[ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX]

[Greg121986]

Hi, I've a problem with Cuda miner.
When i want to start it, i can't because i've this message : json_rpc_call failed.
Please help me.

When I encountered this error, it is because I was running the cudaminer.exe file and not my .BAT file like I was supposed to.

[cbuchner1]

I had two two good reasons to implement an option for low CPU use

-My (t)rusty old AMD Phenom II X6 1055T processor has a TDP of 125 Watts. It was getting loaded 75-80% when my various GPUs were mining at 900 kHash in total. I rather save 100 Watts from my power bill, thanks

-my new mining rig that I am currently building will mine at nearly 1.5MHash/s and it only has a weak dual core i3 processor. So I had to make sure the CPU load is low.

Those who don't like the -H 2 in the 2013-12-10 build can still specify -H 0 or -H 1  

[Force_field]

Hi @ll
My first post here... will try to help as much as I can (and ask as little as I must)
First of all, thanks Buchner for making this possible!
I have two "average" working rigs mining at the moment, so I'm going to share my xp hoping to help all users and developer.

Gainward GTX 570 1280MB phantom @ stock (running on intel E8400, 4GB/1066, Asus formula rampage, all @ stock)
Gigabyte GTX 760 2GB OC @ stock (irrelevant in this case, but i7, 16GB, Z87... too good to make any "noise" anyway)

Anyway... after 4-5 days in testing, I have got the best results with Cudaminer 2013-12-01, 64bit in both cases (2013-12-10 not tested on GTX 570).
Both rigs running on Win7 Pro X64 SP1 with various (different for sure) official updates...

GTX 570 config (max 255KH/s):
cudaminer.exe -i 0 -H 1 -l F15x16 -C 1 -o stratum+tcp://...
All other auto or manual configs work with lower rates, but never below 200KH/s (I'm talking about "correct settings" for this card, not random Fxxxx)

GTX 760 config (max 177KH/s):
cudaminer.exe -i 0 -H 1 -l K96x2 -C 1 -o stratum+tcp://...
this is MANUAL config, because auto NEVER gives me this setup! All other configs are lower, but not lower than 150KH/s.

My next post will be about some odd things on both rigs, which hopefully can be eliminated or "refined" in further development...

Anyone with better configs/setups are welcome... we all love to try something better!

P.S. All of this is obviously very different on every system, so EVERY SINGLE component in our rigs can make a serious change in hashrates.

[scarface]

been using this to mine with my 550 gtx ti, getting around 72khs

is this good or bad? should i change any settings

[qwerty77]

Will it be possible to configure a backup pool for cudaminer like you can do it in cgminer? This planned as a feature?
Lately a lot of pools are having problems, so I think this would be a nice option for pool mining.

[Force_field]

...And now something about weird things on my GTX 570 and GTX 760 rigs:

GTX 570
- Most people claim that system RAM has little or no difference in hashrates, but I have suffered serious problems when i left just 2GB on mobo. These problems started immediately after cudaminer start, because the good-old config (posted in previous post) was NOT WORKING at all! F15x16 could be executed just to start the mining, but with "wrong detection" of over 600KH/s and errors on every single share ("not validated on CPU"). Auto config was giving different setups every time, rates dropped to ~200KH/s, and what is most interesting, auto config was reporting MAX WARPS: 171 (instead of 264 reported with 4GB of RAM)?! My rig was going too WILD and unpredictable for the stuff that is "not so much RAM dependent"!
- My max of ~255 KH/s can be achieved ONLY if browser (chrome in my case) is active BEFORE starting cudaminer.exe, and some page is opened (usually my pool page). After starting cudaminer, I can close everything and hashrate stays the same. Otherwise, same settings without browser give me only ~235 KH/s?!
- Some people were using Kepler kernels on Fermi devices (in previous versions of cudaminer), but at my place it was making a serious drop in KH/s
- Just to note... interactive mode (-i 1) is taking away 7-8KH/s on this rig

GTX 760
- As I mentioned before, auto never reaches the best manual config for me (and there is a better config than my current one for sure), and the auto option is changing config very often (I had like 3 different configs on 5 cudaminer restarts)?!
- This GPU (or whatever) is varying MUCH more in hashrates (160-177KH/s up-down most of the time), than GTX 570 which has pretty much "fixed" rates 252-255KH/s!

BOTH
- Is it possible that I NEVER get better hashrates on any config with 32bit vs 64bit? So many people (developer too) usually has better results with 32bit version, but I never even saw that for a second?! (not complaining, just asking...)

[kb24mvp]

Thanks for the concern guys, my laptop can take it though... def built for it and I have a turbo fan to cool

temps are pretty stable at around 58 degrees on load mining at 100kh/s

You guys have any idea on the settings I could use?   (770M)

My 660 Ti has 7 SMX and Compute 3.0. I run it with

-H 2 -l K28x8 -C 1 -i 0

Yours should have 5 SMX and Compute 3.0. Try -H 2 -l K20x8 -C 1 -i 0

well that got me up to 105kh/s! Will take it... and prob need to read the readme indepth to see what else I can tweak

Thanks

[skyvahaerie]

Double check your overclock. MSI Afterburner (and others) will not report the true clock of the GPU Core. Mine is showing 1359 on MSI Afterburner, but GPUZ shows a boost clock at 1290Mhz. The GPUZ reading is the true speed of the core. I can push the GPUZ reading to 1302Mhz but it does not last more than an hour or so before everything fails and Direct3D crashes. The differences in clock will result in the 5-7hash/s discrepency we are seeing here. I bounce between 199-203Khash/s but as I said before I sometimes dip to 150 for unknown reasons.

I think GPUZ is associated with TechPowerUp, so you should be able to find it via search.

What i posted earlier were the GPUZ readings. Anyway i ran some benchmarks for several hours and it seemed stable @1293.4 core and 1701 mem, but after about 7 hours of mining this night my rig crashed. Unfortunately i didnt run any diagnosis sofware except for GPUZ to check the temps, as i thought its stable after the benchmark results :/
I went back to 1280 core and 1700 mem for now, as i dont have time to tinker at the moment. With -l K95x2 -i 0 -H 2 it nets me ~188-190 kHash/s, still better than 160 kHash/s...

[Valnurat]

How do I use cudaminer through a proxy?

Well I think my problem is the proxy. At home I don't have any problem to run cudaminer, but at office I do and I know we use a proxy.

Can anyone help me?

Anyone?

[paulus51]

guys help me out here

what is the best setting with my GT610 and put in some CPU i have a quad core 2 Q6600
please help me out for optimal settings cudaminer

[vdell]

guys help me out here

what is the best setting with my GT610 and put in some CPU i have a quad core 2 Q6600
please help me out for optimal settings cudaminer

You just really have to go with trial and error. Test different flags and see what autotune suggests. Be aware, though, that you're card will most likely have very limited performance. You could try running cudaminer + cpuminer together.

[cbuchner1]

How do I use cudaminer through a proxy?

Well I think my problem is the proxy. At home I don't have any problem to run cudaminer, but at office I do and I know we use a proxy.

Can anyone help me?

Anyone?

google cpuminer proxy. or just follow that link.

https://bitcointalk.org/index.php?topic=63689.0

this requires that you use the getwork protocol, not stratum.  So your pool URL must start with a http:// , not stratum+tcp://

[cbuchner1]

...And now something about weird things on my GTX 570 and GTX 760 rigs:

GTX 570
- Most people claim that system RAM has little or no difference in hashrates, but I have suffered serious problems when i left just 2GB on mobo. auto config was reporting MAX WARPS: 171 (instead of 264 reported with 4GB of RAM)?!

single memory allocation mode (required for -C 1 and -C 2) has to do a single cudaMalloc() call to get one big chunk of memory.

The WDDM driver model used by Vista, Win 7/8 and later imposes the following restrictions for CUDA.

The maximum size of a single memory allocation created by cudaMalloc() or cuMemAlloc() on WDDM devices is limited to MIN( (System Memory Size in MB - 512 MB) / 2, PAGING_BUFFER_SEGMENT_SIZE ). For Vista, PAGING_BUFFER_SEGMENT_SIZE is approximately 2 GB.

And you've just reduced your "System Memory Size in MB" to 2GB.

To solve it, try either
1) switch to Linux or Windows XP
2) don't use -C 0 or -C 1 so that cudaminer doesn't have to use single memory allocation mode
3) use some card BIOS/hardware hacks to use the nVidia TCC driver (that is made for Tesla and doesn't have these restriction).

I'd vote for trying 2), followed by 1). Don't brick your card with 3)

Christian

[Valnurat]

How do I use cudaminer through a proxy?

Well I think my problem is the proxy. At home I don't have any problem to run cudaminer, but at office I do and I know we use a proxy.

Can anyone help me?

Anyone?

google cpuminer proxy. or just follow that link.

https://bitcointalk.org/index.php?topic=63689.0

this requires that you use the getwork protocol, not stratum.  So your pool URL must start with a http:// , not stratum+tcp://

The link is cpumining. I use cudaminer. There is a parameter:
-x

but I do not know how syntax for this is.

[Force_field]

@cbuchner1

2GB config made for testing purposes only, so no harm done, 4GB were mining again 15 mins later
Not going to change anything - both rigs are mostly used for work/gaming... so mining is done on "idle hours only"
Tnx anyway
Any idea about "browser + cudaminer = high KH/s" situation from my previous post?! Sounds stupid but happens all the time on the first rig only!

[cbuchner1]

The link is cpumining. I use cudaminer. There is a parameter:
-x

but I do not know how syntax for this is.

guess what cudaminer was derived from. That HTTP_PROXY environment variable should still work

Code:

	case 'x':			/* --proxy */
		if (!strncasecmp(arg, "socks4://", 9))
			opt_proxy_type = CURLPROXY_SOCKS4;
		else if (!strncasecmp(arg, "socks5://", 9))
			opt_proxy_type = CURLPROXY_SOCKS5;
#if LIBCURL_VERSION_NUM >= 0x071200
		else if (!strncasecmp(arg, "socks4a://", 10))
			opt_proxy_type = CURLPROXY_SOCKS4A;
		else if (!strncasecmp(arg, "socks5h://", 10))
			opt_proxy_type = CURLPROXY_SOCKS5_HOSTNAME;
#endif
		else
			opt_proxy_type = CURLPROXY_HTTP;
		free(opt_proxy);
		opt_proxy = strdup(arg);
		break;

now looking at that -x or --proxy switch. It appears the code accepts various socks URLs, as well as plain http:// proxy URLs. I did not write that proxy code though and never used it myself, it is taken straight from cpuminer.  Its typical use might look like this, in addition to all the  other command line options for specifying the (http!) pool URL for the getwork protocol.

Code:

-x http://proxy.mycompany.com:8080

I do not know enough about the socks protocol to confirm whether or not it would support a stratum connection. You can try, but you also need to have access to a socks proxy that would permit connections on the stratum port.

[cbuchner1]

Woooha!

Seems like David Andersen took a cpuminer (most likely), modified it for CUDA and boosted the hash rate of an amazon EC2 instance from 150 (cudaminer) to 220 kHash/s (his miner)

http://da-data.blogspot.de/2013/12/briefly-profitable-alt-coin-mining-on.html

and in the next blog posting he will share his optimization techniques with us, I hope.

Christian

[vosovich]

@cbuchner1

2GB config made for testing purposes only, so no harm done, 4GB were mining again 15 mins later
Not going to change anything - both rigs are mostly used for work/gaming... so mining is done on "idle hours only"
Tnx anyway
Any idea about "browser + cudaminer = high KH/s" situation from my previous post?! Sounds stupid but happens all the time on the first rig only!

The first thing I would do check is the power savings settings. You're most likely using windows (the gaming gave it away :p) and I wouldn't be completely surprised if windows doesn't recognize an app running in the console as a priority, given a conservative power savings profile.

[vdell]

@cbuchner1

2GB config made for testing purposes only, so no harm done, 4GB were mining again 15 mins later
Not going to change anything - both rigs are mostly used for work/gaming... so mining is done on "idle hours only"
Tnx anyway
Any idea about "browser + cudaminer = high KH/s" situation from my previous post?! Sounds stupid but happens all the time on the first rig only!

The first thing I would do check is the power savings settings. You're most likely using windows (the gaming gave it away :p) and I wouldn't be completely surprised if windows doesn't recognize an app running in the console as a priority, given a conservative power savings profile.

I noticed the same thing just now. Rebooted machine (Win7 x64 SP1), started cudaMiner, and the speed was much worse than what it was supposed to be. Also, GPU-Z said that GPU load was only at 78%. Immediately after I started Chrome the GPU load rose to 99% and cudaMiners' speed returned to the expected level. Would be interesting to know what's causing this...

[blackraven1425]

I noticed the same thing just now. Rebooted machine (Win7 x64 SP1), started cudaMiner, and the speed was much worse than what it was supposed to be. Also, GPU-Z said that GPU load was only at 78%. Immediately after I started Chrome the GPU load rose to 99% and cudaMiners' speed returned to the expected level. Would be interesting to know what's causing this...

I'd double down on power saving settings, since IIRC, Chrome is hardware accelerated by default on most machines.