Pages:
Author

Topic: [XPM] Working on a GPU miner for Primecoin, new thread :) - page 17. (Read 166591 times)

sr. member
Activity: 364
Merit: 253
I'm only going to say the following and then I'm done.  I said the same thing in #eligius-prime.

Everyone donated because mtrlt led them to believe his miner was going to be faster than hpX.  He knew this wasn't the case and he knew that certain people were going to leak his code.  Yet, he kept accepting "donations" up until the very end, September 2nd.

If he had any decency, he would have said, "I know all of you donated in hopes that the gpu miner would be faster than hpX.  Unfortunately, (due to whatever reason, personal issues, lack of time, too hard), the software I produced is slower than hpX. Therefore, I'm giving back everyone's money, because the "early beta access" that I promised is now worthless, because you can get better performance out of hp10."

Who in their right mind would want "early beta access" to an inferior miner?

If individuals wanted to give him donations in return for nothing, because he put time into coding a miner.  Then, that's one thing.  But, that's not how he marketed it.  He was giving "early beta access" in return for donations.  He knew the miner was inferior before he released it.  Why would he release it before it was capable of providing the "early beta access" advantage to the people who supported him?

How much did you donate that you wanted your bitcoins back?
sr. member
Activity: 644
Merit: 250
I'm only going to say the following and then I'm done.  I said the same thing in #eligius-prime.

Everyone donated because mtrlt led them to believe his miner was going to be faster than hpX.  He knew this wasn't the case and he knew that certain people were going to leak his code.  Yet, he kept accepting "donations" up until the very end, September 2nd.

If he had any decency, he would have said, "I know all of you donated in hopes that the gpu miner would be faster than hpX.  Unfortunately, (due to whatever reason, personal issues, lack of time, too hard), the software I produced is slower than hpX. Therefore, I'm giving back everyone's money, because the "early beta access" that I promised is now worthless, because you can get better performance out of hp10."

Who in their right mind would want "early beta access" to an inferior miner?

If individuals wanted to give him donations in return for nothing, because he put time into coding a miner.  Then, that's one thing.  But, that's not how he marketed it.  He was giving "early beta access" in return for donations.  He knew the miner was inferior before he released it.  Why would he release it before it was capable of providing the "early beta access" advantage to the people who supported him?
legendary
Activity: 1901
Merit: 1024
it depend on your cpu, but for me using CPU for fermat test speed it up by 2x...
usinig 6 threads for i5
2x 7950
max i got is ~650 K 2-chains/s

sieva percentage ~18 it does or default almost same, rest all same

If i use only GPU 2x 7950 max i got is ~350 k 2-chains/s

So how long do you typically get a block at this difficulty with that setup?

I don`t know, atm is really hard to get block at diff so high, you keep in mind that 9.0 (2-3 week ago) and ~9.85 is like >800% harder
sr. member
Activity: 364
Merit: 253
it depend on your cpu, but for me using CPU for fermat test speed it up by 2x...
usinig 6 threads for i5
2x 7950
max i got is ~650 K 2-chains/s

sieva percentage ~18 it does or default almost same, rest all same

If i use only GPU 2x 7950 max i got is ~350 k 2-chains/s

So how long do you typically get a block at this difficulty with that setup?
legendary
Activity: 1901
Merit: 1024
it depend on your cpu, but for me using CPU for fermat test speed it up by 2x...
usinig 6 threads for i5
2x 7950
max i got is ~650 K 2-chains/s

sieva percentage ~18 it does or default almost same, rest all same

If i use only GPU 2x 7950 max i got is ~350 k 2-chains/s
member
Activity: 61
Merit: 10
Ok now that its out, time to get a github going to bring this project to completion.
sr. member
Activity: 364
Merit: 253
please guys post your PPS and setup information so we could see how it works.
newbie
Activity: 41
Merit: 0
It's already released, it's safe and you can compile it yourself. Go grab it: http://cryptobro.com/reaperprime.zip

Edit:
Don't forget to say "Thanks" to all the good people who donated, without their sacrifice the rest of us would not have this.

No one has anything now!  This miner is garbage. He didn't develop anything worthy.

Calm down man, I was just ironic pissing off those greedy fools who hoped to gain advantage over somebody by donating. Hats off to honest donators who did it not only for themselves but for the community. Asking for donations back is so ridiculous beyond words to describe.

I'm sorry, but you must be a complete fool.  Why would I donate money to develop something without reaping a financial gain???  It's not like he's saving babies or feeding the homeless.

Mtrlt obviously made a large financial gain off of this scam.  Why is it so terrible that everyone that gave him money expected a financial gain?

I donated because I expected him to make an effort to deliver a decent product.  It's obvious that he didn't even try.

2Kool4Skewl -> Have you ever even been associated with a software development project? You're whining about performance on a early release of software? LOL

Mtrlt, hats off to you, I am quite impressed with the progress you've made, and I hope to see improvements on this fine accomplishment.
full member
Activity: 147
Merit: 100
seems things are coming along nicely
member
Activity: 104
Merit: 10
1/9 seems about right. With use_gpu_fermat_test, it reports the amount of candidates left after doing Fermat tests on the GPU. Without it, it just reports whatever the GPU sieve returned.
full member
Activity: 210
Merit: 100
I not use any kind of messenger beware of scammers
The TOTAL counts aren't comparable between use_gpu_fermat_test true/false, should look at 2-chains. I'll remove the TOTAL count in the next version.
Good to know, thanks! Looking at 2/3 chains then the gpu_fermat_test is faster by ~ 1.4x (I do have a weak CPU in there).  Is the candidate difference a red herring (1/9 as many) or just the way it is reported?
member
Activity: 104
Merit: 10
The TOTAL counts aren't comparable between use_gpu_fermat_test true/false, should look at 2-chains. I'll remove the TOTAL count in the next version.
full member
Activity: 210
Merit: 100
I not use any kind of messenger beware of scammers
Since this is pretty much public domain now, any chance of getting the code put on sourceforge/git? Makes it easier to see what has changed between releases.

---
Thanks for the new stats, it definitely points out what we have noticed: When the fermat test (use_gpu_fermat_test) is done on the GPU it is quite a bit slower.

use_gpu_fermat_test false : 94ms for GPU stuff! 1899 candidates found!    (9209.4 fermats/s, 811.647 gandalfs/s. 52.7538M TOTAL)
use_gpu_fermat_test true  :  327ms for GPU stuff! 228 candidates found!   (0 fermats/s, 1201 gandalfs/s.  11.6282M TOTAL)

Throwing away more candidates?

(self compiled 32-bit exe version of reaper beta 2, Windows 7 x64, 2 x 7950, Cata 13.5 beta ).

Time to start tweaking I think (would like to find a way around OpenCL 1.2 requirement, Nvidia is not planning on supporting that, especially on Ubuntu )

Crashing also reminds me a lot of the issues with getting cgminer stable with all the different catalyst releases (likes to crash if something is not right). Could easily be a driver issue.
member
Activity: 104
Merit: 10

It is hard to get the entire output as once the program froze enough text has gone by that the original details are gone.
Is there anyway to capture the whole output in an easy way?
(I have little knowledge within coding and such, so please be patient Tongue)
//DeaDTerra
Yeah just run it from the command line / terminal like this:
Code:
reaper > output.txt
And it'll write the output to the file output.txt.
donator
Activity: 1064
Merit: 1000
DeaDTerra, could you try with:
Code:
cpu_mining_threads 1
debug true
opencldebug true
and put the entire output to pastebin.com?
It is hard to get the entire output as once the program froze enough text has gone by that the original details are gone.
Is there anyway to capture the whole output in an easy way?
(I have little knowledge within coding and such, so please be patient Tongue)
//DeaDTerra
member
Activity: 104
Merit: 10
DeaDTerra, could you try with:
Code:
cpu_mining_threads 1
debug true
opencldebug true
and put the entire output to pastebin.com?
donator
Activity: 1064
Merit: 1000
For all the people with crashes, dll issues aside, I found that deleting the generated .bin files, setting "save_binaries no" in reaper.conf and worksize to 256 in primecoin.conf reduces crashes to a minimum, at least on my system.

YMMV

cK'
I gave it a try, didn't lead to any improvements unfortunately.
//DeaDTerra
full member
Activity: 213
Merit: 100
I'm getting "The procedure entry point clEnqueueFillBuffer could not be located in the dynamic link library" and crash. Win8 x64 & GTX 560

You need to upgrade your amd/nvidia drivers to fix that one, but it will probably still crash after that.

Same here on Win8 x64 Nvidia C2075. The driver was updated in July and should support  openCL 1.1

You assume clEnqueueFillBuffer works in OCL 1.1

It seems clEnqueueFillBuffer indeed requires OpenCL 1.2. I could do without it but it's kind of a pain in the ass. Nvidia doesn't have drivers that support OpenCL 1.2?

Ahhhh so this is why it crashes on OSX 10.8, which ships with OpenCL 1.1 and there is no newer driver available.

According to Wikipedia, when OS X 10.9 Mavericks ships this Fall, OpenCL 1.2 and OpenGL 4.1 will be built in.

So people with Nvidia cards are just SOL until 10.9 comes out?
hero member
Activity: 639
Merit: 500
Somebody, please, make a working 64 bit binary file.
Are you using Linux or Windows?
Both. Windows on first PC and Ubuntu 13.04 on second.
newbie
Activity: 35
Merit: 0
For all the people with crashes, dll issues aside, I found that deleting the generated .bin files, setting "save_binaries no" in reaper.conf and worksize to 256 in primecoin.conf reduces crashes to a minimum, at least on my system.

YMMV

cK'
Pages:
Jump to: