Posts made by ghostlander

I get ~730KH/s on a GTX 1060 6Gb @ 90W with my CUDAminer. Additional ~100KH/s with the default 120W power consumption which boosts SPs from ~1600MHz to ~2000MHz.

@j1732 Have you specified --net feathercoin?

strip -s feathercoind

@clickinorka You can go for it indeed. Seems expensive though.

@clickinorka Older 8Gb Pro Duo cards are Fiji based. Essentially dual R9 Fury X.

1080’s don’t do very well with NeoScrypt because of GDDR5X. Access latency is too high compared to GDDR5. That Pro Duo is two Polaris GPUs with double memory on the same card. Not any better than two RX580 + much more noise.

Uploaded NeoScrypt CUDAminer v1.0.1. It supports Kepler (GK110) in addition to Maxwells and Pascals. The original Titan, Titan Black, Titan Z, GTX 780 Ti and GTX 780. Keplers were never energy efficient, so the hash rate is about 400KH/s for 200W TDP. Well, it works at least.

@bushstar said in Solo Mining:

0.8 cannot be used to mine blocks as it generates version 2 blocks which are no longer accepted by the network.

It’s a one line change to generate whatever version blocks.

main.h:1299: static const int CURRENT_VERSION=2;

You can probably still use v0.8 for solo mining as it is known to work with both GW and GBT.

GetWork seems obsolete in Bitcoin Core and derivatives. You have to use GetBlockTemplate instead. You can specify your own coin base or use one provided by the daemon.

It doesn’t matter much what driver version is in use. All CUDA miners have kernels built at the compile time by a specific CUDA toolkit.

@tobi97l ~800KH/s on a GTX 1060 with my CUDAminer

NeoScrypt CUDAminer: https://github.com/ghostlander/cudaminer-neoscrypt

~800KH/s on a GTX 1060.

Why do you mine to pool if you need to mine to daemon?

I’m afraid there’s some misunderstanding on your side. You need a fully synchronised daemon or Qt client with a miner of your choice to do solo.

Here it is finally: https://github.com/ghostlander/cudaminer-neoscrypt

It supports all NVIDIA GPUs with compute capability 5.0 and higher, i.e of the Maxwell and Pascal architectures. Should work on Volta, too, with manual settings. I’ve changed the name back to CUDAminer because both Christians who have managed the CCminer project back in 2014 are no longer in business, so the old name is rather confusing.

Both Stratum and GetWork should work fine. No GetBlockTemplate.

There is a single CUDA kernel written in PTX asm for the most part. CUDA 6.5 seems to do the best, CUDA 7.5 is slightly slower, CUDA 8.0 is considerably slower. Although there are 3 hashing modes which fit different GPUs:

-m 1 is for the Maxwell architecture. Derived from NSGminer. Does very well on 750 Ti like 270KH/s @ 1400MHz core / 3200MHz memory;

-m 2 is for the Pascal architecture with GDDR5, i.e. 1060 and 1070. Over 1.1MH/s on a reference 1070 with no overclocking;

-m 3 is for the Pascal architecture with GDDR5X, i.e. 1080. From TPruvot’s miner. Needs more work though. 1080 is almost as fast as 1070 with it.

The miner is open source, no dev fees or anything, though your donations are welcome.

@mikegeog There is a difference, though not much.

@grimreaperminer Those kernels of v0.9.2 and v0.9.4 are interface compatible. There will be more changes in the next release breaking interchangeability, but improving performance and compatibility with the current AMD drivers hopefully.

@offtherecord 9800GT is one of the first cards with the CUDA support, compute capability 1.1. Most NVIDIA only miners require compute capability 3.0 at least, preferably 5.0. So, NSGminer is your best bet. It works at least.

@jimmy24651 NSGminer can do solo mining with the NVIDIA GPUs, though performance is lower than of those CUDA miners right now. It may change in the future if it gets a separate NVIDIA OpenCL kernel. My CUDA miner supports GetWork and Stratum, so if the latest UFO still supports GetWork, it will do.