Author

Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 818. (Read 3426922 times)

hero member
Activity: 938
Merit: 1000

- Calculating prime number with GPU is a given problem from long ago. Almost a few decades and nobody was able to do it. I don't think we will have it soon.


I wouldn't make so broad statements.

I have implemented a GPU accelerated sieve of erathostenes myself as a tutorial project, and it was faster than on the CPU. But it was for small prime numbers (think <  2^^32) and the acceleration wasn't that great.

But doing the same for BIG numbers (integers with 1024 bits or so) requires specialized maths libraries, ported over to GPU. This is effort.

there is prior art in this area though: the gpuPrime and prime grid projects (I think its a Boinc project). Maybe one could take some inspiration from there (and also code, license permitting)

Christian


You're right, there are some improvements in this field but in general it doesn't improve much for a few decades. Prime number is still tough for GPU, except we totally change the current GPU design.
Sorry if I made some broad statements but IMO it is still nothing in comparison with the speed up of another GPUminers
member
Activity: 79
Merit: 10
my card is gtx780 How can I make a bat to mine max and utc? please give me two bat Thanks.

I exchange two bat for two coins.Hey body do u have utc max and vet bat

hero member
Activity: 756
Merit: 502
[2014-02-10 18:30:12] Accepted: 5035/5046 (100%), 86787 kH/s  (yaaaaaay!!!!!!!!!)

I would say commit 176 is perfect!

and a single boo! will spoil the perfection Wink  oh wait you already had 11 boo's, but they were drowned in rounding the percentage.
legendary
Activity: 2002
Merit: 1051
ICO? Not even once.
[2014-02-10 18:29:45] Accepted: 5031/5042 (100%), 86678 kH/s  (yaaaaaay!!!!!!!!!)
[2014-02-10 18:30:04] Accepted: 5032/5043 (100%), 86833 kH/s  (yaaaaaay!!!!!!!!!)
[2014-02-10 18:30:04] Accepted: 5033/5044 (100%), 81057 kH/s  (yaaaaaay!!!!!!!!!)
[2014-02-10 18:30:05] Accepted: 5034/5045 (100%), 83938 kH/s  (yaaaaaay!!!!!!!!!)
[2014-02-10 18:30:12] Accepted: 5035/5046 (100%), 86787 kH/s  (yaaaaaay!!!!!!!!!)

I would say commit 176 is perfect!
hero member
Activity: 756
Merit: 502
@cbuchner1

Good work on your CUDA miner for maxcoin on such a short notice.

Nvidia's GTX 780 Ti Classified is about 8x (overall) faster than Intel's i7-4770K (Haswell using AVX2 instructions)
Sieving performance on the Ti was a little bit disappointing when compared to Haswell.

Karatsuba followed by Montgomery Reduction seems to work best on Haswell (GMP modular exponentiation)

1) Thanks ;-)

2) this is what I was just saying. The expected speed-up over CPU isn't that great. 8 times is respectable tough.

3) hey, I did a 1024 bit karatsuba multiplication once, based on 24 bit integer multiplication intrinsics for Compute 1.x devices. I wanted to use it for some deep zoom Mandelbrots, but never got around to actually code up some Mandelbrots around it.
hero member
Activity: 756
Merit: 502
Hey Christian, just a question


Are you thinking about implementing Protoshare or not at all ?

nope. DGA has done it (cudapts) and I couldn't do it any better.

https://bitcointalksearch.org/topic/ann-cudapts-gpu-mining-for-protoshares-pts-404915

Plus it's by definition a memory hard problem and doesn't provide a huge advantage over the CPU when run in CUDA
newbie
Activity: 59
Merit: 0
Hey Christian, just a question


Are you thinking about implementing Protoshare or not at all ?
hero member
Activity: 756
Merit: 502

- Calculating prime number with GPU is a given problem from long ago. Almost a few decades and nobody was able to do it. I don't think we will have it soon.


I wouldn't make so broad statements.

I have implemented a GPU accelerated sieve of erathostenes myself as a tutorial project, and it was faster than on the CPU. But it was for small prime numbers (think <  2^^32) and the acceleration wasn't that great.

But doing the same for BIG numbers (integers with 1024 bits or so) requires specialized maths libraries, ported over to GPU. This is effort.

there is prior art in this area though: the gpuPrime and prime grid projects (I think its a Boinc project). Maybe one could take some inspiration from there (and also code, license permitting)

Christian
legendary
Activity: 2002
Merit: 1051
ICO? Not even once.
compulsory hoarding, eh?

I'm always so happy to see a new commit that I just drop everything and compile it  Cheesy
sr. member
Activity: 278
Merit: 250
@cbuchner1

Good work on your CUDA miner for maxcoin on such a short notice. Somehow I missed the party, but I hope that it is not too late to join in without purchasing an AMD GPU.

I already implemented Primecoin/Datacoin mining on an Nvidia GPU. I started with a GTX 580 Hydro @ 1.8 GHz (shader) but now running on GTX 780 Ti Classified @ 1.02 GHz with 1.3x speed-up over the former. Here is some helpful information for those who are interested:

Nvidia's GTX 780 Ti Classified is about 8x (overall) faster than Intel's i7-4770K (Haswell using AVX2 instructions)
Sieving performance on the Ti was a little bit disappointing when compared to Haswell. In my opinion, Haswell has the best L1/L2 cache implementation on a CPU. Hybrid implementation works best.
Bit array works best on the Ti for sieving but you will need to use the atomicOr function wisely.
Multiple-precision at 320-bit and at 384-bit using Montgomery Reduction (CIOS) works best. (inline PTX  assembly)
Karatsuba followed by Montgomery Reduction seems to work best on Haswell (GMP modular exponentiation)
I cheated a little with the modular exponentiation on the GPU, I performed a partial reduction on the CPU first to reduce the 50% branching penalty on the GPU.

My code is closed source for now but I will be more than happy to help with an open source version when time permits. Also, because of how I implemented the modular exponentiation on the GPU, it can not be used in a pool without a lot of work.

I do not have an AMD GPU, but after looking at the design specifications, this is one area that the AMD GPU guys will be playing catch-up.
legendary
Activity: 1400
Merit: 1050
Jesus. BTC just took a tumble  Huh

Either someone had fat fingers and entered the wrong number, bots got confused, or a bunch of coins were stolen and dumped. Today's going to be a rough day...
It was worst this morning... I went to buy a new graphic card, at first the bitcoin was at 413euros... hopefully when I was about to pay, it was at 470euro... well this isn't a really good deal but still...
Finally bought a MSI R9-290x Gaming edition... hope there won't be any problem to have it in the same computer as the gtx780ti...


where are you currently buying hardware for bitcoin in the Euro zone?

Christian

It is achatnet.pro (a french site... rather surprise it even exists...)
hero member
Activity: 756
Merit: 502
yes, that is a github feature. I think they have RSS feeds and email notification.

I commit a lot of dysfunctional crap though,  so be warned. Wink


I just checked and my cudaMiner folder takes up 13 GB's and I have 165 different cudaMiner binaries compiled since December.

compulsory hoarding, eh?

hero member
Activity: 938
Merit: 1000
cbuchner1: Awesome work on the maxcoin release. I will surely be sending over some coins when I get my internet back (bad weather here, just barely online via mobile). You make the 'green team' proud!  Grin

Earlier there was a post about new/next cudaminer development opportunities, and I have two suggestions to make:

Protoshares - popular, highly valued, already has a gpu miner available (only binary I think), though it appears to be far from reaching optimal performance on nvidia, so likely opencl only at the moment. My single GTX 770 earns me a little over .5 PTS a day (1GH pool) and only using 70% power. (1PTS trades for 0.015BTC)

Primecoin - popular, currently CPU only (suspecting this one would prove to be a challenge in terms of optimizing for GPU at all, but perhaps cuda specifically would make it worthwile). It's not highly valued, but stable and also very popular. The potential on this one is for a GPU implementation to break into new performance figures, thus making it largely profitable for a relative time.

I hope this comes across as helpful. I would love to help with code but lack the skill-set, so instead I'm looking to save Christian's time in going through all the altcoins to work out the most interesting (or challenging) opportunities ahead. I'll be happy to to test any development version to support new coins, however broken or buggy a release is - but I only take binaries though Smiley

On Riecoin that was just mentioned: It's technically very similar to Primecoin, but yet to be launched. Interesting opportunity if it is found to reach a good coin value for the relative performance in GPU mining, but I don't see that it has the hype or following to produce the same returns as the ones I mentioned above (or even a fraction of Maxcoin for that matter).

Cheers!

- PTS cudaminer is originally opensource from dga. 1gh GPU miner is just a derivation from it, I suspected (like cgminer_keccak inspired by our mighty Christian's cudaminer). Try to inform yourself first before making some statements.
- Calculating prime number with GPU is a given problem from long ago. Almost a few decades and nobody was able to do it. I don't think we will have it soon.
legendary
Activity: 2002
Merit: 1051
ICO? Not even once.
yes, that is a github feature. I think they have RSS feeds and email notification.

I commit a lot of dysfunctional crap though,  so be warned. Wink


I just checked and my cudaMiner folder takes up 13 GB's and I have 165 different cudaMiner binaries compiled since December.
hero member
Activity: 756
Merit: 502
my card is gtx780 How can I make a bat to mine max and utc? please give me two bat Thanks.

I exchange two bat for two coins.
sr. member
Activity: 350
Merit: 250
Scan.co.uk
Aria.co.uk

Both accept bitcoin
hero member
Activity: 756
Merit: 502
Jesus. BTC just took a tumble  Huh

Either someone had fat fingers and entered the wrong number, bots got confused, or a bunch of coins were stolen and dumped. Today's going to be a rough day...
It was worst this morning... I went to buy a new graphic card, at first the bitcoin was at 413euros... hopefully when I was about to pay, it was at 470euro... well this isn't a really good deal but still...
Finally bought a MSI R9-290x Gaming edition... hope there won't be any problem to have it in the same computer as the gtx780ti...


where are you currently buying hardware for bitcoin in the Euro zone?

Christian
hero member
Activity: 756
Merit: 502
Christian, is it possible to subscribe to your git account, so that I get an update for the new versions?

yes, that is a github feature. I think they have RSS feeds and email notification.

I commit a lot of dysfunctional crap though,  so be warned. Wink
hero member
Activity: 756
Merit: 502
With that being said, mining UTC at 400kh/s gives me 0.016btc/day, VTC at 300kh/s 0.014btc and maxcoin at 200mh/s 0.017btc a day vs scrypt's 0.012btc/day with the absolute best coins and a way higher power draw. (UTC is really light on power, like all scrypt jane coins, same with max, vtc and protoshares). Yeah, forgot about protoshares, that's pretty dan good too: 2100cpm = 0.73pts/day.

that is a really nice summary - and it shows how close the cryptocurrencies get to each other in terms of profitability. People jump to whereever it's the most proftable - equalizing the profitability rather quickly.
legendary
Activity: 2002
Merit: 1051
ICO? Not even once.
Thanks for the help guys...

Cudaminer only crashes when mining vtc or utc, maxcoin is fine...1150mv 1306mhz core and 3665mhz mem, 70c or less whilst mining.

From experience, if it crashes after several hours, the OC is too high.
Also, different algorithms need different OC profiles because they are not stressing the card the same way.

Some examples (not exact values):
Max core OC running 3D benchmark (rough estimate with older BIOS): ~60 Mhz before crashing;
Max core OC running scrypt mining: ~80 Mhz;
Max core OC running scrypt-jane:14: ~120 Mhz;
Max core OC running keccak: ~132 Mhz.
Jump to: