Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 48. (Read 3426930 times)

now if undervolting, you need to down clock otherwise it will crash very often...
actually instead of undervolting, you can decrease the max tdp as everything is taken care by the card...

Thanks, i'll have to try it.

Undervolting does not hurt cards.

Doesn't undervolting hurt the cards?  Do you get any shorter life of the cards doing that?

Thank you bigjme. That's for my next rig.
Haven't seen you around for a while.  

I believe the single 970 wins if your using enough 750tis to need to MB's but its been a while

Wow 

Forgetting the price of the card, is the 970 and the 980 the new 750ti? Meaning watts per hash. If you had a free choice, would you take 3 750ti or 1 970 or 1 980 keeping in mind watts per hash only.

.20 USD per kWh here in the uk

How big and what kind of PSU do you have?

Just curious I had a problem with one before that caused all kinds of issues until I replaced it.

.16 USD per kWh here. 

A bit off topic, but I have a card which keeps crashing not just the miner, but the whole computer after a few minutes/hours randomly;
The card is a 780 Ti sitting with 5 x 750 Ti and whatever I do or whichever fork I use, it crashes the miner which start stressing the cpu at 100% (at which point the GPU usage is 0% for all cards) and after a few minutes without forcibly closing the miner, they PC just freezes, no BSOD, no restart, just completely loses interactivity. If I close the miner before that, the System process completely hogs one CPU core (25%, i3) and the PC becomes very unresponsive and it can barely do anything until a restart. Tried different risers, a few different drivers, replaced the GPU BIOS, downclocked but none of that helped. It only does that during mining and the PSU can't be the culprit.

The event log wasn't much of a help:
The description for Event ID 13 from source nvlddmkm cannot be found.
Graphics Exception: ESR 0x506224=0x80000041 0x506228=0x10a40f 0x50622c=0x280 0x506234=0x0
NVRM: Graphics TEX Exception on (GPC 0, TPC 4):     TEX NACK / Page Fault

Edit: The latest official BIOS (F4 - 2014/09/19) which has an 'Enhance stability' entry in the change log, so far seem to fix the issue, at least I hope so. Still not sure why an another exact same card haven't had this issue, though.

Just to check, is there a NeoScrypt miner for CUDA, or are we stilled bound to the CGminer?

Hey guys,

I'm trying to download v1.2U-D8 - 25-08-2014 from NVminer from Cudamining.co.uk, but it keeps telling me that the 'Tonido Enabled Device Unavailable' and I can't download it.
Could this please be fixed, and could someone link me a working version, please?

Edit: The DJM34 download of m7v7 does work, so I'll be using that.

Things have changed a lot :p I don't know half the names posting haha.

Neoscrypt looks interesting but I don't see me using my gpu's any time soon :/

I meant all of you. Since Bigjme has not been posting we have had new miners/versions from DJM, SP, Epsylon.

You have done a cracking job with the Neoscypt miner though.........now if only we could persude u to come back to CUDA and optimize the 970/980.

sorry for the link(I was on mobile when I was writing that post): here is the correct one: http://wwwmath.uni-muenster.de/u/cl/WS2007-8/mult.pdf

It is a novel use of FFT for integer multiplication. (based on  the Schönhage–Strassen algorithm)

These guys just recently published a improvement to Furer's algo:  http://arxiv.org/ftp/arxiv/papers/1407/1407.3360.pdf

Yes FFT is a hog, but with CUDA 6.5 performance has been increased: http://devblogs.nvidia.com/parallelforall/10-ways-cuda-6-5-improves-performance-productivity/

Does this mean that it will work? I don't know; It depends upon the application.

part of the problem... need to decide which algo depending of the lenght...
we need to write gmpg (gmp for gpu)

I can't access the article, it says page not found...
The problem with divide and conquer is to find an economical way (in terms of memory load/store) and efficient (don't want to rewrite everything for each number) to split the operands. and for the moment this 2 parts are a big no go.
May-be fft is the way, seems easier  ?

was looking at the table with calculation on cuda, there are some really strange things in 50 compared to 35

I think they are the same.

But
 If we use the divide and conquer algorithm proposed by Dr. Fürer described here: http://wwwmath.uni-muenster.de/u/cl/WS20078/mult.pdf

Modular Multiplication. Maybe?