Author

Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 1059. (Read 3426921 times)

newbie
Activity: 20
Merit: 0
Looks like my GTX580 doesn't like the latest build (both x64 and x86). For some reason the core temperature increases with ~8 degrees, while performing slightly slower. GPU load is 99% all the time.

 Cry
newbie
Activity: 10
Merit: 0
Peaked out at 643 KH/s!!!!!!

You and Dave are the freakin BOMB! Now I have to really fine tune my overclock since this actually uses 100% of my TDP, and temps are about 10 deg higher.

Previous version I was maxing out at 433 kh/s with 112x2. Now with this version, autotuning used K14x4 for GPU1 and K14x14 for GPU2. Set it to -l K14x14 for both and I'm getting 643 kh/s from 2 EVGA 670 FTWs!

-H 1 -i 0 -C 1 -m 1 -l K14x14
sr. member
Activity: 408
Merit: 250
ded
My 660Ti was getting 215 (+/- 5) khash/s  before.
Started new version and was getting 265 (+/- 5).
I left my computer for ~40 mins and now its at a steady 285 (+/- 5) !!!

would you share your kernel launch configuration there? Mine currently only does 247 kHash/s.

Christian


auto, I used: -i 0 -H 1 -C 2 -D -m 1
I think the kernel conf was K5x12 but not too sure, so I stopped and started it again using the same settings as before. This time it generated K31x6 and its running at 310 khash/s, dropping down 2-10  every so often but bouncing right back up to 310.

Not too sure whats going on lol

Will post an update if something happens
hero member
Activity: 756
Merit: 502
My 660Ti was getting 215 (+/- 5) khash/s  before.
Started new version and was getting 265 (+/- 5).
I left my computer for ~40 mins and now its at a steady 285 (+/- 5) !!!

would you share your kernel launch configuration there? Mine currently only does 247 kHash/s.

Christian

sr. member
Activity: 408
Merit: 250
ded
My 660Ti was getting 215 (+/- 5) khash/s  before.
Started new version and was getting 265 (+/- 5).
I left my computer for ~40 mins and now its at a steady 285 (+/- 5) !!!

Great Work!
sr. member
Activity: 388
Merit: 250
Sitting at around 178kH on autotune with the new version with my GTX 660, compared to 140 with optimised settings on the previous version and around 125 with autotune on the previous version. Going to put in the previous settings and see if it makes a noticeable difference.

Edit: Sitting around 196kH with optimised settings, so it's approximately a 40% jump in output compared to the previous version. Very pleased with the performance. Cheesy
sr. member
Activity: 378
Merit: 250
i see massive jump on gtx 760 from 152 kh/s to 250 kh/s.

thanks

-H 1 -d 0 -i 1 -l auto -C 2 -m 1
member
Activity: 98
Merit: 10
Seriously, what is the deal with the desktop. My cards are cranking like they've never before and desktop is more responsive than the old -i 1 mode even though I am set at -i 1?

I'm noticing this too. There's slightly less stutter on video (had Pawn Stars on this whole time while testing configs). Not the very slight nearly constant stutter I've always seen, but the larger occasional stutters seem to be gone.
full member
Activity: 173
Merit: 100
Seriously, what is the deal with the desktop. My cards are cranking like they've never before and desktop is more responsive than the old -i 1 mode even though I am set at -i 1?
member
Activity: 98
Merit: 10
Settings: -i 1 -l K90x2 -C 1 -H 1 -m 1

You should really autotune again. The x2 is probably inefficient. Also the -C (texture cache) argument is currently ignored with David's new kernels. I will try to integrate this feature again if it brings an additional speed benefit.


Yeah, it was just a quick check before I went into a repeated autotune frenzy.

Speaking of which, there's some really good news on that front. With K42x5, I'm now at a whopping 230khash, still on interactive mode! Incredible job here!

Side question, do you have a Bitcoin addy? I don't mine LTC.

EDIT: 28x8 is at 235khash. Still have a few more to check out, but this is probably where I'm going to top out.

EDIT2: The oddball 14x24 is just a hair faster at 236.
newbie
Activity: 28
Merit: 0
My hashrate improved by about 8% on a Fermi kernel, so I'm not complaining.
hero member
Activity: 756
Merit: 502
sr. member
Activity: 406
Merit: 250
No love for fermi cards? Smiley
hero member
Activity: 756
Merit: 502
Settings: -i 1 -l K90x2 -C 1 -H 1 -m 1

You should really autotune again. The x2 is probably inefficient. Also the -C (texture cache) argument is currently ignored with David's new kernels. I will try to integrate this feature again if it brings an additional speed benefit.
full member
Activity: 173
Merit: 100
Went from 152 per card to 212!!! Actually uses the max power limit now. Cards are hotter but only 66c, before they were 60c.

K5x12 on two MSI GTX 660 TF3/OC cards.

THANK YOU!!!!

edit: Bumped up fans to 56% and I am back under 60c. Desktop performance is as if I don't have cudaminer running at all! Amazing update!!
hero member
Activity: 756
Merit: 502
And having looked for hours, having posted the above I find a way that should query those things.

/usr/bin/nvidia-smi -a

Sadly it is mostly N/A with my card (GTS 250)

https://github.com/CFSworks/nvml_fix

fixed this for me. I had to install gcc 4.4 to compile this hack.

Christian
hero member
Activity: 756
Merit: 502
Any advice on x64 vs x86 with the new version?

I stick to x86, as generally the register pressure is lower (CUDA kernels run faster) and the CPU is mostly idle, anyways.
member
Activity: 98
Merit: 10
Any advice on x64 vs x86 with the new version?

EDIT: With a 670 GTX and the same settings as before, I bumped from 160ish khash to 190 khash. Awesome work, cbuchner! Also, thanks to David Anderson!

Settings: -i 1 -l K90x2 -C 1 -H 1 -m 1
hero member
Activity: 756
Merit: 502
Here are some configs that work for me (there may exist better ones, of course)

GTX 780Ti: -l T15x24    550 kHash/s  (106% TDP, +100 core clock, +200 mem clock)
GTX 660Ti: -l K14x8      250 kHash/s
GT 640 (GK208 based): -l T9x20     92 kHash/s

all of this is significantly above what previous cudaminer releases would achieve.

Careful, your cards may run hotter than before.
hero member
Activity: 756
Merit: 502
Does the massive update on GitHub mean a release is close HuhShocked Shocked

I present the 2013-12-18 release with all new Kepler, Titan kernels. Huge thanks to David Andersen who came up with a more efficient way to do scrypt mining on the Kepler architecture.

We now use CUDA 5.5, which has increased driver requirements. Be sure to use a recent nVidia driver.

When mining on Kepler devices, please autotune again and report your findings.

Christian

Jump to: