Folding Forum

intel Iris 6200 supports openCL 2.0
device ID:1622 (showed on intel HD Graphic control panel)

is there any plan to support it for calculations?

I have no specific information about support for any Intel GPUs, but they've always been somewhere on the wish-list.

The last time Intel support was tested internally, the OpenCL drivers were considerably less that reliable, and the potential output was pretty small. More recently, the total development backlog has grown considerably and it's usually more important to finish important projects that have been started than to re-look at other work that might be undertaken.

Then, too, developing and testing a new FahCore is not a simple task, even after the project gets an official go-ahead.

In other words
"Any plan?" -- yes, sort of.
"Any time soon?" -- probably not.

... but as I said, I don't have any specific information to support those hunches.

--------------

Are there any sources indicating how many of these GPUs have been deployed and/or any reports from other OpenCL developers?

The on-chip GPUs tend to be under-powered, and like bruce said the size of the target audience is unknown, but likely (small and) growing. Also, the current GPU projects in FAH are very large, targeted towards the very powerful NV and AMD GPUs. Not only would they need to develop new cores to run on Intel GPUs, but also make new projects sized towards the lesser powered chips.

But contrast the power of Folding on a phone, to folding on an iGPU, the iGPU is a magnitude or two more powerful.

I have read where Dr. Pande has commented about folding on Intel chips. Pande Group is keeping an eye on the progress Intel is making (as with all new and emerging tech), and when there is both an opportunity, and a cost vs benefit advantage, building an Intel core would be strongly considered. So bruce's ETA sounds about right as well.

Hi - I think Intel's Iris 6200 should be sufficiently powerful. As a comparison, I use NVidia 740M on a Windows laptop. According to (the web did not permit me to add the site) 740M has performance of 770 G3D mark. Iris Pro 6200 has a value of 1550 of G3D mark. It does show good progress.

While it might not be a top priority, it's performance show good potential.

Of course, it becomes complex to define which chip has good enough internal video card in order to be included, but there is some computational performance to be picked up.

As mentioned, eventually the Intel Iris graphics may get a folding core developed for them. But the OpenCL drivers from Intel would need to be provided first that are sufficiently reliable to perform the calculations, and PG would have to see a benefit that justified the cost in time and developer resources needed to create, test, and support a new GPU core for Intel.

Thanks for the quick response.

I will try to see if I can help.

Clarification is in order: since I have done some OpenCL work earlier, I might give that a try.

If you're feeling adventurous, you can try building OpenMM from source and experiment with running the tests inside the solution.

It should be noted that iGPUs, whether from Intel or NVidia or AMD/ATI, are typically designed to use low-power (for laptops, etc) and are quite reasonable for producing good Graphics but they're not outstanding performers compared to the capabilies of discrete GPUs in a desktop environment. While all contributions are valuable, the FAH developers have emphasized code designed to make maximum use of highly parallel floating point calculations on those discrete GPUs.

Personally, i'm not familiar with the various iGPUs from Intel. In a casual search for Stream Computing performance benchmarks (often via OpenCL) I found rather sparse information comparing the performance of Intel/AMD/NV, including the Iris 6200. (Video performance is not particularly useful to FAH, so should not be used to determine the ultimate contribution of this particular platform to the efforts of developing new FAH software and new FAH infrastructure.

Hi Bruce - Intel Iris 6200 was introduced (maybe) ~6 months ago in a few select models. Outside of the new MacBook Pro 15 late 2015 model, I do not know of a consumer product offering it. I think there are a few others - I just don't know about them Regardless, the iris pro is the top-of-the line Intel graphics, so it is a reasonable choice.

On a personal note, I do value lower power products, since they can do good performance on very good performance-per-watt. I.e. I ran Folding at home regularly during winters, since I have to heat my apartment regardless, so the heat stays within the apartment. I understand this sounds unconventional.

Ben - yes, I have to give compiling openMM a try. I have been putting it off for a long time, I should not postpone it again.

I know that the other folding project poem@home has an Intel OpenCL core in alpha stage.
Maybe they can share some of their knowledge. But I expect OpenMM to be a world of its own.

foldy - there is openCL as part of openMM. Nevertheless, there were issues building it on the machine that I have available.

Unfortunately, there was this ominious message on http://docs.openmm.org/6.1.0/userguide/ ... ing-openmm:

Important: A serious bug was introduced in Mac OS X 10.7.5 that prevents OpenMM’s OpenCL platform from working correctly. At the time of this writing, the bug is present in all versions from 10.7.5 onward. The CUDA platform (see below) is not affected by the bug, so if you have an affected version of OS X, you should use it instead of the OpenCL platform.

The hardware that I had available was a Mac Air 2013, whose graphics are Intel's openCL 1.2 standard (Haswell's generation). Maybe 2/3 of OpenCL test examples failed - I'd either have to find a non-Mac OS platform to test, or look for the reasons the message warns against Mac OS 10.7.x+/openCL bug. While I might be able to do both, I doubt I am going to be quick.

Anandtech indicates that Intel Iris 6200 supports double-precision [1]. Though, it might differ on mac platforms [2] and I can't find whether the article refers to Khronoses's double precision standard or not.

I do not have an Iris Pro-CPU easily available to investigate at this point. Once I assemble a new machine, I should be able to test it further.

[1] http://www.anandtech.com/show/9960/inte ... crete-gpus)
[2] https://github.com/SRA-SiliconValley/we ... issues/536

Any technical hurdles aside, Pande Group's limited development resources are spent on a cost vs benefit basis. High volume high performance discreet GPUs is a good bet. Low volume low power non-discreet GPUs already running a CPU client on the other half of the chip, not so attractive.

Pande Group has already said they are watching the progression of iGPUs. If and when it becomes viable, or maybe when a development partner like Sony on the phone client steps up, we will see a client for this. Standard operating procedure.

holdingpattern wrote: Unfortunately, there was this ominious message on http://docs.openmm.org/6.1.0/userguide/ ... ing-openmm:

Important: A serious bug was introduced in Mac OS X 10.7.5 that prevents OpenMM’s OpenCL platform from working correctly. At the time of this writing, the bug is present in all versions from 10.7.5 onward. The CUDA platform (see below) is not affected by the bug, so if you have an affected version of OS X, you should use it instead of the OpenCL platform.

There was a later notice posted nearly a year ago that the OS X bug with using OpenCL had been fixed. It has been mentioned in posts asking about whether a GPU core would become available for OS X.