elective-stereophonic
elective-stereophonic
NxtMint Java minter singapore
Please login or register.

Login with username, password and session length
Advanced search  

News:

Latest Nxt Client: Nxt 1.11.15

Pages: 1 ... 13 14 [15] 16 17 ... 20  All

Author Topic: NxtMint Java minter  (Read 50951 times)

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #280 on: February 03, 2015, 03:31:43 pm »

For some reason 1.4 only yields about 2 MH/s.
I tried 0,256,0 with 1024 intensity on 1.4 vs 0,256,44000 intensity 10 on 1.3
Note the different GPU devicename.

11:29:23 FINE GpuKnv25.<init>: GPU 0: Local size 256, Global size 1048576, Passes 1
11:29:23 FINE GpuKnv25.<init>: GPU 1: Local size 256, Global size 1048576, Passes 1

1.3 Gives 43MH/S

11:33:09 FINE GpuKnv25.<init>: GPU local size 256, global size 11264000
11:33:09 FINE GpuKnv25.<init>: GPU local size 256, global size 11264000
The global size is the number of work items executed in a single pass.  You have 1/10 the number of executions for the 1.4 test.    I didn't know that people were specifying such large intensity values.  I'll change NxtMint to allow intensity to go up to 1024*1024 (1,048,576).  The global size field is a 32-bit integer, so there is an upper limit on the size.
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #281 on: February 03, 2015, 04:17:09 pm »

Well I am increasing the intesity up the the level where there is no increase in hashrate.
So I am not setting these high values on purpose.

To be honest I still have no idea what the values represent, and how to apply the specs of my cars on the values.
e.g gpuspeed, memory, shaders etc.

All I know is that the 6990 matches the 290x when scryptmining  in cgminer.

If you need any specifications please let me know and i am happy to beta your code.

Grts.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #282 on: February 03, 2015, 04:37:29 pm »

A test version of NxtMint is available on Google Driver https://drive.google.com/open?id=0B1312_6UqRHPS3JZQUFDQUNTSEU&authuser=0.  You can now set gpuIntensity up to 1 million.

I updated the README to try and explain gpuIntensity and gpuDevice.  A suggestion is to start with gpuDevice=0,256,0 and then keep increasing gpuIntensity until the hash rate stops improving or your display becomes too sluggish.  It is possible that hash rate might improve if the work group size is equal to the number of cores in a compute unit.  NxtMint will display the number of compute units for your adapter but has no way to get the number of cores.  Often the device driver information displayed by the system device manager will contain the number of cores.  Or you can check the technical specifications for your card (note that AMD and NVIDIA might license the card to another manufacturer who can tweak the specifications by adding additional compute units to the card).

Scrypt is a special case since it requires a large memory allocation for each work item.  This places an upper limit on the global size that can be supported.
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #283 on: February 03, 2015, 04:55:42 pm »

Thanks, I will test and report back.. :)
Since I run linux with no gui I can not report on graphics sluggishness.
I will hold an eye on the temperature..

And about my specs, clinfo gives :

  Device Type:                CL_DEVICE_TYPE_GPU
  Vendor ID:                1002h
  Board name:                AMD Radeon HD 6900 Series
  Device Topology:             PCI[ B#8, D#0, F#0 ]
  Max compute units:             24
  Max work items dimensions:          3
    Max work items[0]:             256
    Max work items[1]:             256
    Max work items[2]:             256
  Max work group size:             256
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #284 on: February 03, 2015, 05:11:48 pm »

Whatever value I choose for intensity the hashrate stays at around 2.5MH/S
Seems there is a bottleneck somewhere...?
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #285 on: February 03, 2015, 06:34:47 pm »

Whatever value I choose for intensity the hashrate stays at around 2.5MH/S
Seems there is a bottleneck somewhere...?
What did you set gpuDevice to?  What did NxtMint display for local size, global size and passes?  The goal is to have 1 pass with as large a global size as possible (global size is the number of work items being executed).  If you set the group count to 0, NxtMint will calculate the global size based on gpuIntensity (you want the work group size to divide evenly into 1024 in this case).

The AMD HD 6900 spec says there are 1408 stream processors.  I don't know the exact number on your particular card (depends on the card manufacturer).  The Linux device configuration might have this information.  So a global size that is a multiple of 1408 might yield some improvement.  Also, all work items in the same work group execute on the same compute unit.  So picking a work group size that is a multiple of the number of processors per compute unit might also yield some improvement.  Since 1408 isn't a multiple of 24, I suspect you have more than 1408 stream processors on your card.

My NVIDIA GT 720 gets 32 MH/s for EGOLD with gpuIntensity=1024.  The card has 192 cores per compute unit but just 2 compute units.  I get slightly better performance with a work group size of 192 instead of 256.

Code: [Select]
C:\Bitcoin\NxtMint>java.exe -Xmx256m -Djava.library.path=\Bitcoin\NxtMint\jni -jar \Bitcoin\NxtMint\NxtMint-1.4.0.jar
01:09:53 INFO Main.main: Java Nxt Mint Version 1.4.0
01:09:53 INFO Main.main: Application data path: C:\Users\Ronald\Appdata\Roaming\NxtMint
01:09:53 INFO Main.main: Using Nxt node at scripterron.dyndns.biz:7876
01:09:53 INFO Main.main: Minting 60.000000 units of EGOLD for account NXT-XM86-4ZNA-65L5-CDWUE: 0 CPU threads, 1024 GPU  intensity
01:09:53 INFO Nxt.init: API node=scripterron.dyndns.biz, API port=7876
01:09:54 INFO Main.buildGpuList: GPU device 0: GeForce GT 720, Driver 347.09
  1024MB global memory, 47KB local memory, 2 compute units, Max work group size 1024
01:09:55 INFO HashFunction.<init>: JNI library NxtMint_x86_64 loaded - using native CPU hash routines
01:09:55 FINE GpuKnv25.<init>: GPU 0: Local size 192, Global size 1048704, Passes 1
01:09:55 INFO MintWorker.run: GPU worker 0 starting on GPU 0
01:09:55 FINE MintWorker.run: Worker 0 starting on counter 255
01:10:55 FINE MintWorker.run: Worker 0: 1,925.42 MHash, 32.0903 MHash/s
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #286 on: February 03, 2015, 06:48:24 pm »

Hmm - I get even better performance with gpuDevice=0,128,0.  So you really need to just experiment to see what works best.  I think NVIDIA cards always prefer a work group size that is a multiple of 32.  I don't know about AMD cards.  It all depends on the internal architecture of the chips.

Code: [Select]
C:\Bitcoin\NxtMint>java.exe -Xmx256m -Djava.library.path=\Bitcoin\NxtMint\jni -jar \Bitcoin\NxtMint\NxtMint-1.4.0.jar
01:43:08 INFO Main.main: Java Nxt Mint Version 1.4.0
01:43:08 INFO Main.main: Application data path: C:\Users\Ronald\Appdata\Roaming\NxtMint
01:43:08 INFO Main.main: Using Nxt node at scripterron.dyndns.biz:7876
01:43:08 INFO Main.main: Minting 60.000000 units of EGOLD for account NXT-XM86-4ZNA-65L5-CDWUE: 0 CPU threads, 1024 GPU intensity
01:43:08 INFO Nxt.init: API node=scripterron.dyndns.biz, API port=7876
01:43:09 INFO Main.buildGpuList: GPU device 0: GeForce GT 720, Driver 347.09
  1024MB global memory, 47KB local memory, 2 compute units, Max work group size 1024
01:43:10 INFO HashFunction.<init>: JNI library NxtMint_x86_64 loaded - using native CPU hash routines
01:43:10 FINE GpuKnv25.<init>: GPU 0: Local size 128, Global size 1048576, Passes 1
01:43:10 INFO MintWorker.run: GPU worker 0 starting on GPU 0
01:43:10 FINE MintWorker.run: Worker 0 starting on counter 255
01:44:10 FINE MintWorker.run: Worker 0: 2,063.60 MHash, 34.3933 MHash/s
01:45:10 FINE MintWorker.run: Worker 0: 4,126.15 MHash, 34.3846 MHash/s
Logged

Wolf0

  • Jr. Member
  • **
  • Karma: +7/-0
  • Offline Offline
  • Posts: 97
    • View Profile
Re: NxtMint Java minter
« Reply #287 on: February 03, 2015, 08:05:48 pm »

Whatever value I choose for intensity the hashrate stays at around 2.5MH/S
Seems there is a bottleneck somewhere...?
What did you set gpuDevice to?  What did NxtMint display for local size, global size and passes?  The goal is to have 1 pass with as large a global size as possible (global size is the number of work items being executed).  If you set the group count to 0, NxtMint will calculate the global size based on gpuIntensity (you want the work group size to divide evenly into 1024 in this case).

The AMD HD 6900 spec says there are 1408 stream processors.  I don't know the exact number on your particular card (depends on the card manufacturer).  The Linux device configuration might have this information.  So a global size that is a multiple of 1408 might yield some improvement.  Also, all work items in the same work group execute on the same compute unit.  So picking a work group size that is a multiple of the number of processors per compute unit might also yield some improvement.  Since 1408 isn't a multiple of 24, I suspect you have more than 1408 stream processors on your card.

My NVIDIA GT 720 gets 32 MH/s for EGOLD with gpuIntensity=1024.  The card has 192 cores per compute unit but just 2 compute units.  I get slightly better performance with a work group size of 192 instead of 256.

Code: [Select]
C:\Bitcoin\NxtMint>java.exe -Xmx256m -Djava.library.path=\Bitcoin\NxtMint\jni -jar \Bitcoin\NxtMint\NxtMint-1.4.0.jar
01:09:53 INFO Main.main: Java Nxt Mint Version 1.4.0
01:09:53 INFO Main.main: Application data path: C:\Users\Ronald\Appdata\Roaming\NxtMint
01:09:53 INFO Main.main: Using Nxt node at scripterron.dyndns.biz:7876
01:09:53 INFO Main.main: Minting 60.000000 units of EGOLD for account NXT-XM86-4ZNA-65L5-CDWUE: 0 CPU threads, 1024 GPU  intensity
01:09:53 INFO Nxt.init: API node=scripterron.dyndns.biz, API port=7876
01:09:54 INFO Main.buildGpuList: GPU device 0: GeForce GT 720, Driver 347.09
  1024MB global memory, 47KB local memory, 2 compute units, Max work group size 1024
01:09:55 INFO HashFunction.<init>: JNI library NxtMint_x86_64 loaded - using native CPU hash routines
01:09:55 FINE GpuKnv25.<init>: GPU 0: Local size 192, Global size 1048704, Passes 1
01:09:55 INFO MintWorker.run: GPU worker 0 starting on GPU 0
01:09:55 FINE MintWorker.run: Worker 0 starting on counter 255
01:10:55 FINE MintWorker.run: Worker 0: 1,925.42 MHash, 32.0903 MHash/s

The minter's gonna kinda suck for 6xxx cards - VLIW4 have hardware vectors, and the OpenCL compiler SUCKS with ulong, especially for them.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #288 on: February 03, 2015, 08:35:52 pm »

The minter's gonna kinda suck for 6xxx cards - VLIW4 have hardware vectors, and the OpenCL compiler SUCKS with ulong, especially for them.

What do you suggest?  Switching to uint didn't make a noticeable difference for my NVIDIA card, but I can do that if it will help AMD cards.
Logged

Wolf0

  • Jr. Member
  • **
  • Karma: +7/-0
  • Offline Offline
  • Posts: 97
    • View Profile
Re: NxtMint Java minter
« Reply #289 on: February 03, 2015, 08:44:33 pm »

The minter's gonna kinda suck for 6xxx cards - VLIW4 have hardware vectors, and the OpenCL compiler SUCKS with ulong, especially for them.

What do you suggest?  Switching to uint didn't make a noticeable difference for my NVIDIA card, but I can do that if it will help AMD cards.

Vectorization will help 6xxx. But hardly anyone uses pre-GCN cards anymore, anyway.
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #290 on: February 03, 2015, 10:09:24 pm »

I have tried:

gpuIntensity=1024
gpuDevice=0,256,0
gpuDevice=1,256,0

Gives 2.5MH/S





Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #291 on: February 03, 2015, 10:38:45 pm »

I have tried:

gpuIntensity=1024
gpuDevice=0,256,0
gpuDevice=1,256,0

Gives 2.5MH/S
Are you using the test version that I posted today?  Increase gpuIntensity to 44000 and see if the hash rate improves.  If it doesn't, I don't know what is causing the problem since the Keccak25 algorithm didn't change between 1.3 and 1.4.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #292 on: February 03, 2015, 10:53:18 pm »

I have tried:

gpuIntensity=1024
gpuDevice=0,256,0
gpuDevice=1,256,0

Gives 2.5MH/S
I installed NxtMint 1.3.0 and I'm getting 1/2 of the hash rate that I get with 1.4.0.  I'm getting 34.3 MH/s on 1.4.0 and 17.9 MH/s on 1.3.0.  The gpuIntensity and global size are the same, so each test should be performing the same number of hashes per kernel execution.  But evidently Aparapi does something special for AMD cards that I'm not doing, but I don't know what it is.

Code: [Select]
C:\Bitcoin\NxtMint\1.3.0>java -Xmx256m -Djava.library.path="aparapi;jni" -jar NxtMint-1.3.0.jar
05:41:01 INFO Main.main: Java Nxt Mint Version 1.3.0
05:41:01 INFO Main.main: Application data path: C:\Users\Ronald\Appdata\Roaming\NxtMint
05:41:01 INFO Main.main: Using Nxt node at scripterron.dyndns.biz:7876
05:41:01 INFO Main.main: Minting 60.000000 units of EGOLD for account NXT-XM86-4ZNA-65L5-CDWUE: 0 CPU threads, 1024 GPU
intensity
05:41:01 INFO Nxt.init: API node=scripterron.dyndns.biz, API port=7876
05:41:02 INFO Main.lambda$null$2: GPU device 0: NVIDIA CUDA
  1024MB global memory, 47KB local memory, 2 compute units, Max work group size 1024
05:41:03 INFO HashFunction.<init>: JNI library NxtMint_x86_64 loaded - using native hash routines
05:41:03 FINE GpuKnv25.<init>: GPU local size 128, global size 1048576
05:41:03 INFO MintWorker.run: GPU worker 0 starting on GPU 0
05:41:03 FINE MintWorker.run: Worker 0 starting on counter 255
05:42:03 FINE MintWorker.run: Worker 0: 1,075.84 MHash, 17.9306 MHash/s
Logged

Wolf0

  • Jr. Member
  • **
  • Karma: +7/-0
  • Offline Offline
  • Posts: 97
    • View Profile
Re: NxtMint Java minter
« Reply #293 on: February 03, 2015, 11:02:34 pm »

I have tried:

gpuIntensity=1024
gpuDevice=0,256,0
gpuDevice=1,256,0

Gives 2.5MH/S
I installed NxtMint 1.3.0 and I'm getting 1/2 of the hash rate that I get with 1.4.0.  I'm getting 34.3 MH/s on 1.4.0 and 17.9 MH/s on 1.3.0.  The gpuIntensity and global size are the same, so each test should be performing the same number of hashes per kernel execution.  But evidently Aparapi does something special for AMD cards that I'm not doing, but I don't know what it is.

Code: [Select]
C:\Bitcoin\NxtMint\1.3.0>java -Xmx256m -Djava.library.path="aparapi;jni" -jar NxtMint-1.3.0.jar
05:41:01 INFO Main.main: Java Nxt Mint Version 1.3.0
05:41:01 INFO Main.main: Application data path: C:\Users\Ronald\Appdata\Roaming\NxtMint
05:41:01 INFO Main.main: Using Nxt node at scripterron.dyndns.biz:7876
05:41:01 INFO Main.main: Minting 60.000000 units of EGOLD for account NXT-XM86-4ZNA-65L5-CDWUE: 0 CPU threads, 1024 GPU
intensity
05:41:01 INFO Nxt.init: API node=scripterron.dyndns.biz, API port=7876
05:41:02 INFO Main.lambda$null$2: GPU device 0: NVIDIA CUDA
  1024MB global memory, 47KB local memory, 2 compute units, Max work group size 1024
05:41:03 INFO HashFunction.<init>: JNI library NxtMint_x86_64 loaded - using native hash routines
05:41:03 FINE GpuKnv25.<init>: GPU local size 128, global size 1048576
05:41:03 INFO MintWorker.run: GPU worker 0 starting on GPU 0
05:41:03 FINE MintWorker.run: Worker 0 starting on counter 255
05:42:03 FINE MintWorker.run: Worker 0: 1,075.84 MHash, 17.9306 MHash/s

It probably changes the code somewhat, but it's still pretty bad.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #294 on: February 04, 2015, 01:04:43 am »

I have tried:

gpuIntensity=1024
gpuDevice=0,256,0
gpuDevice=1,256,0

Gives 2.5MH/S
I changed the test version of NxtMint on Google Drive to use scalars instead of arrays in the Keccak25 algorithm.  If you get a chance, try it out and let me know if that improves the AMD performance.
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #295 on: February 04, 2015, 09:10:05 am »

Ok ran the latest google drive version:

1.4 Latest
09: 38: 59 INFO Main. main: Java Nxt Mint Version 1.4.0
09: 38: 59 INFO Main. main: Using Nxt node at localh0st: 7876
09: 38: 59 INFO Main. main: Minting 40.000000 units of EGOLD for account NXT-Z4RD-3PZW-RA7Y-4JNUR: 1 CPU threads, 44000 GPU intensity
09: 38: 59 INFO Nxt. init: API node=localh0st, API port=7876
09: 38: 59 INFO Main. buildGpuList: GPU device 0: Cayman, Driver 1573.4 (VM)
  1904MB global memory, 32KB local memory, 24 compute units, Max work group size 256
09: 38: 59 INFO Main. buildGpuList: GPU device 1: Cayman, Driver 1573.4 (VM)
  1953MB global memory, 32KB local memory, 24 compute units, Max work group size 256
09: 39: 00 INFO HashFunction. <init>: JNI library NxtMint_x86 loaded - using native CPU hash routines
09: 39: 00 INFO MintWorker. run: CPU worker 0 starting
09: 39: 01 FINE GpuKnv25. <init>: GPU 0: Local size 256, Global size 45056000, Passes 1
09: 39: 01 INFO MintWorker. run: GPU worker 1 starting on GPU 0
09: 39: 01 FINE GpuKnv25. <init>: GPU 1: Local size 256, Global size 45056000, Passes 1
09: 39: 01 INFO MintWorker. run: GPU worker 2 starting on GPU 1
09: 39: 01 FINE MintWorker. run: Worker 0 starting on counter 36
09: 39: 01 FINE MintWorker. run: Worker 1 starting on counter 36
09: 39: 01 FINE MintWorker. run: Worker 2 starting on counter 36
09: 40: 04 FINE MintWorker. run: Worker 0: 23.07 MHash, 0.3721 MHash/s
09: 40: 11 FINE MintWorker. run: Worker 2: 180.22 MHash, 2.5746 MHash/s
09: 40: 11 FINE MintWorker. run: Worker 1: 180.22 MHash, 2.5746 MHash/s
09: 41: 06 FINE MintWorker. run: Worker 0: 46.14 MHash, 0.3691 MHash/s
09: 41: 21 FINE MintWorker. run: Worker 2: 360.45 MHash, 2.5746 MHash/s
09: 41: 21 FINE MintWorker. run: Worker 1: 360.45 MHash, 2.5746 MHash/s


1.3
[
Starting NxtMint
09: 53: 29 INFO Main. main: Java Nxt Mint Version 1.3.0
09: 53: 29 INFO Main. main: Using Nxt node at localh0st: 7876
09: 53: 29 INFO Main. main: Minting 40.000000 units of EGOLD for account NXT-Z4RD-3PZW-RA7Y-4JNUR: 0 CPU threads, 10 GPU intensity
09: 53: 29 INFO Nxt. init: API node=localh0st, API port=7876
09: 53: 30 INFO Main. lambda$null$2: GPU device 0: AMD Accelerated Parallel Processing
  1904MB global memory, 32KB local memory, 24 compute units, Max work group size 256
09: 53: 30 INFO Main. lambda$null$2: GPU device 1: AMD Accelerated Parallel Processing
  1953MB global memory, 32KB local memory, 24 compute units, Max work group size 256
09: 53: 31 INFO HashFunction. <init>: JNI library NxtMint_x86 loaded - using native hash routines
09: 53: 31 FINE GpuKnv25. <init>: GPU local size 256, global size 11264000
09: 53: 31 FINE GpuKnv25. <init>: GPU local size 256, global size 11264000
09: 53: 31 INFO MintWorker. run: GPU worker 0 starting on GPU 0
09: 53: 31 INFO MintWorker. run: GPU worker 1 starting on GPU 1
09: 53: 31 FINE MintWorker. run: Worker 1 starting on counter 36
09: 53: 31 FINE MintWorker. run: Worker 0 starting on counter 36
09: 54: 31 FINE MintWorker. run: Worker 1: 2,601.98 MHash, 43.3664 MHash/s
09: 54: 31 FINE MintWorker. run: Worker 0: 2,601.98 MHash, 43.3664 MHash/s


The thing that caught my eye is the CPU load..
xorg starts with 1.4 and consumes cpu resources, 1.3 does not shows this behaviour.

top - 10:00:23 up 2 days, 23:18,  6 users,  load average: 0.85, 0.78, 1.06
Tasks: 215 total,   2 running, 211 sleeping,   1 stopped,   1 zombie
Cpu(s): 16.6%us,  0.8%sy,  0.0%ni, 81.7%id,  0.2%wa,  0.0%hi,  0.8%si,  0.0%st
Mem:   4129320k total,  3084376k used,  1044944k free,   324060k buffers
Swap:  4190204k total,      288k used,  4189916k free,  1685748k cached

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND                                                                                                             
 1300 root      20   0  530m 287m 238m R   97  7.1  27:05.59 Xorg                                                                                                                 
   38 root      20   0     0    0    0 S    4  0.0   2:55.42 kworker/0:1                                                                                                         
25218 jasper    20   0 1404m 183m  10m S    2  4.6   0:17.19 java                                                                                                                 
    1 root      20   0  3664 2072 1300 S    0  0.1   0:00.81 init
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #296 on: February 04, 2015, 09:18:11 am »

Also 1.4 gives :

09:59:38 INFO HashFunction.<init>: JNI library NxtMint_x86 loaded - using native CPU hash routines

1.3 :

10:15:19 INFO HashFunction.<init>: JNI library NxtMint_x86 loaded - using native hash routines

Does this implies that the CPU in involved in 1.4?
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #297 on: February 04, 2015, 12:01:54 pm »

I also found this comment on page  7 :
Want to try the 1.1.0 public released, for the sake of test.

Quote from: Jimmy2011 on January 21, 2015, 09:22:51 am

Quote
Quote from: Jimmy2011 on January 21, 2015, 09:22:51 am


    I have 16 compute units as following.  BTW, it seems that the latest version has poor performance as minting EGOLD compared with the public released 1.1.0 version.

    Code: [Select]

    05:17:34 INFO Main.buildGpuList: GPU device 0: NVIDIA CUDA, 3072MB global memory
    , 48KB local memory, 16 compute units

I also have poor performance, in a 6990, each GPU now gives me 45mhash and before it was giving me 100mhash, so I don't gain anything using the whole card
Each GPU has 1536 cores and 24 compute units
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #298 on: February 04, 2015, 03:04:12 pm »

Ok ran the latest google drive version:
Since there is a single pass, the CPU time is in OpenCL as it schedules additional kernel executions.  Java is involved when it is necessary to schedule multiple passes.  There comes a point where increasing the global size does not increase the hash rate.

I don't know what else to do for your card.  I thought the bottleneck was memory references, but switching to scalar values didn't change that.  You should stay with 1.3 since that works best for you.  I'll keep 1.3 available for download for other users who are experiencing the same problem.
Logged

dubmannnn

  • Jr. Member
  • **
  • Karma: +0/-0
  • Offline Offline
  • Posts: 16
    • View Profile
Re: NxtMint Java minter
« Reply #299 on: February 04, 2015, 04:25:02 pm »

But can you pinpoint a codechange in the post where i quote a the user also using the 6990? He experienced 1/2 drop in hashrate somewhere in the 1.1.0 releases.

i searched the git, but i could not find a full version history.
Logged
Pages: 1 ... 13 14 [15] 16 17 ... 20  All
 

elective-stereophonic
elective-stereophonic
assembly
assembly