Pages:
Author

Topic: Open Source XPM (Primecoin) GPU Miner & Pool xpmforall.org - page 46. (Read 110139 times)

legendary
Activity: 1510
Merit: 1003
eXtremal
Any news about otimization for NVIDIA ?
same question
with opencl it is very hard (nearly impossible) to reach top performance on nvidia. Better ask for cuda primecoin miner )))
Hix
legendary
Activity: 1971
Merit: 1036
eXtremal
Any news about otimization for NVIDIA ?
same question
newbie
Activity: 21
Merit: 0
full member
Activity: 151
Merit: 100
eXtremal
I have trouble to mine on xpmclminer.
Tried 9.4.1 and 10 beta + my addess (local and market).
W7 x64, driver 14.6
Xpmclminer default + + my addess working with 1 or 5 GPU but not "Share accepted" to rig with 3,6 GPU (7850,270).
I need change any parameters in config.txt or .... ?
Thanks!

check the computer name.
newbie
Activity: 21
Merit: 0
eXtremal
I have trouble to mine on xpmclminer.
Tried 9.4.1 and 10 beta + my addess (local and market).
W7 x64, driver 14.6
Xpmclminer default + + my addess working with 1 or 5 GPU but not "Share accepted" to rig with 3,6 GPU (7850,270).
I need change any parameters in config.txt or .... ?
Thanks!
newbie
Activity: 20
Merit: 0
Hi all, I dig http://xpmforall.org/, address for payments made from his wallet, a miner digs, but there is still no statistics (2 hours)
member
Activity: 92
Merit: 10
eXtremal
Any news about otimization for NVIDIA ?
hero member
Activity: 799
Merit: 1000
First 14ch share found by pool! Block height is 1144207.

So I'm assuming that long latency times can drop your earnings as the GPU goes for periods without getting work. Is there any way to improve this? It seems like it happens a lot. Do another pools have this problem?
High latency is a problem of original primecoind application. I have already tried to fix it, current pool version after optimizations works better, but not ideal Sad

Quote from: merc84
Also 8.5CPD with both 14.4 modded and 14.7rc3
Ok, I'll try test different versions of drivers.
Also you can try use different OS, Linux (and Windows 8.1?) does not require modded drivers for run 5+ GPUs.



I'm trying to understand why it worked perfectly before but now gives error even though I made no changes to drivers. Also why recompiling kernel.bin removes error but does not give same speed as precompiled kernel.bin? I notice a slight size difference in newly compiled kernel.bin vs precompiled one. If it helps here is a link to the newly compiled kernel.bin https://www.dropbox.com/s/qlbi7z815dnbhly/kernel.bin?dl=0
For what its worth Claymore xpm miner also gives same error.
sr. member
Activity: 473
Merit: 250
Sodium hypochlorite, acetone, ethanol
5870

Code:
found platform[0] name = 'AMD Accelerated Parallel Processing'
Found 1 devices
Using device 0 as GPU 0
Compiling ...
Source: 236814 bytes
binsize = 1318152 bytes
prepare_adl success
Failed to ADL_Adapter_ID_Get. Error -5Failed to ADL_Adapter_ID_Get. Error -5ADL
found less devices than opencl!There is possibly more than one display attached
to a GPUUse the gpu map feature to reliably map OpenCL to ADLWARNING: Number of
OpenCL and ADL devices did not match!Hardware monitoring may NOT match up with d
evices!Set powertune not supported
set_powertune(0, -1) failed.
Cypress; 20 compute units
square 320 bits: 120.000ms (1118.481M ops/sec)
square 320 bits: 119.000ms (1127.880M ops/sec)
multiply 320 bits: 139.000ms (965.595M ops/sec)
square 352 bits: 124.000ms (1082.401M ops/sec)
multiply 352 bits: 163.000ms (823.422M ops/sec)
element index: 0
gmp: A13A81BA 6AFB6EC4 D5F93DF9 EE1F8DC0 6C08796C 8B29C976 5DF3364E 671DEF36 09F
CC100 00000000
gpu: D3A14E6C 0065A8B1 130EC568 55976CF1 855A01F2 8657BA6F 880780B3 053C3EB6 019
89E83 00000000
results differ!
Fermat tests 320 bits: 110.000ms (1.192M ops/sec)
element index: 0
gmp: 9C07BBCF CF6E4015 84CCD446 D94FC2E9 F3F607B2 0A8DE0F2 EC950808 B2E1EADA 596
FB284 26A0F830 00000000
gpu: 117547F1 E9A416D7 C99C4C05 12EEE851 13D3BFB8 855FB2DC 91E1C13D 9879362F 516
901F9 DE6254E6 00000000
results differ!
Fermat tests 352 bits: 129.000ms (1.016M ops/sec)

 *** hashmod benchmark ***
 MHash per second: 179.977
 Hash per iteration: 38.047 (0.000454 %)
 Average hash multiplier size: 30.629

 *** sieve (check) benchmark ***
 * [OK] found candidates by CPU: 3505 by GPU: 3508
 * [OK] invalid candidates: 0
 * [OK] CPU/GPU candidates difference: 0

 *** sieve (performance) benchmark ***
 * scan speed: 13.458 G
 * iteration time: 40.906ms
 * candidates per second: 89372.422
 * candidates per iteration: 3655.89 (1579.16 320bit, 2076.73 352bit)
 * 320bit/352bit ratio: 0.760/1
Hix
legendary
Activity: 1971
Merit: 1036
First 14ch share found by pool! Block height is 1144207.

So I'm assuming that long latency times can drop your earnings as the GPU goes for periods without getting work. Is there any way to improve this? It seems like it happens a lot. Do another pools have this problem?
High latency is a problem of original primecoind application. I have already tried to fix it, current pool version after optimizations works better, but not ideal Sad

Quote from: merc84
Also 8.5CPD with both 14.4 modded and 14.7rc3
Ok, I'll try test different versions of drivers.
Also you can try use different OS, Linux (and Windows 8.1?) does not require modded drivers for run 5+ GPUs.


Any news about optimization version for NVidia cards?
sr. member
Activity: 2106
Merit: 282
👉bit.ly/3QXp3oh | 🔥 Ultimate Launc
First 14ch share found by pool! Block height is 1144207.

So I'm assuming that long latency times can drop your earnings as the GPU goes for periods without getting work. Is there any way to improve this? It seems like it happens a lot. Do another pools have this problem?
High latency is a problem of original primecoind application. I have already tried to fix it, current pool version after optimizations works better, but not ideal Sad

Quote from: merc84
Also 8.5CPD with both 14.4 modded and 14.7rc3
Ok, I'll try test different versions of drivers.
Also you can try use different OS, Linux (and Windows 8.1?) does not require modded drivers for run 5+ GPUs.

newbie
Activity: 39
Merit: 0
So I'm assuming that long latency times can drop your earnings as the GPU goes for periods without getting work. Is there any way to improve this? It seems like it happens a lot. Do another pools have this problem?
hero member
Activity: 799
Merit: 1000
Quote from: TheIllusiveMan
Any suggestions about 970/980 performance after optimizations?
After 2-3 days I'll release new beta version with optimizations for all NV cards.

Quote from: CoffeeCat
So to answer your question, I'm running an R9 290. I was getting about 9 CPD with the new version and the older version I'm getting about 14 CPD.
Use stable 9.4.1 version for AMD cards, 10.0beta don't have any AMD optimizations.
My tests says that 10.0beta have same speed as 9.4.1 on Linux/catalyst 14.4  and Windows/catalyst 14.6.

Quote from: merc84
It seems the average cpd is lower after recompiling kernel.bin 8.5 cpd instead of 9.4 cpd
8.5 cpd with drivers 14.4 modded, or 14.7rc3? Or both ?

Quote
With some other miners it is possible to put the opencl files into the folder with the miner and prevent the need to change drivers, is this something you would consider doing?
What opencl files do you mean ? Sources in 'gpu' folder, compiled kernel 'kernel.bin' in root miner's directory.

The actual driver files like amd_opencl32.dll etc, for example https://www.dropbox.com/s/wb7jihr5ibe5zu7/14.9cl.rar?dl=0 With some miners it's possible to place these in the dir with the miner so you can use different drivers without the need to install new ones.

Also 8.5CPD with both 14.4 modded and 14.7rc3

sr. member
Activity: 2106
Merit: 282
👉bit.ly/3QXp3oh | 🔥 Ultimate Launc
Quote from: TheIllusiveMan
Any suggestions about 970/980 performance after optimizations?
After 2-3 days I'll release new beta version with optimizations for all NV cards.

Quote from: CoffeeCat
So to answer your question, I'm running an R9 290. I was getting about 9 CPD with the new version and the older version I'm getting about 14 CPD.
Use stable 9.4.1 version for AMD cards, 10.0beta don't have any AMD optimizations.
My tests says that 10.0beta have same speed as 9.4.1 on Linux/catalyst 14.4  and Windows/catalyst 14.6.

Quote from: merc84
It seems the average cpd is lower after recompiling kernel.bin 8.5 cpd instead of 9.4 cpd
8.5 cpd with drivers 14.4 modded, or 14.7rc3? Or both ?

Quote
With some other miners it is possible to put the opencl files into the folder with the miner and prevent the need to change drivers, is this something you would consider doing?
What opencl files do you mean ? Sources in 'gpu' folder, compiled kernel 'kernel.bin' in root miner's directory.
hero member
Activity: 799
Merit: 1000
With some other miners it is possible to put the opencl files into the folder with the miner and prevent the need to change drivers, is this something you would consider doing? It would be extremely useful in particular for rigs that need to use modded drivers.
hero member
Activity: 799
Merit: 1000
Got some weird issue with drivers,
WARNING: Linking two modules of different data layouts!
WARNING: Linking two modules of different target triples: 'amdil64-pc-unknown-am
dopencl' and 'amdil-pc-unknown-amdopencl'

Now this wouldn't be unexpected if i had say changed the driver version since the last time i ran the miner (which i haven't)
So drivers are the same 14.4 modded (so i can run 5 cards) but now they throw this error at the start and no longer perform at 9.4 cpd but instead 6 cpd what in the world could cause this???

Edit: This is happening on 2 rigs both previously working and without any change to drivers; one has 3 x 280x and running 14.7 rc3 the other 5 x 280x 14.4 modded drivers. On the 3x 280x rig i ran amd cleanup utility to remove drivers and re-installed 14.7rc3 but i still get the same warning and speed is only 6cpd. On the 5x 280x rig i ran amd cleanup and re-installed 14.4 modded drivers still get same warning and same 6 cpd. #mindfuck!

Edit2: Simple solution, delete kernel.bin forcing it to recompile everything works again.
It seems the average cpd is lower after recompiling kernel.bin 8.5 cpd instead of 9.4 cpd
member
Activity: 92
Merit: 10
*** sieve (check) benchmark ***
OpenCL error: -54 at /HDD/build/projects/xpmclient/xpmclient/benchmarks.cpp:836
Hmm.. what driver version and config.txt you use ?

Nvidia 750ti Driver version 347.88
standard config:

Code:
# Platform:
#  "amd": AMD GCN cards
#  "amd legacy": AMD HD5xxx & HD6xxx cards
#  "nvidia": NVidia Maxwell cards
platform = "nvidia";

# GPU crash handling (properly untested feature):
# 0 - do nothing
# 1 - close miner application
# 2 - reboot system (need superuser rights)
onCrash = "0";

# - Valid cpuload range: 0 to 4
# - More than 2 does not make sense
# - You can use 0 for really slow CPUs
cpuload = "1";

# Target Cunningham chain length, default = 10
target = "10";

# Sieve size in stripes (1 stripe = 16384 bits), recommended 420 for AMD, 210 for NVidia
# Big sieve size gives more candidates with less prime probability
sieveSize = "210";

# Weave depth (primes number count using by sieve), recommended 40960 for AMD, 65536 for NVidia
# Big weave depth increases prime probability, but sieve generate less candidates
weaveDepth = "65536";

# Sieve layers number, optimal value is target*2, default = 20
width = "20";

# Sieve window size. Recommended values:
# AMD GPUs: 4096
# NVidia GeForce 750Ti: 8188
# Other NVidia Maxwell GPUs: 12284
windowSize = "8188";

# - The following settings have one entry per GPU
# - If you have more than 4 just add more entries or leave it as is to use defaults
# - All entries must be separated by commas and set between double quotes

# 0 = don't use device
# 1 = use device
devices =        ["1", "0", "0", "0", "0"];

# Primorial 13 is optimal
sieveprimorial = ["13", "13", "13", "13", "13"];

# sieves per round value, default = 5
sievePerRound = ["5", "5", "5", "5", "5"];

# -1 means don't change
corefreq =       ["-1", "-1", "-1", "-1", "-1"];
memfreq =        ["-1", "-1", "-1", "-1", "-1"];
powertune =      ["-1", "-1", "-1", "-1", "-1"];
fanspeed =       ["-1", "-1", "-1", "-1", "-1"];
newbie
Activity: 39
Merit: 0
Quote from: CoffeeCat
Can you explain with is different with this build? I tried it and I'm getting slower performance than with the previous version. I'm running 14.4 drivers. Thanks.
What GPU you use and how much CPD you see?
Can you run benchmarks (xpmclient -b) with versions 9.4.1 and 10.0?

GeForce GTX 750Ti results:
Quote
[GPU 0] T=-1C A=-1% E=0 primes=0.108085 fermat=92557/sec cpd=1.74/day
(ST/INV/DUP): 1369x 7ch(29/0/7) 154x 8ch(3/0/0) 13x 9ch(0/0/0) 3x 10ch(1/0/0)
Work received: height=1136229 diff=10.940961 latency=44ms
GPU 0 found share: 7-ch type 2
Share accepted.
GPU 0 found share: 7-ch type 3
Share accepted.
[GPU 0] T=-1C A=-1% E=0 primes=0.108085 fermat=93735/sec cpd=1.76/day
(ST/INV/DUP): 1371x 7ch(29/0/7) 154x 8ch(3/0/0) 13x 9ch(0/0/0) 3x 10ch(1/0/0)

XPM mining with 750Ti can be profitable after optimizations, if performance reaches 4+ CPD.. I think, it possible Smiley

So to answer your question, I'm running an R9 290. I was getting about 9 CPD with the new version and the older version I'm getting about 14 CPD. (I know this isn't the benchmark you were looking for but I'm away from my computer.) Again, I'm running the 14.4 drivers.

What performance are you seeing with an R9 290? It the performance lower because I'm using older drivers? If so, which version would you recommend?

Thanks!
sr. member
Activity: 2106
Merit: 282
👉bit.ly/3QXp3oh | 🔥 Ultimate Launc
780ti

 *** sieve (performance) benchmark ***
 * scan speed: 11.071 G
Same problem as HD5xxx & HD6xxx, need another implementation of sieve.

*** sieve (check) benchmark ***
OpenCL error: -54 at /HDD/build/projects/xpmclient/xpmclient/benchmarks.cpp:836
Hmm.. what driver version and config.txt you use ?

Quote
Is there any chance for optimization in near future for Nvidia 750ti ?
I don't know speed limit of 750Ti and other cards, I'll try find it. It's easy to make 2+ CPD with 750Ti (now only 1.75), I'll release new version with optimizations for NVidia soon.
member
Activity: 92
Merit: 10
hoze
Thanks.
I see that HD6xxx cards require whole kernel optimization, not only sieve Sad For compare, R9 290:

Quote

square 320 bits: 40.524ms (3312.055M ops/sec)
multiply 320 bits: 49.931ms (2688.064M ops/sec)
square 352 bits: 48.891ms (2745.244M ops/sec)
multiply 352 bits: 60.490ms (2218.842M ops/sec)
Fermat tests 320 bits: 38.432ms (3.410M ops/sec)
Fermat tests 352 bits: 47.956ms (2.733M ops/sec)

 *** hashmod benchmark ***
 MHash per second: 550.495
 Hash per iteration: 37.938 (0.000452 %)
 Average hash multiplier size: 30.712

 *** sieve (check) benchmark ***
 * [OK] found candidates by CPU: 8077 by GPU: 8082
 * [OK] invalid candidates: 0
 * [OK] CPU/GPU candidates difference: 0

 *** sieve (performance) benchmark ***
 * scan speed: 94.933 G
 * iteration time: 5.799ms
 * candidates per second: 1282165.428
 * candidates per iteration: 7435.11 (2711.41 320bit, 4723.70 352bit)
 * 320bit/352bit ratio: 0.574/1
HD5970 shows good results on Fermat test, only 3.5 times slower than R9 290.

I also intrest in benchmarks of old GeForce GTX 6xx/7xx cards, 980Ti and Fury X Smiley

Is there any chance for optimization in near future for Nvidia 750ti ?
Pages:
Jump to: