liomojo1
|
|
February 18, 2014, 10:34:59 PM |
|
New nvidia driver out 334.89
|
|
|
|
cbuchner1 (OP)
|
|
February 18, 2014, 10:36:17 PM Last edit: February 18, 2014, 10:50:50 PM by cbuchner1 |
|
Hi all. I posted a 2014-02-18 release that will not barf when it finds a Maxwell. This doesn't mean that it was optimized for Maxwell (it wasn't!), but it runs. Previously you had to force it to use either the T kernel (scrypt, scrypt-jane) or the K kernel (best for keccak) in order to run. Now it's selected automatically instead of just crashing. Also CTRL-C terminates the program much faster when mining on a stratum server. Same for the --time-limit option. Oh, and Autotune does something on Keccak. I just don't think it's of much use You still need to pass a large -L parameter because otherwise the (unnecessarily allocated) scrypt scratchpad limits the maximum number of warps we can run this with. Let me know if my changes broke anything else for you... Christian
|
|
|
|
ivanlabrie
|
|
February 18, 2014, 11:10:19 PM |
|
Thanks! I'll give it a shot asap...
|
|
|
|
chapmanjw
Newbie
Offline
Activity: 53
Merit: 0
|
|
February 18, 2014, 11:10:43 PM |
|
Hi all. I posted a 2014-02-18 release that will not barf when it finds a Maxwell. This doesn't mean that it was optimized for Maxwell (it wasn't!), but it runs. Previously you had to force it to use either the T kernel (scrypt, scrypt-jane) or the K kernel (best for keccak) in order to run. Now it's selected automatically instead of just crashing. Also CTRL-C terminates the program much faster when mining on a stratum server. Same for the --time-limit option. Oh, and Autotune does something on Keccak. I just don't think it's of much use You still need to pass a large -L parameter because otherwise the (unnecessarily allocated) scrypt scratchpad limits the maximum number of warps we can run this with. Let me know if my changes broke anything else for you... Christian The 2014-02-18 release for Mac OS X is available here: http://www.johnchapman.net/crypto-currency/cudaminer-2-18-2014-release-now-available-for-os-x-10-6-10-7-10-8-and-10-9/
|
|
|
|
antonio8
Legendary
Offline
Activity: 1400
Merit: 1000
|
|
February 18, 2014, 11:16:49 PM |
|
Hey Christian,
At work but wanted to ask for when I get home.
Just got the MSI gtx750ti non powered OC. I am seeing from the results it already does the exact same as my 670 but a lot less power. Awesome.
Want would be a proper bat file since Maxwell is new? Should I use K or T setting in the bat? I want to try scrypt, jane and kecca to see how it competes with my 670.
I want to test it out with PTS also and see if it will do about 1,000 c/m like my 670. I will be thrilled if it does for the price of the card and the power draw. If it can I am going back to Fry's and picking up 3 more. I would really like the powered edition but this is all they have.
Thanks
Which Frys did you go to? The 750 isn't showing up on their website yet. Indianapolis. They had 8 in stock this morning. I had them price match Newegg and I got the rebate also.
|
If you are going to leave your BTC on an exchange please send it to this address instead 1GH3ub3UUHbU5qDJW5u3E9jZ96ZEmzaXtG, I will at least use the money better than someone who steals it from the exchange. Thanks
|
|
|
JimmyD
Newbie
Offline
Activity: 19
Merit: 0
|
|
February 18, 2014, 11:34:40 PM |
|
Sorry guys just wanted to ask - with the new Nvidia drivers released today (334.89) my hashing performance went down on my 2 machines. Did you have the same experience?
|
|
|
|
Pustul
|
|
February 18, 2014, 11:40:36 PM |
|
Those 750 looks promising. I'd try them if I had't just ordered 3x r9 270. R9 270: 475khs at 140W for 170$ = 2.79khs/$ but 0.29W/khs GTX 750: 300khs at 60W for 130$ = 2.30khs/$ but 0.20W/khs
I like the low consumption but problems are space and cost of the rig due to the low hashrate per card. Some kind of multiple gpu on one board like the asus mars would be nice.
PS: Prices are as found in Switzerland
|
|
|
|
I_M
|
|
February 18, 2014, 11:47:14 PM |
|
Christian, What kind of hopeful things do you think you can squeeze out with an M(axwell) kernel? I was looking at this, and it all seems pretty interesting.
|
|
|
|
cbuchner1 (OP)
|
|
February 18, 2014, 11:51:49 PM |
|
Christian, What kind of hopeful things do you think you can squeeze out with an M(axwell) kernel? I was looking at this, and it all seems pretty interesting. more Keccak (Maxcoin) and a good Blake256 (BlakeCoin) performance, hopefully. I still do not fully understand all architectural differences between Kepler and Maxwell. For scrypt I do not think that a significant leap can be made. We're memory limited here. the bigger L2 cache makes a difference, I guess. But you would already see this benefit in the current benchmarks (297 kHash/s overclocked which rivals my GTX 660Ti's ) I will have to study some Maxwell white papers and tuning guides, once published by nVidia. Christian
|
|
|
|
Morgahl
Member
Offline
Activity: 70
Merit: 10
|
|
February 19, 2014, 12:17:43 AM Last edit: February 19, 2014, 12:31:26 AM by Morgahl |
|
Soo.. TITAN BLACK edition anyone? I'd be interested to know how these get on!
basically a 780Ti with 6 Gigs of RAM and a slightly higher base clock. I'll pass. It will probably have a slight advantage in Yacoin, but the price tag is too high. Christian Doubt you will get better then a 780 Ti, you won't be able to properly allocate the beyond then 3 GB, unless the DAC is higher then 384 just won't work correctly, unless you can allocate and use two scratch pads on a single instance of CUDAMiner lol. 8 and I know someone maxing this out about the drivers? dunno. I remember having seen some forum posting (nvidia forums) of a guy having 14 cards recognized. But during operation there were issues. Christian I think it was 11 cards and he was using a expandable backplane that had 18 PCI-e slots avail. I'm confused - I thought I read that the cards had 640 cuda cores - the msi specifications show 960. (5x128=640 - they would need 7.5 SMM units to hit 960) 5*192=960 This said, I'm seeing NewEgg confused on this. The MSI cards are listed as 960 CUDA cores (the rest show 640) until you view the item itself and then it shows 640 Cuda Cores so your 5x128 is probably correct. Also CTRL-C terminates the program much faster when mining on a stratum server. Same for the --time-limit option.
Christian lol, yay for quick fixes The quotes for OCing the new Maxwell kind of bug me, I've never seen a benefit for up-clocking the the memory for Scrypt or Scrypt-Jane... I'd be interested how far you can push the core if the memory was floored to -502... and still stay under the 100% TDP limit. This one seems to think +650 Memory is needed?
|
|
|
|
apluscarp
Newbie
Offline
Activity: 12
Merit: 0
|
|
February 19, 2014, 12:43:03 AM |
|
anyone have a good config for scrypt on a 560 ti using 2014-2-9?
|
|
|
|
ManiacMiner
|
|
February 19, 2014, 01:31:30 AM |
|
anyone have a good config for scrypt on a 560 ti using 2014-2-9?
For scrypt on 560Ti try 2013-12-18. With this config -H 1 -i 0 -C 1 -m 0 -l F8x16 i have ~188kHash on this version, other versions ~165kHash
|
(つ ͡๏ ͜১ ͡๏ )つ[̲̅$̲̅(̲̅5̲̅)̲̅$̲̅]ε=ʕ ͡๏ ͜১ ͡๏ʔ=з
|
|
|
|
djm34
Legendary
Offline
Activity: 1400
Merit: 1050
|
|
February 19, 2014, 02:16:23 AM |
|
They had a chance over two... better luck next time
|
djm34 facebook pageBTC: 1NENYmxwZGHsKFmyjTc5WferTn5VTFb7Ze Pledge for neoscrypt ccminer to that address: 16UoC4DmTz2pvhFvcfTQrzkPTrXkWijzXw
|
|
|
sub-80
Member
Offline
Activity: 103
Merit: 10
|
|
February 19, 2014, 03:27:47 AM |
|
Started mining pandacoin getting 300khash/s with my 780. Is that the highest one can get with the 780?
This is my bat file: "cudaminer.exe --algo=scrypt:2048 -o stratum+tcp://panda.nitro.org:3338 -O user:pwd"
Launch config is T12x20 Bench hash/sec: 329142 hash/s Max total warps (BxW): 322
-i 1 -C 0 -H 0 (Default)
|
|
|
|
ouchmefoot
|
|
February 19, 2014, 03:32:07 AM |
|
Get 315-320 kh/s with a GTX680 (not overclocked, no cudaminer parameters) mining standard Scrypt
|
PostCoin:PYrnF6rLzvERtS73rRc76fpP4QCKP2WJS4
|
|
|
bacon_wrangler
Newbie
Offline
Activity: 26
Merit: 0
|
|
February 19, 2014, 03:45:13 AM |
|
This is weird: I just upgraded to Nvidia drivers 334.89 and to cudaminer 2-18, and get this on startup:
Unable to query CUDA driver version! Is an nVidia driver installed?
Running a 650 ti boost.
On second thought, bet a reboot fixes this.
EDIT: yep.
|
|
|
|
paulk11087
|
|
February 19, 2014, 04:02:41 AM |
|
so, i've been messing around with a 9800GTX and can't seem to figure this out... when i tried L32x8, it says "requires too much memory" and then switches to F32x8. i'm trying to mine MAX with this... it's going at roughly 3mh/s. does that sound right? anyone with a 9800 GTX get this working? haven't found any shares yet at this rate, and when i autotune it only hashes at about 1.2mh/s. my GPU load right now is sitting at 12% so i don't think it's really working.
any help would be great!
|
|
|
|
cwizard
Member
Offline
Activity: 98
Merit: 10
|
|
February 19, 2014, 04:11:06 AM |
|
I can get 10 zotac and 17 gigabyte out of D&H, but I read that you want cards with 6 pin power for use with risers.
|
|
|
|
sub-80
Member
Offline
Activity: 103
Merit: 10
|
|
February 19, 2014, 04:21:42 AM |
|
Get 315-320 kh/s with a GTX680 (not overclocked, no cudaminer parameters) mining standard Scrypt
For mining ltc and other scrypt coins seems to be in the range of the google spreadsheet Christian presented. The highest was a 415 Khash/sec.
|
|
|
|
|