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 ... 6 7 [8] 9 10 ... 20  All

Author Topic: NxtMint Java minter  (Read 50816 times)

794259332

  • Newbie
  • *
  • Karma: +0/-0
  • Offline Offline
  • Posts: 7
    • View Profile
Re: NxtMint Java minter
« Reply #140 on: January 22, 2015, 08:35:08 am »

error:

exception in thread "main" java.lang.OutOfMemoryError: Java heap space


how to set up?


i have used -Xmx1024m
Logged

coretechs

  • Sr. Member
  • ****
  • Karma: +161/-1
  • Offline Offline
  • Posts: 436
    • View Profile
Re: NxtMint Java minter
« Reply #141 on: January 22, 2015, 01:50:44 pm »

error:

exception in thread "main" java.lang.OutOfMemoryError: Java heap space


how to set up?


i have used -Xmx1024m

Is this with scrypt hashing?  I tried minting some ELEMS on an r290x and had the same error with 2048m.  If I increase it to 4096m it is able to start hashing, but it crashes within a few seconds every time.  I think the card may have been overclocked at the time and I haven't had a chance to tinker with it any further yet so YMMV.
Logged
https://ardorportal.org - Ardor blockchain explorer | https://nxtportal.org - Nxt blockchain explorer | http://bitcoindoc.com - The Rise and Rise of Bitcoin
ARDOR-T43P-R2K9-8W79-9W2AL | NXT-WY9K-ZMTT-QQTT-3NBL7

794259332

  • Newbie
  • *
  • Karma: +0/-0
  • Offline Offline
  • Posts: 7
    • View Profile
Re: NxtMint Java minter
« Reply #142 on: January 22, 2015, 02:19:07 pm »

error:

exception in thread "main" java.lang.OutOfMemoryError: Java heap space


how to set up?


i have used -Xmx1024m

Is this with scrypt hashing?  I tried minting some ELEMS on an r290x and had the same error with 2048m.  If I increase it to 4096m it is able to start hashing, but it crashes within a few seconds every time.  I think the card may have been overclocked at the time and I haven't had a chance to tinker with it any further yet so YMMV.

yes,ELEMS, i use two r280x cards
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #143 on: January 22, 2015, 03:16:59 pm »

error:

exception in thread "main" java.lang.OutOfMemoryError: Java heap space


how to set up?


i have used -Xmx1024m
Please post the exception stack trace so I can see where it happened.  Also post the startup messages which show information about the graphics card and the calculated local and global sizes.

A 1GB heap should be plenty.  Did you specify -Xmx1024m before -jar?
Logged

coretechs

  • Sr. Member
  • ****
  • Karma: +161/-1
  • Offline Offline
  • Posts: 436
    • View Profile
Re: NxtMint Java minter
« Reply #144 on: January 22, 2015, 05:08:29 pm »

I spent some time trying to get the r290x working with scrypt, but the display driver crashes every time with no java error thrown.  I tried stock settings as well as lower cpu/memory clocks, intensity 0-XXX, but it always crashes after about 10-20 seconds of hashing.  KNV25 is hashing fine even with overclocking (310Mh/s !) so I don't think its an issue with the card.  For ref. this is on Windows 7 x64 and the card is reporting 4096m global, 32k local, 44 compute units.
Logged
https://ardorportal.org - Ardor blockchain explorer | https://nxtportal.org - Nxt blockchain explorer | http://bitcoindoc.com - The Rise and Rise of Bitcoin
ARDOR-T43P-R2K9-8W79-9W2AL | NXT-WY9K-ZMTT-QQTT-3NBL7

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #145 on: January 22, 2015, 05:22:14 pm »

I spent some time trying to get the r290x working with scrypt, but the display driver crashes every time with no java error thrown.  I tried stock settings as well as lower cpu/memory clocks, intensity 0-XXX, but it always crashes after about 10-20 seconds of hashing.  KNV25 is hashing fine even with overclocking (310Mh/s !) so I don't think its an issue with the card.  For ref. this is on Windows 7 x64 and the card is reporting 4096m global, 32k local, 44 compute units.
I just put Version 1.2.0 out on github.  Give it a try and set the work group size to 8 (gpuDevice=0,8).
The problem might be the 32K of local memory since the default work group size of 256 will require 32K of local memory, leaving nothing for other uses.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #146 on: January 22, 2015, 05:26:16 pm »

I put Version 1.2.0 out on github https://github.com/ScripterRon/NxtMint/releases.

The gpuDevice configuration parameter has changed from the test version of NxtMint.  The second parameter now specifies the work group size instead of the number of cores on the card.  You can vary the work group size to reduce local memory contention.  I got the best Scrypt results on my card with a work group size of 8.  For Keccak25 and Sha256, I would suggest setting the work group size to the number of cores in a compute unit (cores on the card divided by the number of compute units)
Logged

cbw

  • Newbie
  • *
  • Karma: +2/-0
  • Offline Offline
  • Posts: 12
    • View Profile
Re: NxtMint Java minter
« Reply #147 on: January 22, 2015, 06:51:52 pm »

Video cards - 3xr9270x. On version 1.2.0 also tried start with 3gpuDevice (out of memory) and with only 1gpuDevice. With only 1gpuDevice tried set "gpuDevice=0", "gpuDevice=0,8", "gpuDevice=0,4", "gpuDevice=0,2" and "gpuDevice=0,1" - not working, videodriver fails.
Logged

coretechs

  • Sr. Member
  • ****
  • Karma: +161/-1
  • Offline Offline
  • Posts: 436
    • View Profile
Re: NxtMint Java minter
« Reply #148 on: January 22, 2015, 07:23:01 pm »

Same result here, tried with gpuDevice=0,8 / 0,4 / 0,1 and intensity 1,10, video driver crashes.  This is only with the AMD, I tested an Nvidia card and it hashes with no problems.

Code: [Select]
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x0000000002a94a2b, pid=2620, tid=404
#
# JRE version: Java(TM) SE Runtime Environment (8.0_31-b13) (build 1.8.0_31-b13)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (25.31-b07 mixed mode windows-amd64 compressed oops)
# Problematic frame:
# J 1238 C1 java.util.ArrayList.toArray([Ljava/lang/Object;)[Ljava/lang/Object; (60 bytes) @ 0x0000000002a94a2b [0x0000000002a945e0+0x44b]
...
Logged
https://ardorportal.org - Ardor blockchain explorer | https://nxtportal.org - Nxt blockchain explorer | http://bitcoindoc.com - The Rise and Rise of Bitcoin
ARDOR-T43P-R2K9-8W79-9W2AL | NXT-WY9K-ZMTT-QQTT-3NBL7

crimi

  • Hero Member
  • *****
  • Karma: +122/-11
  • Offline Offline
  • Posts: 863
    • View Profile
Re: NxtMint Java minter
« Reply #149 on: January 22, 2015, 07:45:40 pm »

I spent some time trying to get the r290x working with scrypt, but the display driver crashes every time with no java error thrown.  I tried stock settings as well as lower cpu/memory clocks, intensity 0-XXX, but it always crashes after about 10-20 seconds of hashing.  KNV25 is hashing fine even with overclocking (310Mh/s !) so I don't think its an issue with the card.  For ref. this is on Windows 7 x64 and the card is reporting 4096m global, 32k local, 44 compute units.
No problems here 2 x 290x, intensity 512. But seems a little bit one of the workers is a lot slower.
« Last Edit: January 22, 2015, 07:52:43 pm by crimi »
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #150 on: January 22, 2015, 08:14:45 pm »

Same result here, tried with gpuDevice=0,8 / 0,4 / 0,1 and intensity 1,10, video driver crashes.  This is only with the AMD, I tested an Nvidia card and it hashes with no problems.

Code: [Select]
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x0000000002a94a2b, pid=2620, tid=404
#
# JRE version: Java(TM) SE Runtime Environment (8.0_31-b13) (build 1.8.0_31-b13)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (25.31-b07 mixed mode windows-amd64 compressed oops)
# Problematic frame:
# J 1238 C1 java.util.ArrayList.toArray([Ljava/lang/Object;)[Ljava/lang/Object; (60 bytes) @ 0x0000000002a94a2b [0x0000000002a945e0+0x44b]
...
That looks like an error in the Aparapi runtime library.  I've seen that kind of crash when there is a problem accessing adapter local memory.  SCRYPT is the only algorithm that uses local memory, KECCAK25 and SHA256 use just global memory.

Have you checked your device driver and OpenCL levels?
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #151 on: January 22, 2015, 08:16:22 pm »

Video cards - 3xr9270x. On version 1.2.0 also tried start with 3gpuDevice (out of memory) and with only 1gpuDevice. With only 1gpuDevice tried set "gpuDevice=0", "gpuDevice=0,8", "gpuDevice=0,4", "gpuDevice=0,2" and "gpuDevice=0,1" - not working, videodriver fails.
What is the GPU information displayed by NxtMint when it starts?
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #152 on: January 22, 2015, 08:19:19 pm »

I spent some time trying to get the r290x working with scrypt, but the display driver crashes every time with no java error thrown.  I tried stock settings as well as lower cpu/memory clocks, intensity 0-XXX, but it always crashes after about 10-20 seconds of hashing.  KNV25 is hashing fine even with overclocking (310Mh/s !) so I don't think its an issue with the card.  For ref. this is on Windows 7 x64 and the card is reporting 4096m global, 32k local, 44 compute units.
No problems here 2 x 290x, intensity 512. But seems a little bit one of the workers is a lot slower.

Try reducing the work group size: gpuDevice=0,8 for example.  This improved performance for me because it reduces the local memory contention.
Logged

cbw

  • Newbie
  • *
  • Karma: +2/-0
  • Offline Offline
  • Posts: 12
    • View Profile
Re: NxtMint Java minter
« Reply #153 on: January 22, 2015, 09:17:37 pm »

Video cards - 3xr9270x. On version 1.2.0 also tried start with 3gpuDevice (out of memory) and with only 1gpuDevice. With only 1gpuDevice tried set "gpuDevice=0", "gpuDevice=0,8", "gpuDevice=0,4", "gpuDevice=0,2" and "gpuDevice=0,1" - not working, videodriver fails.
What is the GPU information displayed by NxtMint when it starts?
11:16:47 INFO Main.lambda$null$2: GPU device 0: AMD Accelerated Parallel Processing 2048MB global memory, 32KB local memory, 20 compute units, Max work group size 256
11:16:47 INFO Main.lambda$null$2: GPU device 1: AMD Accelerated Parallel Processing 2048MB global memory, 32KB local memory, 20 compute units, Max work group size 256
11:16:47 INFO Main.lambda$null$2: GPU device 2: AMD Accelerated Parallel Processing 2048MB global memory, 32KB local memory, 20 compute units, Max work group size 256

If starts with gpuDevice=0,8 gpuDevice=1,8 gpuDevice=2,8
Quote
11:19:40 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:19:40 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:19:40 INFO MintWorker.run: GPU worker 0 starting on GPU 0
11:19:41 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:19:41 INFO MintWorker.run: GPU worker 1 starting on GPU 1
Exception in thread "main" java.lang.OutOfMemoryError: Java heap space
        at org.ScripterRon.NxtMint.GpuScrypt.<init>(GpuScrypt.java:207)
        at org.ScripterRon.NxtMint.GpuFunction.factory(GpuFunction.java:59)
        at org.ScripterRon.NxtMint.MintWorker.<init>(MintWorker.java:86)
        at org.ScripterRon.NxtMint.Mint.mint(Mint.java:93)
        at org.ScripterRon.NxtMint.Main.main(Main.java:306)

If starts with gpuDevice=0,8 gpuDevice=1,8
Quote
11:21:32 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:21:32 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:21:32 INFO MintWorker.run: GPU worker 0 starting on GPU 0
11:21:33 INFO MintWorker.run: GPU worker 1 starting on GPU 1
11:21:33 FINE MintWorker.run: Worker 1 starting on counter 1
11:21:33 FINE MintWorker.run: Worker 0 starting on counter 1
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x000000006c6e7957, pid=6980, tid=5196
#
# JRE version: Java(TM) SE Runtime Environment (8.0_40-b21) (build 1.8.0_40-ea-b21)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (25.40-b24 mixed mode windows-amd64 compressed oops)
# Problematic frame:
# V  [jvm.dll+0x187957]
#
# Failed to write core dump. Minidumps are not enabled by default on client versions of Windows
#
# An error report file with more information is saved as:
# C:\nxtmint\hs_err_pid6980.log
Logfile: pastebin.com/sfWAfa7U

If starts with gpuDevice=0,8 - for 15-20 seconds I see load of my 1 GPU and after 15-20 seconds I got system message, video driver crashed. GPU load 0%. But in NxtMnt cmd window no errors:
Quote
11:28:26 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:28:26 INFO MintWorker.run: GPU worker 0 starting on GPU 0
11:28:26 FINE MintWorker.run: Worker 0 starting on counter 1

Changes in gpuIntensity given nothing, try set 1 and 10.

Start command:
java -Djava.library.path=C:\nxtmint -Xmx2048m -Dnxt.datadir=C:\nxtmint -jar C:\nxtmint\NxtMint-1.2.0.jar
In c:\nxtmint I'm copy all files and folders from aparapi archive and all files and folders from AMD APP SDK x64 - C:\Program Files (x86)\AMD APP SDK\3.0-0-Beta\bin\x86_64 (in previous I'm copy only OpenCL.dll from AMD APP SDK folder). Drivers - AMD Catalyst Omega 14.12.
« Last Edit: January 22, 2015, 09:44:39 pm by cbw »
Logged

cbw

  • Newbie
  • *
  • Karma: +2/-0
  • Offline Offline
  • Posts: 12
    • View Profile
Re: NxtMint Java minter
« Reply #154 on: January 22, 2015, 10:33:42 pm »

No problems here 2 x 290x, intensity 512. But seems a little bit one of the workers is a lot slower.

4 CPU threads, 0 GPU intensity
12:23:25 INFO Nxt.init: API node=localhost, API port=7876
12:23:27 INFO MintWorker.run: CPU worker 0 starting
12:23:27 INFO MintWorker.run: CPU worker 1 starting
12:23:27 INFO MintWorker.run: CPU worker 2 starting
12:23:27 INFO MintWorker.run: CPU worker 3 starting
12:24:38 FINE MintWorker.run: Worker 0: 0,11 MHash, 0,0018 MHash/s
12:24:38 FINE MintWorker.run: Worker 3: 0,11 MHash, 0,0018 MHash/s
12:24:38 FINE MintWorker.run: Worker 2: 0,11 MHash, 0,0018 MHash/s
12:24:38 FINE MintWorker.run: Worker 1: 0,11 MHash, 0,0018 MHash/s

Also tried to set gpuDevice=0,512 and gpuIntensity=512 - not working.
My Athlon X4 750k (weak CPU) doing 0,0072MHash/s and you 2x 290x doing around 0,0088MHash/s at SCRYPT. You say "no problems", really?
Logged

crimi

  • Hero Member
  • *****
  • Karma: +122/-11
  • Offline Offline
  • Posts: 863
    • View Profile
Re: NxtMint Java minter
« Reply #155 on: January 22, 2015, 11:00:07 pm »

No problems here 2 x 290x, intensity 512. But seems a little bit one of the workers is a lot slower.

4 CPU threads, 0 GPU intensity
12:23:25 INFO Nxt.init: API node=localhost, API port=7876
12:23:27 INFO MintWorker.run: CPU worker 0 starting
12:23:27 INFO MintWorker.run: CPU worker 1 starting
12:23:27 INFO MintWorker.run: CPU worker 2 starting
12:23:27 INFO MintWorker.run: CPU worker 3 starting
12:24:38 FINE MintWorker.run: Worker 0: 0,11 MHash, 0,0018 MHash/s
12:24:38 FINE MintWorker.run: Worker 3: 0,11 MHash, 0,0018 MHash/s
12:24:38 FINE MintWorker.run: Worker 2: 0,11 MHash, 0,0018 MHash/s
12:24:38 FINE MintWorker.run: Worker 1: 0,11 MHash, 0,0018 MHash/s

Also tried to set gpuDevice=0,512 and gpuIntensity=512 - not working.
My Athlon X4 750k (weak CPU) doing 0,0072MHash/s and you 2x 290x doing around 0,0088MHash/s at SCRYPT. You say "no problems", really?

Good point this is really slow. But i dont have any errors like you and did not say it is fast either. Both gpus are loaded 99%.

@ScripterRon

gpuDevice=0,8 etc doesnt even start the minting process, always get out of memory(current java memory usage 2,7 gb). local size: 256, global size 11264 hmm,
« Last Edit: January 22, 2015, 11:26:02 pm by crimi »
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #156 on: January 22, 2015, 11:44:52 pm »

11:16:47 INFO Main.lambda$null$2: GPU device 0: AMD Accelerated Parallel Processing 2048MB global memory, 32KB local memory, 20 compute units, Max work group size 256
11:16:47 INFO Main.lambda$null$2: GPU device 1: AMD Accelerated Parallel Processing 2048MB global memory, 32KB local memory, 20 compute units, Max work group size 256
11:16:47 INFO Main.lambda$null$2: GPU device 2: AMD Accelerated Parallel Processing 2048MB global memory, 32KB local memory, 20 compute units, Max work group size 256

If starts with gpuDevice=0,8 gpuDevice=1,8 gpuDevice=2,8
Quote
11:19:40 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:19:40 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:19:40 INFO MintWorker.run: GPU worker 0 starting on GPU 0
11:19:41 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:19:41 INFO MintWorker.run: GPU worker 1 starting on GPU 1
Exception in thread "main" java.lang.OutOfMemoryError: Java heap space
        at org.ScripterRon.NxtMint.GpuScrypt.<init>(GpuScrypt.java:207)
        at org.ScripterRon.NxtMint.GpuFunction.factory(GpuFunction.java:59)
        at org.ScripterRon.NxtMint.MintWorker.<init>(MintWorker.java:86)
        at org.ScripterRon.NxtMint.Mint.mint(Mint.java:93)
        at org.ScripterRon.NxtMint.Main.main(Main.java:306)

If starts with gpuDevice=0,8 gpuDevice=1,8
Quote
11:21:32 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:21:32 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:21:32 INFO MintWorker.run: GPU worker 0 starting on GPU 0
11:21:33 INFO MintWorker.run: GPU worker 1 starting on GPU 1
11:21:33 FINE MintWorker.run: Worker 1 starting on counter 1
11:21:33 FINE MintWorker.run: Worker 0 starting on counter 1
#
# A fatal error has been detected by the Java Runtime Environment:
#
#  EXCEPTION_ACCESS_VIOLATION (0xc0000005) at pc=0x000000006c6e7957, pid=6980, tid=5196
#
# JRE version: Java(TM) SE Runtime Environment (8.0_40-b21) (build 1.8.0_40-ea-b21)
# Java VM: Java HotSpot(TM) 64-Bit Server VM (25.40-b24 mixed mode windows-amd64 compressed oops)
# Problematic frame:
# V  [jvm.dll+0x187957]
#
# Failed to write core dump. Minidumps are not enabled by default on client versions of Windows
#
# An error report file with more information is saved as:
# C:\nxtmint\hs_err_pid6980.log
Logfile: pastebin.com/sfWAfa7U

If starts with gpuDevice=0,8 - for 15-20 seconds I see load of my 1 GPU and after 15-20 seconds I got system message, video driver crashed. GPU load 0%. But in NxtMnt cmd window no errors:
Quote
11:28:26 FINE GpuScrypt.<init>: GPU local size 8, global size 5120
11:28:26 INFO MintWorker.run: GPU worker 0 starting on GPU 0
11:28:26 FINE MintWorker.run: Worker 0 starting on counter 1

Changes in gpuIntensity given nothing, try set 1 and 10.

Start command:
java -Djava.library.path=C:\nxtmint -Xmx2048m -Dnxt.datadir=C:\nxtmint -jar C:\nxtmint\NxtMint-1.2.0.jar
In c:\nxtmint I'm copy all files and folders from aparapi archive and all files and folders from AMD APP SDK x64 - C:\Program Files (x86)\AMD APP SDK\3.0-0-Beta\bin\x86_64 (in previous I'm copy only OpenCL.dll from AMD APP SDK folder). Drivers - AMD Catalyst Omega 14.12.
Each work item takes 128KB.  With 5120 work items (256*20), that is 660MB.  If you start three cards, that means you need a Java heap size of 3*660MB or 1920MB.  I can change the global size to be (number of compute units) x (local size) instead of always using 256.  This would reduce the storage requirement in your example to (8*20) work items or 20MB.  Running 3 cards would then require just 60MB.

The JVM crash might be related to the large data allocation as well.  So the above change might fix that problem.

The drawback is that the hash rate will decrease since you won't be keeping all of the cores busy if you set the local size less than the number of cores per compute unit.  So it is a trade-off between local memory contention and huge data allocations.  I think we need another parameter on gpuDevice to specify the global size so you can vary the both local and global sizes.  That way, if you have 32 cores per compute unit and 20 compute units, you could specify gpuDevice=0,8,640.  You Java heap size would be just 80MB and all of the cores would be busy.

I'll put up a test version once I have made the change.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #157 on: January 22, 2015, 11:48:40 pm »

gpuDevice=0,8 etc doesnt even start the minting process, always get out of memory(current java memory usage 2,7 gb). local size: 256, global size 11264 hmm,
Yep - I'm going to have to change that.  Multiplying the compute unit number by 256 leads to outrageous group sizes and global storage requirements.  I'm going to change gpuDevice to have 3 parameters: device number, local size, global size.
Logged

cbw

  • Newbie
  • *
  • Karma: +2/-0
  • Offline Offline
  • Posts: 12
    • View Profile
Re: NxtMint Java minter
« Reply #158 on: January 23, 2015, 12:01:47 am »

Each work item takes 128KB.  With 5120 work items (256*20), that is 660MB.  If you start three cards, that means you need a Java heap size of 3*660MB or 1920MB.  I can change the global size to be (number of compute units) x (local size) instead of always using 256.  This would reduce the storage requirement in your example to (8*20) work items or 20MB.  Running 3 cards would then require just 60MB.
If I understand correctly - it have setted up 2048MB of RAM in config and each GPU have 2048MB - 2048MB>1920MB, it couldn't be a problem.

The JVM crash might be related to the large data allocation as well.  So the above change might fix that problem.
OK, waiting for fix and test again.
Logged

ScripterRon

  • Hero Member
  • *****
  • Karma: +75/-2
  • Offline Offline
  • Posts: 523
    • View Profile
Re: NxtMint Java minter
« Reply #159 on: January 23, 2015, 12:58:32 am »

If I understand correctly - it have setted up 2048MB of RAM in config and each GPU have 2048MB - 2048MB>1920MB, it couldn't be a problem.
Java needs heap space for itself as well.   Try 1920+192=-2122MB or greater.  But that is still a lot of storage to allocate and might be causing PCI-E bus problems.  The fact that you get a JVM crash for a small allocation size indicates there is still a problem.

I'll post a test version of NxtMint which will allow you to set the global size.  I'm also going to look into directly running native OpenCL programs.  This would eliminate the Java heap usage since storage will be allocated only on the adapter.
Logged
Pages: 1 ... 6 7 [8] 9 10 ... 20  All
 

elective-stereophonic
elective-stereophonic
assembly
assembly