ccminer This is the first official build of ccminer and there are no known issues at this time. We are currently only providing a CUDA 7.5 build due to lack of demand for CUDA 6.5. cgminer Intensity has been changed to be more inline with ccminer/sgminer. Previously the maximum intensity was 14 and now the maximum intensity is 31. Please adjust your -I/--intensity flags accordingly. Hash rate has been significantly improved. Several issues remain with cgminer (including high CPU usage) (GitHub issue tracker). General Notes We are currently only providing 64-bit binaries for Linux and Windows due to lack of demand for other platforms. These builds are now generated by scripts which are contained in the repository (except ccminer for WIndows which is built in Visual Studio 13). This removes a lot of room for errors or omissions when building by hand which occurred with some of the previous builds. We hope these builds are relatively error free but without wide testing, it's hard to be certain. Please report your findings here on the forum. It would also be beneficial to open issues and feature requests on GitHub (ccminer | cgminer) so they can more easily be tracked. We view automated builds as an important first step to improving the mining software stack and strive to continuously improve it as time goes on now that releases can more easily be made. Thank you! Sincerely, Decred Project
Comparison on a rig with 4 cards. cgminer 3.7.2 (decred cgminer v0.0.4): 2.15GH/s per card, 8.60GH/s total, WU 950/min total with I=14 cgminer 0.1.0-decred: 1.75GH/s per card, 7.00GH/s total, WU 650/min total with I=31 Except Intensity everything else identical in the batch file. Edit: v0.0.4 running here since day one with the above mentioned performance.
The one you mentioned back in December with 16.3.1. I would have assumed the new miner to work perfectly with that one since it seems to be your favorite card. v0.0.4 did.
Hmm, I'm assuming you're referring to the R9 Nano since that's the only AMD card I have. You should be able to hit closer to ~2.3GH/s with the new code and powertune set to 20%. Both myself on Arch Linux w/Catalyast 15.12 and a Windows tester have clocked around the 2.3GH/s mark. We have single card setups though so maybe that's what is causing the discrepancy.
Thanks, but that doesn't explain the above mentioned bad performance compared to v0.0.4. There is no tuning whatsoever on my cards, just stock out of the box and connected.
Solo PoW mining. About one hour testing. With I 30 performs about 200-250Mh/card lower hash rate than v0.0.4. Total combined hash rate is about 2.900Gh, with old CGminer 3.550Gh. After few minutes begins to throttle up fans to maintain GPU temp. arround 65 °C, old one maintains temp. 72-75°C. For now I will stay with ancient v0.0.4. 2x R9 Nano Sapphire, Catalyst 16.150, Windows 7 Ultimate x64, Intel Pentium G3220, 4Gb RAM
Thanks, that confirms my results. 0.1.0-decred not working well on this card, even at full Intensity.
Thanks for giving it a try and reporting results. We're going to procure/setup some multi-GPU machines so we can diagnose.
Thanks for your great work and effort ! Just tested ccminer-decred-1.0 versus ccminer from github with compute52. After 2 minutes running -a decred --benchmark I got: [2016-04-28 17:12:49] GPU #0: GeForce GTX 970, 1445.04 MH/s [2016-04-28 17:12:51] GPU #1: GeForce GTX 960, 953.90 MH/s [2016-04-28 17:12:51] Total: 2402.60 MH/s versus [2016-04-28 17:16:59] GPU #0: Gigabyte GTX 970, 1452.42 MH/s [2016-04-28 17:16:59] GPU #1: Gigabyte GTX 960, 953.25 MH/s [2016-04-28 17:16:59] Total: 2403.12 MH/s
I also have a drop in performance with this cgminer release on AMD fury X cards (based on the Fiji GPU like the Nano)
Running the new ccminer binary, was 505 MH/s now is ~518 MH/s, not too shabby EVGA NVidia 750 Ti 2Gb with Windows 10. http://yiimp.ccminer.org/?address=DsWg6rBSDoZ4gBCWpDe1LHhJK8cpsv1WNSa (yiimp reports totally different values, guess network latency is at play or something funkier).
@jolan Well not only multi-GPU. Latest cgminer v0.1.0 (cgminer-decred-win64-v0.1.0-20160427.zip) seems to be a disappointment for Windows miners. With an AMD R9 380 the hashare I get with cgminer v0.1.0 is about 1.05Gh/s with intensity at 31 while with the older v0.0.4 I get about 1.20Gh/s with intensity at 14. GPU usage is almost constant on 100% with both cases. sgminer also gives me lower hashrate than cgminer v0.0.4, so there is no benefit of using stratum on pools. I hope we will see some improvements with miners for AMD VGAs in the future. A new version of cgminer with at least the same hashrate as v0.0.4 but lower CPU usage will be great and since there is demand for stratum a better optimized sgminer for AMD VGAs will also be good since from what I read ccminer works fine for NVIDIA VGAs. Looking forward too for the gominer with which I hope we'll all get better results and stability.
I tried it again, with only one card and the same sad result. Won't hash higher than 1.75GH/s at I=31. v0.0.4 is doing 2.15GH/s at I=14, as already mentioned. That's nearly a 20% loss in hashing power. Interesetingly the cards run at the same temperature and fan speed despite the much lower hashrate.
I did not see an improvement in hashrate using the new cgminer on several different multi-gpu AMD rigs. In fact I saw lower performance across the board - most notably on a rig with a single R9 280x. This card does about 700 Mh/s in the 0.0.4 Decred release of cgminer 3.7.2, but with the new v0.1.0-decred version, this card will only do up to 76 Mh/s. On another rig with 2 AMD HD6950s, these cards will do about 650 Mh/s each with v3.7.2, but in v0.1.0 they do around 170 Mh/s each. On another rig with several GTX 980 Ti cards, I did not see an improvement in using the new ccminer vs ccminer v1.7.4. Hashrate was the same or less with the new version. All the AMD machines are running Windows 8.1 x64 and the Nvidia machine is running Windows 7 x64. I'm happy to test out new releases and report my findings for the developers of the project and the community, but unfortunately this release appears to be a bust, for me at least. I'll be sticking to the older cgminer and ccminer versions for now.
@David: Did you use the same intensity on both? Remember that e.g. I=14 on v0.0.4 isn't I=14 on 0.1.0-decred. With I=14 on 0.1.0-decred I too got only a tiny fraction of my previous hashrate. I had to use I=31 to get at least 80% of the previous one.
Okay, so this means that those using the latest AMD cards have no choice but to use the official decred cgminer v0.0.4. The new official cgminer is missing 20% hashrate, ccminer is only for nvidia and sgminer is reportedly also missing appr. 20% hashrate.
Thanks for the reminder! I set intensity to 31 and got about the same level of performance on the rig with two 6950s (650-660 Mh/s each). On the rig with the single 280x, I did see a slight improvement of about 10%