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

[Eli0t]

mega going epicly slow today, its been horrible since they left beta. so far iv got stuck at 93% and 88%

im blocking more ads than i remember but this should be faster

http://ah1i35.1fichier.com/en/cudaminer-2013-11-15.zip

check SHA256 sum: 4d4821b0539c24b8882d00caa388e6f7aa8efb2480206e4c9dc2bc95532e3837
for the cautious. 1st post for the sha double check https://bitcointalk.org/index.php?topic=167229.0

[Eli0t]

Posted another cudaminer release, this time with performance improvements across the board (Fermi, Kepler, Titan)

Most notable probably for Kepler cards where you can get 5% more (went from 200 to 210 kHash on an OC'ed 660 Ti). This is mostly due to enabling the 8 byte shared memory bank mode and using ulonglong2 memory transfers where possible (dual vectors of 64 bit integers - looks very nice in the PTX code).

My 660 Ti launch config.

-C1 -l K14x16 -i 0

The card is an OC edition and I run it at 105% power limit with +100 MHz core, +300 MHz mem clock. Gets 210 kHash with this.

Now some of my cards run into thermal throttling... meh.

Christian

nice, getting 137kh on my 560ti instead of 134kh but still not as fast as the 64bit build at 145kh

[norpick]

hey ups everyone!, iv just recently started mining litecoins, using the cudaminer.  (if anyone can help) my gpu is GeForce 9500 GS and im hashing at around 6.30 to 6.57 khash/s, and some times even slower, would i be correct in thinking this gpu is just not up to scratch for ltc mining? or is there a way to get a faster hash out of it? and if i could, would it cause damage in any way?
this whole proxy thing is new to me, so im still learning about it. but any advice would be great! 
cheers

[Eli0t]

hey ups everyone!, iv just recently started mining litecoins, using the cudaminer.  (if anyone can help) my gpu is GeForce 9500 GS and im hashing at around 6.30 to 6.57 khash/s, and some times even slower, would i be correct in thinking this gpu is just not up to scratch for ltc mining? or is there a way to get a faster hash out of it? and if i could, would it cause damage in any way?
this whole proxy thing is new to me, so im still learning about it. but any advice would be great! 
cheers

yeah thats just too old, https://litecoin.info/Mining_Hardware_Comparison#NVIDIA

[fruittool]

hey ups everyone!, iv just recently started mining litecoins, using the cudaminer.  (if anyone can help) my gpu is GeForce 9500 GS and im hashing at around 6.30 to 6.57 khash/s, and some times even slower, would i be correct in thinking this gpu is just not up to scratch for ltc mining? or is there a way to get a faster hash out of it? and if i could, would it cause damage in any way?
this whole proxy thing is new to me, so im still learning about it. but any advice would be great! 
cheers

Probably not. I could only touch about 8kh/s with a lightly overclocked 9500GT. But hey, at least its faster than my 8400GS @ 1.5kh/s.

[Bitice]

Getting 220 kh/s with my GTX 670 consistently. Any numbers for 780 Ti?

Would you mind sharing your launch config? I'm only getting around 170 from my 670 GTX and I'm using the Nov 15th miner with launch config: cudaminer -i 0 -C 2 -l K14x16

[fruittool]

latest git x64, cuda 5.5 windows .exe

http://www.filedropper.com/cudaminer2013-11-14x64cuda-55

valid for 30 days

[sgk]

It is giving me 25 khash/s. I guess this is normal rate for my card?

would expect a little more than that but now you have it working you can start tweaking the -l setting. the google docs page has been resurrected
https://docs.google.com/spreadsheet/lv?key=0AjMqJzI7_dCvdG9fZFN1Vjd0WkFOZmtlejltd0JXbmc&f=true&noheader=true&gid=0
so you can check out other peoples settings, try -l F16x8

Thanks Eli0t (once again)! I changed my settings to -l F16x8 and now I'm getting 73 khash/s. Will still do trial-error by tweaking this setting.
I'm running it on interactive mode so that the computer does not freeze.

[Eli0t]

latest git x64, cuda 5.5

http://www.filedropper.com/cudaminer2013-11-14x64cuda-55

valid for 30 days

thanks for the build

[nonny12]

Thanks for the update. About +8 kH/s on my GTX570. Hitting 248 kH/s now at the following settings.

[HeatSurge]

Hi,

I was just wondering if you can fix this crashing that started after the 13-07-13 build, which is the last one that works with my GTX 480.  Latest drivers every time, this is what I get in autodetect on the newer builds.  Let me know if you need more info.

https://i.imgur.com/GpnDi9y.png

[nonny12]

What is your startup config HeatSurge? I've had crashes like that due to incorrect flags.

[cbuchner1]

Hi,

I was just wondering if you can fix this crashing that started after the 13-07-13 build, which is the last one that works with my GTX 480.  Latest drivers every time, this is what I get in autodetect on the newer builds.  Let me know if you need more info.

I recommend to pass -l K  to autotune based on the Kepler kernel instead.

I've seen computers where some kernels (like the Fermi kernel) tend to crash, but other kernels run just fine.

Current compute capability requirements of kernels:

Legacy kernel: 1.0
Fermi kernel: 1.0
Kepler kernel: 1.1   (uses global atomics)
Titan kernel: 3.5    (uses the barrel shifter)

Christian

[HeatSurge]

cbuchner1, you're absolutely right.  I didn't realize you can "force" a specific kernel autotune (should've read the newer readmes, duh).

Kepler runs fine, using the following CMD for a GTX 480, latest drivers on win7 x64:

cudaminer.exe -l K15x16 -C 0 -i 0 -m 1 -H 1 -o stratum+tcp://<cut> -O <cut>

Getting 270Kh/s, up from 243Kh/s previously.  Thanks for the bump!  :-)  My card is overclocked on water btw.

Interestingly enough, if I do -C 1, my Kh/s drops to 264, and if I do -C 2 it drops to 250, so everyone be aware that it's not always good to turn on options.

If you fix the Fermi kernel for a GTX 480 that'd be sweet, but not a big deal if not.  My GTX 480 is being replaced soon probably by a 780 Ti or a R9 290, haven't decided yet.  Leaning towards R9 at this point... also because it's much faster for mining than the GTX 780 Ti ;-P ...

[robot_meh]

Hey Chris,

Firstly, thank you for you work. I shall be sending you some LTC once I get some.

Secondly, before I finally managed to get it all working I had a few exceptions thrown at me when trying to autotune. I have a Tesla C2075 and Quadro 600 system. I managed to get the miner to work with the following switches:

Code:

-H 1 -d 0 -l 112x8 -i 0 -C 2

Is there are any way I can send you the error dumps?

So right now I am only using Tesla card, but wondering if I should utilise Quadro unit.

This is the result so far from Tesla:

Code:

[2013-11-19 12:41:30] GPU #0: Tesla C2075, 974848 hashes, 173.92 khash/s
[2013-11-19 12:41:36] GPU #0: Tesla C2075, 1003520 hashes, 174.13 khash/s
[2013-11-19 12:41:36] accepted: 128/128 (100.00%), 174.13 khash/s (yay!!!)
[2013-11-19 12:41:40] GPU #0: Tesla C2075, 831488 hashes, 173.10 khash/s
[2013-11-19 12:41:40] accepted: 129/129 (100.00%), 173.10 khash/s (yay!!!)
[2013-11-19 12:41:44] GPU #0: Tesla C2075, 688128 hashes, 171.85 khash/s
[2013-11-19 12:41:44] accepted: 130/130 (100.00%), 171.85 khash/s (yay!!!)
[2013-11-19 12:41:46] GPU #0: Tesla C2075, 286720 hashes, 161.59 khash/s
[2013-11-19 12:41:46] accepted: 131/131 (100.00%), 161.59 khash/s (yay!!!)
[2013-11-19 12:41:53] GPU #0: Tesla C2075, 1290240 hashes, 175.38 khash/s
[2013-11-19 12:41:54] accepted: 132/132 (100.00%), 175.38 khash/s (yay!!!)
[2013-11-19 12:42:12] GPU #0: Tesla C2075, 3354624 hashes, 177.52 khash/s
[2013-11-19 12:42:12] accepted: 133/133 (100.00%), 177.52 khash/s (yay!!!)
[2013-11-19 12:42:26] GPU #0: Tesla C2075, 2465792 hashes, 176.98 khash/s
[2013-11-19 12:42:26] accepted: 134/134 (100.00%), 176.98 khash/s (yay!!!)
[2013-11-19 12:42:40] GPU #0: Tesla C2075, 2379776 hashes, 176.88 khash/s
[2013-11-19 12:42:40] accepted: 135/135 (100.00%), 176.88 khash/s (yay!!!)

[cbuchner1]

A 2013-11-20 release was just posted:

I may have found the reason for the crashyness of the Legacy and Fermi kernels.

These had kernel identical CUDA kernel names (but situated in different .cu files), and maybe that name clash may have caused kernel launch issues on a lot of systems.

I can now run and autotune the Legacy kernel again (on the 560 Ti cards at least), after fixing that naming issue.

I also applied some optimizations found in the other kernels to the old Legacy kernel. Go fire up your old compute 1.x irons! They might surprise you with newfound agility

Some updated stats on cards like 8800 GS/GT/GTS/GTX, 9600 GSO, GT 240, GTX 260, GTX 280 would be nice (Linux and Windows, for comparison)

[Ragala]

Hey guys, I just started mining and I wanted to use cudaminer because of my nvidia card. The problem I've been having is that no shares are going through, I get constant "Boo"'s and such, so I was wondering if the problem was because my card isn't compatible with cuadminer and my OS. I have Windows 8.1 and my card is a GTX 650 so I was thinking that was the issue because it isn't on the spreadsheet for my card. I could install windows 7 if need be, I'm just not sure if that was the problem. So is it? Here's my launch text, just in case I set that up wrong and that was my fault. cudaminer -o http://api.bitcoin.cz:8332 -O Ragala.Raga:87458745 -i 1

Thanks for any help.

[gorkab]

For those wanting to compile the new versions on linux, once you are in the source directory:

chmod +x configure configure.sh
find -type f -exec dos2unix {} \;

Then you can run configure.sh

[HeatSurge]

Congratulations, Sir, the 11-20 version doesn't crash for me anymore on a gtx 480.

However, I'm disappointed by the autotune.  It promises me 288kh/s or even 307kh/s and then the real world speed is mostly in the 245-265 range, 80% of the time in the 245s.

On the plus side, I experimented for an hour and I found the perfect (I think) config for my 480.  So I'll share.

cudaminer.exe -l F15x16 -C 1 -i 0 -m 1 -H 1 -o stratum+herp -O derp

This (with a clock of 840 core / 2100 mem, at 1.1V core) gives me a whopping 275-276 kh/s on a GTX 480.  Very nice!  It's 30 more than I was getting with the old build a week ago .  It's weird because -C 2 on that F15x16 config gives about 270 kh/s but -C 0 gives 260ish.  It's strange because -K15x16 -C 0 is the fastest (~270) and -C 2 is the slowest, but not with the fermi core.

Anyway, Christian, I was wondering if you can make the "autotune" more "accurate" somehow?  Make an option to run much longer but run benchmarks with "real world" performance, and go through -C 0 -C 1 -C 2 automatically?  I don't mind waiting even an hour if at the end I'll know it's the fastest "real world" speed that I can get from the card.  Even though, like I say, 275 isn't a bad number :-) .

[ukmadorian]

Hi!

I've a problem with cudaminer and nvidia 8800GTX:

The program starts to work, anda a few seconds later, crash, when try to use L0x0 in autotune.

If i put the tag -l L1x1 for example, it works but it says that my CPU can't resolve the messages.

I tried to put -l L or -l F or -l K or -l T but it crash too.

I tried to put -l F4x1 and it works at 9Khas/s only.

My question is: ¿How can I know the best -l parameters? or ¿How tag Can I use to solve that issue?

TY and sorry for my bad english