[ANN] Opensource primecoin GPU miner beta-testing started! (Crossplatform)

[primer10]

@eXtremal.ik7

Btw, what is your rig like? Looks like you got good results

[zppwxz]

How to calculate the ch/day?
I Mining a night without any reaction。

** xpmclminer started 22-04-2014 14:37:05 192.168.1.106:9914 AR9dmNAWL1Siymqenf
** block: 505238, difficulty: 10.811

[1] 1.262G, average: 1.271G

• speed: 1.262G, average: 1.271G

  • chains/1: 37886777 11950.184/sec
  • chains/2: 2892549 912.363/sec
  • chains/3: 337843 106.562/sec
  • chains/4: 26591 8.387/sec
  • chains/5: 2120 0.669/sec
  • chains/6: 161 0.051/sec
  • chains/7: 10 0.003/sec 11.355/hour
  • chains/8: 2 0.001/sec 2.271/hour
  • chains/9: 1 0.000/sec 1.136/hour

22-04-2014 15:01:45: new block detected: 505211
22-04-2014 15:02:21: new block detected: 505212
22-04-2014 15:02:38: new block detected: 505213
22-04-2014 15:03:54: new block detected: 505214

[lnt900]

@eXtremal.ik7

i tried the latest build, still same error. but i figured out that because of my hardware. all of cards are R9 280x but not all the same. i removed 2 cards (which is different brand from the rest) from rig and the remaining 3 going well. that’s the problem, so what’s solution for this case ?

[minerador]

Hi everyone,

I can also confirm the speed-up of the new version on my R9 270X cards. Got from 1.5G to:

[code][1] 1.782G, average: 1.770G
[2] 1.792G, average: 1.770G
[3] 1.717G, average: 1.686G

• speed: 5.291G, average: 5.226G[/code]
  Still, 3 days running and not a single block found. Bad luck?

A suggestion to eXtremal.ik7
IMHO you should now focus on adding pooled mining instead of the performance improvements. This way you could set up a pool and let us all join the party. I believe it would be a win-win for everyone:
a) eXtremal.ik7 could set up a fair fee (3-5% ?) to reward him for his work. I believe everyone here would be willing to donate such a tiny amount.
b) Miners would all benefit from a steady income of XPM.
c) Primecoin itself would benefit from having less centralization. Right now there are only 3 pools working properly (ypool, beeeeer and XRamPool).

Yup, I know that setting up a Primecoin pool is not as straight forward as in other coins due to issues with fairness. Though others have done it and you could take inspiration:
https://github.com/thbaumbach/primecoin/tree/experimental
https://github.com/clintar/jhPrimeminer-Aero

What do you all think about this?

[okzx]

Also gpool.net and RapidPrime.com.

[lnt900]

diff too high or i have very bad luck ? all these 10 chains but found just 1 block

@eXtremal.ik7 : Due to XPM needs 3200 blocks to confirm, i sent you donation in LTC, many thanks and keep up your good works man

[primer10]

@Int900, unfortunately, all the groups of cards are working independently … you should try to group them into lesser group with more cards…

[zppwxz]

@primer10

@Int900, unfortunately, all the groups of cards are working independently … you should try to group them into lesser group with more cards…

How to group them into lesser group with more cards？

[lnt900]

@primer10

@Int900, unfortunately, all the groups of cards are working independently … you should try to group them into lesser group with more cards…

my main board have only 4 PCIE sockets, can’t group more of them. but all rigs are mining on same rpc wallet server on LAN.

I’m still can’t make all my 5 R9280X cards working on 1 rig got 5 PCIE sockets, so only 3 working now. someone suggest some idea for help please. don’t know the cause of that prob, i have 1 rig of 270x with VGAs from different brands still working fine, the 280X won’t work

@lnt900

i tried the latest build, still same error. but i figured out that because of my hardware. all of cards are R9 280x but not all the same. i removed 2 cards (which is different brand from the rest) from rig and the remaining 3 going well. that’s the problem, so what’s solution for this case ?

[primer10]

@Int900, zppwxz

Btw my rationale was that the wallet does not function like a pool (distributing work out) and any RPC clients connecting to it works independently …

How to increase the number of cards per system?

1. PCI-E Multiplier: https://bitcointalk.org/index.php?topic=367829.0
2. You prob would also need to daisy chain PSUs.

Obviously there is going to be a limit to how much you can place on one board.
http://www.asrock.com/mb/intel/h81%20pro%20btc/ -> 6 PCI slots
with 6 PCIE multipliers, you could have up to 24 cards.

[lnt900]

[quote=“primer10, post:90, topic:2223”]@Int900, zppwxz

Btw my rationale was that the wallet does not function like a pool (distributing work out) and any RPC clients connecting to it works independently …

How to increase the number of cards per system?

1. PCI-E Multiplier: Anyone using PCI-e Multipliers?
2. You prob would also need to daisy chain PSUs.

Obviously there is going to be a limit to how much you can place on one board.
http://www.asrock.com/mb/intel/h81%20pro%20btc/ → 6 PCI slots
with 6 PCIE multipliers, you could have up to 24 cards.[/quote]

thanks for the info, i’ll try to get some of them

btw, a lot of this, is it normal for over 10 chains just 1 block ?

ThreadRPCServer method=submitblock ERROR: CheckPrimeProofOfWork() : failed prime chain test target=0a.d08a00 length=(05.09e326 06.d68960 0a.09e326) ERROR: CheckProofOfWork() : check failed for prime proof-of-work ERROR: ProcessBlock() : proof of work failed

[zppwxz]

I think working in the mine pool is good, but do not support.

[eXtremal.ik7]

@SuperComputer

NVidia needs PTX Assembly for performance. OpenCL code running on NVidia hardware for large integer arithmetic seems slow.

Yes… sieve running fast on 780Ti, it shows same speed as Radeon R9 290, but multiprecision arithmetic too slow. I think, with PTX assembly 780Ti will be faster than R9 280X.

[quote=“zppwxz”]How to calculate the ch/day?
I Mining a night without any reaction。

** xpmclminer started 22-04-2014 14:37:05 192.168.1.106:9914 AR9dmNAWL1Siymqenf
** block: 505238, difficulty: 10.811

[1] 1.262G, average: 1.271G

• speed: 1.262G, average: 1.271G

  • chains/1: 37886777 11950.184/sec
  • chains/2: 2892549 912.363/sec
  • chains/3: 337843 106.562/sec
  • chains/4: 26591 8.387/sec
  • chains/5: 2120 0.669/sec
  • chains/6: 161 0.051/sec
  • chains/7: 10 0.003/sec 11.355/hour
  • chains/8: 2 0.001/sec 2.271/hour
  • chains/9: 1 0.000/sec 1.136/hour[/quote]
    You need about 3 days to found 1 chain with your GPU.

@lnt900

i tried the latest build, still same error. but i figured out that because of my hardware. all of cards are R9 280x but not all the same. i removed 2 cards (which is different brand from the rest) from rig and the remaining 3 going well. that’s the problem, so what’s solution for this case ?

Try make debug build (-DCMAKE_BUILD_TYPE=Debug) and show gdb output with last release from github, may be this helps me to find a bug.

[quote=“minerador, post:84, topic:2223”]I can also confirm the speed-up of the new version on my R9 270X cards. Got from 1.5G to:

[code][1] 1.782G, average: 1.770G
[2] 1.792G, average: 1.770G
[3] 1.717G, average: 1.686G

• speed: 5.291G, average: 5.226G[/code]
  Still, 3 days running and not a single block found. Bad luck?[/quote]
  How many 10-number chains found? With current diffuculty only 1 from 5 found chains is block.

[quote=“minerador”]IMHO you should now focus on adding pooled mining instead of the performance improvements. This way you could set up a pool and let us all join the party. I believe it would be a win-win for everyone:
a) eXtremal.ik7 could set up a fair fee (3-5% ?) to reward him for his work. I believe everyone here would be willing to donate such a tiny amount.
b) Miners would all benefit from a steady income of XPM.
c) Primecoin itself would benefit from having less centralization. Right now there are only 3 pools working properly (ypool, beeeeer and XRamPool).[/quote]
Now Claymore miner about 80% faster than xpmclminer, its not good.
After windows support (I already have windows build, it works except time measuring) and some speedup I will think about XPT protocol support… it allows a pool mining.

@primer10

Btw my rationale was that the wallet does not function like a pool (distributing work out) and any RPC clients connecting to it works independently …

By default all clients use their own block templates, block finding probability same with solo and pool mining. But pool pays you for all shares found (not only blocks), and your earnings on pool more steady…

@lnt900

btw, a lot of this, is it normal for over 10 chains just 1 block ?

No, normal is 2 blocks per 10 chains now

[lnt900]

about pool, i have this idea, just edit the code that make client submit from chain/8. then open a pool based on MPOS, edited to fit with protocol. reward can be calculate some like this :

• 1 chain/8 = 1 share
• 1 chain/9 = 12 shares
• 1 chain/10 = 150 shares.
• chain/10 is block = 250 shares.

Pool will verify and cache the submited shares, if duplicate with the ones has been submited will be mark as stale and then calculate PPLNS… so we can reach our goal that everyone mining will have their shares when blocks found.

Those estimate reward just for sample. i think when the miner going stable, the rest work to bring it in to pool ming is easy. surely the pool will grow up stronger than ypool because miner is free. it’s all on your hand to reach eXtremal.ik7. but if you not willing to do that i’ll do and cut you shares from mining fees

Try make debug build (-DCMAKE_BUILD_TYPE=Debug) and show gdb output with last release from github, may be this helps me to find a bug.

i’ll do that and report back to you ASAP

[gee4xpm]

@eXtremal.ik7

[quote=“gee4xpm, post:73, topic:2223”]but until now, i have not at least one 10-chain)

14 chains/9 and 0 chains/10, its unluck Usually 11 chains/9 = 1 chains/10.

@gee4xpmi

was wondering if i close miner for a while, that all my past work is gone/disappear? In another word, restart miner will clean all past work or will continue work?

No, if you close miner, you don’t loose any results (same as scrypt mining).[/quote]
nice…, thanks!

Finally, after 4 days mining, my first block was find! hope next block more luckly… donation send, thanks for ur work

[primer10]

@eXtremal.ik7

Yes… sieve running fast on 780Ti, it shows same speed as Radeon R9 290, but multiprecision arithmetic too slow. I think, with PTX assembly 780Ti will be faster than R9 280X.

How about 750Ti?

@eXtremal.ik7

Now Claymore miner about 80% faster than xpmclminer, its not good.

That’s not good …

@eXtremal.ik7

By default all clients use their own block templates, block finding probability same with solo and pool mining. But pool pays you for all shares found (not only blocks), and your earnings on pool more steady…

To put it this way, if we have multiple GPUs in one machine, your miner code will spread the calculation across these GPUs?

So if you have 1 machine with 10 cards, you are hashing at a faster rate. If you have 2 machines with 5 cards each, it is like 2 separate individuals with 2 machines trying to find a block. So, in this case, it is the same for pool mining but not so for solo.

[masterOfDisaster]

[quote=“minerador, post:84, topic:2223”][ …]
A suggestion to eXtremal.ik7
IMHO you should now focus on adding pooled mining instead of the performance improvements. This way you could set up a pool and let us all join the party. I believe it would be a win-win for everyone:
a) eXtremal.ik7 could set up a fair fee (3-5% ?) to reward him for his work. I believe everyone here would be willing to donate such a tiny amount.
b) Miners would all benefit from a steady income of XPM.
c) Primecoin itself would benefit from having less centralization. Right now there are only 3 pools working properly (ypool, beeeeer and XRamPool).
[ …]
What do you all think about this[/quote]

I appreciate the idea to reward eXtremal.ik7 for this great job.
But how exactly do you counter centralization if you combine a bunch of quite fast mining rigs that currently only operate solo into another pool?
If you want to stay decentralized, stay away from pools. With this software it’s possible, because of the high probability to find a block within a few days…
…until somebody integrates pooled mining this is decentralized!

sent by Tapatalk

[willi]

Hi,

I try to get this good-looking software to run on any of my nvidia cards - without success so far.
Example:

$ xpmclminer --opencl-platform "NVIDIA Corporation" -b
OpenCL platform available: NVIDIA Corporation
ptxas error   : Entry function 'sieve' uses too much shared data (0x4030 bytes, 0x4000 max)
ptxas error   : Entry function 'sieveBenchmark' uses too much shared data (0x4030 bytes, 0x4000 max)            
                              

The driver and card model is shown here:

$ nvidia-smi 
Wed Apr 23 12:32:05 2014       
+------------------------------------------------------+                       
| NVIDIA-SMI 334.21     Driver Version: 334.21         |                       
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GT 330M     Off  | 0000:01:00.0     N/A |                  N/A |
| N/A   46C  N/A     N/A /  N/A |    116MiB /  1023MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

On another workstation with a “Quadro FX 5800” (with 4GB memory) there is the same error message. I assume there is something wrong with my set of software?

dev-util/nvidia-cuda-sdk-5.5.22
dev-util/nvidia-cuda-toolkit-5.5.22
x11-drivers/nvidia-drivers-334.21-r3

Has someone an idea?

[lnt900]

eXtremal.ik7

here is the result when i run gdb, please take a look :

[New Thread 0x7fffe92e5700 (LWP 3655)] [New Thread 0x7fffe8ae4700 (LWP 3656)] [New Thread 0x7fffe3fff700 (LWP 3657)] [New Thread 0x7fffe37fe700 (LWP 3658)]

** xpmclminer started 23-04-2014 12:50:22 192.168.1.100:9912 Ae4hwE6fYgCDNpx8D1bhfdAyWtVtjTRJms worker 1398271677 **
** block: 507366, difficulty: 10.816

[1] -nanG, average: 0.000G
[2] -nanG, average: 0.000G
[3] -884801232607955887981948807806729667691698971088223541459146113072777881484371460611579566893981869853699337899177077006030208222529483283
[4] -585327843878432559527093970393985164076010236408273962226704802379848069847436185991991934096130644505353713414986718266014485558062208576
[5] -329167360415690763737246666005781972111261014350761181848292162016738797349462201682349756772287303904890649164409992489113881968835717622

• speed: -nanG, average: 0.000G
  1569216900201983170838528.000G, average: 0.000G

23-04-2014 12:50:22: new block detected: 507366

Program received signal SIGSEGV, Segmentation fault.
main (argc=9, argv=0x7fffffffe078) at /opt/xpmm1/src/OpenCL/xpmclminer.cpp:465
465 1, 0, globalThreads, localThreads, 0, 0, &event)) == CL_SUCCESS) {
(gdb) bt
#0 main (argc=9, argv=0x7fffffffe078) at /opt/xpmm1/src/OpenCL/xpmclminer.cpp:465

[eXtremal.ik7]

@lnt900

here is the result when i run gdb, please take a look :

Strange behavior… like a next AMD drivers bug. Try add line:

continue;

before

 if ((counter++ % 16) == 0) {

[quote=“willi, post:98, topic:2223”]On another workstation with a “Quadro FX 5800” (with 4GB memory) there is the same error message. I assume there is something wrong with my set of software?

dev-util/nvidia-cuda-sdk-5.5.22
dev-util/nvidia-cuda-toolkit-5.5.22
x11-drivers/nvidia-drivers-334.21-r3

Has someone an idea?[/quote]
NVidia Fermi GPUs have less shared memory than Kepler and Maxwell. Wait a patch that makes a sieve window size configurable.
But I think these GPUs too slow for XPM solo mining…

[quote=“masterOfDisaster, post:97, topic:2223”]But how exactly do you counter centralization if you combine a bunch of quite fast mining rigs that currently only operate solo into another pool?
If you want to stay decentralized, stay away from pools. With this software it’s possible, because of the high probability to find a block within a few days…
…until somebody integrates pooled mining this is decentralized![/quote]
Litecoin mining now is pool-only (I don’t know LTC solo miners), but it is decentralized… There are no pools that can do 51% attack.
When XPM diffuculty reach 11, solo mining will start to die Radeon R9 290X searches chain/11 in 5-6 days continuous work.

[quote=“primer10, post:96, topic:2223”]To put it this way, if we have multiple GPUs in one machine, your miner code will spread the calculation across these GPUs?
So if you have 1 machine with 10 cards, you are hashing at a faster rate. If you have 2 machines with 5 cards each, it is like 2 separate individuals with 2 machines trying to find a block. So, in this case, it is the same for pool mining but not so for solo.[/quote]
Each GPU uses its own extra nonce for block header building, no difference between one rig with 10 GPUs and two rigs with 5 GPUs in each.

@primer10

How about 750Ti?

I don’t know about new Maxwell features and can’t use it in my OpenCL code… With current version, Maxwell GPUs shows same performance as Kepler.

[quote=“lnt900, post:94, topic:2223”]about pool, i have this idea, just edit the code that make client submit from chain/8. then open a pool based on MPOS, edited to fit with protocol. reward can be calculate some like this :

• 1 chain/8 = 1 share
• 1 chain/9 = 12 shares
• 1 chain/10 = 150 shares.
• chain/10 is block = 250 shares.[/quote]
  Pools for sha256/scrypt uses similar fee calculation algorithms But I don’t now about existing XPM pools (ypool, xrampool, etc)…