Also, comment from David Anderson on the request for OpenCL support for Ivy Bridge GPUs in BOINC:
Quote:
]We're handicapped right now by lack of computers with ivy bridge processor. Does anyone have one they can loan or give us? (we'll buy one if needed, but our funding is tight these days)
From looking at the BOINC Git repository, the BOINC developers are working on adding OpenCL - Intel Ivy Bridge GPU detection to the BOINC client. Refer to the notes from Dec 5th - 8th in the below URL:
BOINC 7.0.40 now in testing has some preliminary OpenCL detection for the Ivy Bridge embedded GPUs. It lacks being able to work with XML though, plus there've been at least 2 bugs found already since .40 was built.
BOINC 7.0.40 now in testing has some preliminary OpenCL detection for the Ivy Bridge embedded GPUs. It lacks being able to work with XML though, plus there've been at least 2 bugs found already since .40 was built.
7.0.64 is available as stable for quite a few days now with working Intel OpenCL support.
The lunatics from kwsn already have a working beta for Seti.
Support for features like Intel® Quick Sync Video and OpenCL* in
systems with discrete graphics on Windows 8
Now, one can use both Intel Quick Sync Video and OpenCL even when Intel® HD Graphics is not the primary display adapter
This requires Intel Graphics driver to be installed and will work only on Windows 8 platforms
I can't test what boinc 7.0.64 would detect due to lack of willingness to install W8 but maybe someone else shows us the output with such a combination?
Question, is OpenCL on Intel even fast enough to be useful? What is the power efficiency? The AMD graphics cores are much more powerful yet it's even questionable whether they're worthwhile on OpenCL projects.
The AMD graphics cores are much more powerful yet it's even questionable whether they're worthwhile on OpenCL projects.
The actual A10 APU crunches as fast as a GT430. So where do you want to set the limit for 'worthwhile' ?
The next step could be: port all apps to gpu and remove all cpu-apps. I don't think that this is the right way.
Intel will upgrade the onboard gpu with the next generation (coming soon), tests speak about a performance increase of 50%.
On the other hand some project (like Albert) generate wu's for ARM devices. My Nexus, crunching for pogs, has a RAC of 477 running on 50% of the cpu's and shares the time with Albert. This is a bit better than low-end cpu's.
I mean, this is a science project. Science is the way to understand things and make it happen, not to maximize production.
This is why I would vote 'YES' if asked Intel OpenCL Y/N.
Intel will upgrade the onboard gpu with the next generation (coming soon), tests speak about a performance increase of 50%.
That would be nice, still not fast, but nice.
It's been a while so my memory of this might be slightly off but when I tested a beta-app over at Seti the HD4000 had about the same performance or slightly better as one of the cores on my i7 3770K @ 4.2 GHz with HT turned on. So not in the same league as a dedicated GPU but not to bad and it uses less than 10 W according to GPU-Z.
There are lots of apps that GPUs can't do. Why not use CPUs exclusively for them?
Absolut YES, but I think that is what we're doing when leaving one core free for the gpu's. This is how I see the AMD APU's. A low/mid-range Graphic card with onboard supporting cpu with the power to run 3 more cpu wu's for a fair price on a cheap MB.
The actual developement @ ARM is: 4 low end cpu's for housekeeping and 4 power cores for performance together on one chip. Let's see what happens, maybe the PC-world picks up this idea.
Also, comment from David
)
Also, comment from David Anderson on the request for OpenCL support for Ivy Bridge GPUs in BOINC:
From looking at the BOINC Git
)
From looking at the BOINC Git repository, the BOINC developers are working on adding OpenCL - Intel Ivy Bridge GPU detection to the BOINC client. Refer to the notes from Dec 5th - 8th in the below URL:
http://boinc.berkeley.edu/trac/browser/boinc/checkin_notes
BOINC 7.0.40 now in testing
)
BOINC 7.0.40 now in testing has some preliminary OpenCL detection for the Ivy Bridge embedded GPUs. It lacks being able to work with XML though, plus there've been at least 2 bugs found already since .40 was built.
RE: BOINC 7.0.40 now in
)
7.0.64 is available as stable for quite a few days now with working Intel OpenCL support.
The lunatics from kwsn already have a working beta for Seti.
Any news regarding this from Einstein?
Intel released their newest opencl 1.2 drivers about a month ago together with their new opencl tweaker Intel Vtune: http://software.intel.com/en-us/articles/intel-vtune-amplifier-xe-getting-started-with-opencl-performance-analysis-on-intel-hd-graphics
According to http://downloadmirror.intel.com/22610/eng/ReleaseNotes_Gfx_3071.pdf
the newest driver supports v1.2 on HD4000 and HD2500 igpu's and there is also a note which might fix the nvdia opencl killer:
I can't test what boinc 7.0.64 would detect due to lack of willingness to install W8 but maybe someone else shows us the output with such a combination?
Question, is OpenCL on Intel
)
Question, is OpenCL on Intel even fast enough to be useful? What is the power efficiency? The AMD graphics cores are much more powerful yet it's even questionable whether they're worthwhile on OpenCL projects.
RE: The AMD graphics cores
)
The actual A10 APU crunches as fast as a GT430. So where do you want to set the limit for 'worthwhile' ?
The next step could be: port all apps to gpu and remove all cpu-apps. I don't think that this is the right way.
Intel will upgrade the onboard gpu with the next generation (coming soon), tests speak about a performance increase of 50%.
On the other hand some project (like Albert) generate wu's for ARM devices. My Nexus, crunching for pogs, has a RAC of 477 running on 50% of the cpu's and shares the time with Albert. This is a bit better than low-end cpu's.
I mean, this is a science project. Science is the way to understand things and make it happen, not to maximize production.
This is why I would vote 'YES' if asked Intel OpenCL Y/N.
RE: The actual A10 APU
)
Good question.
There are lots of apps that GPUs can't do. Why not use CPUs exclusively for them?
That would be nice, still not fast, but nice.
Was just asking a question because I'm not sure of the answer. Thanks for the response.
RE: RE: Intel will
)
It's been a while so my memory of this might be slightly off but when I tested a beta-app over at Seti the HD4000 had about the same performance or slightly better as one of the cores on my i7 3770K @ 4.2 GHz with HT turned on. So not in the same league as a dedicated GPU but not to bad and it uses less than 10 W according to GPU-Z.
RE: There are lots of apps
)
Absolut YES, but I think that is what we're doing when leaving one core free for the gpu's. This is how I see the AMD APU's. A low/mid-range Graphic card with onboard supporting cpu with the power to run 3 more cpu wu's for a fair price on a cheap MB.
The actual developement @ ARM is: 4 low end cpu's for housekeeping and 4 power cores for performance together on one chip. Let's see what happens, maybe the PC-world picks up this idea.