Looks like the simplest calculation of the number of cores in a Quad CPU is a hard work. After starting the 6.3.14 only 3 tasks are running. So I had to create a new cc_config.xml to tell the program the real number of cores in my CPU. After the Entry ncpus=4 the program is running fine.

---

Second comment to the 6.3.14:

Finally the 6.3.14 don't eat memory when the BOINC-Manager is running under Vista 64 bit :)

Please change the entry

Aktiv (1.00 CPUs, 1 CUDA) to Aktiv (1.00 Cores, 1 CUDA)

I think, you mean to control the feeding of the GPU is one core needed and not the whole CPU wirh all 4 cores ;)

Question:
Is the number of 2 cores hard coded in this version? I'm wondering, while 2 CPU tasks und one GPU tasks are running on a Quad if no ncpus-entry is available with the real number of cores in Windows.

A team mate told me, on Linux the right number of tasks is started. I will ask back for a cc_config.xml.

EDIT: He has running a cc_config.xml with ncpus=5, so I think, the Linux-Version has the same problem.

---

I've been running x2 hosts for about 3-4 hours. I just installed on 3rd host a few minutes ago. All are Windows XP x86

I did not want to post immediately until some testing had been done. However I have not had any problem so far and been in contact with the developer (Anderson) all day about it. I observed proper task switching when a CPU task ended another was started, so I was always left with max running tasks.

The problem with reading the XML preferences file has been fixed and the proper venue is now used. XML is parsed correctly again.

It seems to be good so far, It starts the proper number of tasks on initial start up, in my case without using ncpus setting I should have 2 and thats what I got. I did not have to do any suspending or other tinkering to get it to work. 1 CPU only and 1 CPU/CUDA task. You will note immediately CUDA tasks show as "Running (1.00 CPUs, 1 CUDA)". CPU only tasks just show as "Running". This amount of CPU used can be set by the project, currently 1.00.

The behavior is suppose to be:
Anderson wrote:

The intended behavior (which I think we've finally achieved)
is that the client runs tasks until all the GPUs and all the CPUs
are utilized.

E.g. if it has a (1 GPU + .75 CPU) job and several 1 CPU jobs,
and your machine has 1 GPU and 2 CPUs,
it will run the GPU job and 2 of the 1-CPU jobs
(so that CPU usage is 2.75, i.e. there will be contention
for the CPUs, but that's better than leaving one of them .25 idle).

BTW, I changed things so that if a job uses a GPU and < 1 CPU,
it gets run at normal process priority.
Hence in this case the GPU job will get .75 CPU
(so that the GPU runs at full speed)
and the two CPU jobs will share the remaining 1.25 CPU.

Second comment to the 6.3.14:

Finally the 6.3.14 don't eat memory when the BOINC-Manager is running under Vista 64 bit :)

Please change the entry

Aktiv (1.00 CPUs, 1 CUDA) to Aktiv (1.00 Cores, 1 CUDA)

I think, you mean to control the feeding of the GPU is one core needed and not the whole CPU wirh all 4 cores ;)

Question:
Is the number of 2 cores hard coded in this version? I'm wondering, while 2 CPU tasks und one GPU tasks are running on a Quad if no ncpus-entry is available with the real number of cores in Windows.

A team mate told me, on Linux the right number of tasks is started. I will ask back for a cc_config.xml.

EDIT: He has running a cc_config.xml with ncpus=5, so I think, the Linux-Version has the same problem.

No linux version of 6.3.14 has been made yet. There hasn't been a linux version made since 6.3.10.

The number of cores is not hard coded in any version. You can limit it with your preferences, cc_config.xml or global_prefs_override.xml files.

Check your preferences and any files you might have modified and make sure they are set to use 100% of CPU's and correct values for such things as ncpus.

I've installed the 6.3.14 on a second PC, also on Vista 64 bit. There are 4 tasks running. This PC has a 8800GT. Don't know, why the other only 3 tasks has running without a cc_config.xml at the first start.

I've deleted the entry in the cc_config of the other PC with the GTX280 and after the restart now also 4 tasks are running. The CPU is set to 100% on both PCs.

---

No linux version of 6.3.14 has been made yet. There hasn't been a linux version made since 6.3.10.

If you have the sources and the changes and a compiler ... ;)

---

Another remark to the 6.3.14:

The temperature of my GTX280 is dropped from 79°C to 72°C. In the past this only happens with ncpus=5 and a longer running time per WU. I'm really agog for the new time per step, the last with the 6.3.10 was 34.438 ms.

---

You will note immediately CUDA tasks show as "Running (1.00 CPUs, 1 CUDA)". CPU only tasks just show as "Running". This amount of CPU used can be set by the project, currently 1.00.

Do you know when/if the CPU amount will be changed to lower than 1, for somebody, such as myself, who prefers to maximize the number of CPU tasks instead of having one tied up in each GPU task?

Once a new BOINC release proves to work we will reduce the cpu usage to less than one. We will also need to upload the new Windows and Linux application. These things are better separated by an interval to detect problems.
It will be up to you to decide if use the CPU it or not (It depends on the speed of your GPU).

GDF

The first WU only running under 6.3.14 is ready:

http://www.gpugrid.net/result.php?resultid=75290

# Time per step: 35.483 ms
# Approximate elapsed time for entire WU: 30160.337 s

Looks good!

---

This is as before, isn't it?

gdf

...... where can I get 6.3.14?

Thanks!

...... where can I get 6.3.14?

Try a look here:
http://boinc.berkeley.edu/dl/?C=M;O=D

---

Member of BOINC@Heidelberg and ATA!

This is as before, isn't it?

gdf

Nearly, a little bit slower ...

6.3.14 - # Time per step: 35.483 ms

... and the last 3 WUs without a ncpu-entry with the 6.3.10
6.3.10 - # Time per step: 34.438 ms
6.3.10 - # Time per step: 34.608 ms
6.3.10 - # Time per step: 34.547 ms

---

.....Got it - thanks.

My first result crunched with 6.3.14 is uploaded - http://www.ps3grid.net/result.php?resultid=75292

It took about 43.6ms/step which is about the same speed than with previous BOINC versions (was always between 43 -45 ms/step).

---

pixelicious.at - my little photoblog

Worked like a charm, 4 tasks, 3 cpu and 1 gpu.

Did the cc_config.xml to 5 cpus and now its 4 cpu, 1 gpu.

Will have to wait a day or two to average the performance.

I installed the new 6.3.14 this morning and it immediately Trashed the 4 PS3 Wu on the Computer and has Trashed every one since then. Going back to 6.3.10 hasn't helped either as it now Trashes any PS3 Wu I download ... All I get is Computation Errors with the following message:

<core_client_version>6.3.10</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
# Using CUDA device 0
Cuda error in file 'deviceQuery.cu' in line 59 : out of memory.

<core_client_version>6.3.14</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
# Using CUDA device 0
Cuda error in file 'deviceQuery.cu' in line 59 : out of memory.

out of memory

Did you reboot your machine?

MrS

---

Scanning for our furry friends since Jan 2002

...did install v6.3.14 yesterday - everything is running fine ... :-) speed is the same (as I supposed). On a 32 bit Windows XP prof. SP 3

nice job!