Bitcoin Forum

I want a full explanation of why you cannot mine on this, other than non opencl comparability(why not)
and any ideas on how to circumvent restrictions and make a miner.

Intel doesn't provide any OpenCL SDK.  There is no way to "program" the GPU.

because its specialized hardware, basically is can only draw lines and circles on the screen, thats why its called a graphics card.
ati and nvidia, have also provided some interface, to use the card to other things then drawing... this is possible becuase much of the stuff on a gpu, is similar to a cpu, except that its much simpler.

Thanks for the reply's guys, is the limitation hardware based or could as super assembly coder create some sort of programming interface?

It is both.

1) the hardware has to be capable of being programmed
2) you need an interface.

no super coder is going to program hardware they have no data on. 

A couple thousand coders with specs provided by AMD haven't made OpenCL capable drivers for AMD GPUs.  Doing it in the dark is never going to happen. 

It isn't much loss.  The GPU is beyond worthless.  The cheapest (and I do mean cheapest as in $49 special) AMD graphics card has much higher computation power than Intel's GPUs. 

All hope is gone...
Thank you for ending my delusion.

It's also slow as fuck, right guys?  Even if it could mine, it wouldn't even make a few cents a day....

its incomparable, its like comparing a toaster and a spoon. you will not get anything useful of comparing.

That's untrue, starting with sandy bridge and up Intel's IGP supports OpenCL.
Their SDK can be downloaded from here:
http://software.intel.com/en-us/articles/vcsource-tools-opencl-sdk/

Cant DL right now, but will definitively look at it tomorrow and possible slap something together even if its shit .

I am sure someone who knows more could estimate the speed, but if someone went through all of the work with programming just for it, it might do a few times more then a good CPU. 

Manufacturer   Intel

HD Graphics 3000 12@350-1350MHz


Codename   Sandy Bridge
Pipelines   12 - unified
Core Speed *   350-1350 MHz
Shader Speed *   350-1350 MHz
Shared Memory   yes
DirectX   DirectX 10.1, Shader 4.1
technology   32 nm

Do you read before you claim things are untrue....


http://software.intel.com/en-us/articles/opencl-release-notes/


See any GPU in there?  Intel has long supported OpenCL on CPU ONLY.  I will wait for your retraction.  If you still don't believe me install the SDK and try to run the samples on an Intel GPU.  No need for a miner to prove it is possible just use their samples.

No intel GPU to date supports OpenCL.  No ifs, and, or buts.  That may change in the future but given the lackluster performance I doubt Intel wants a direct comparison.  They would much rather companies buying $8,000 server CPUs.

Rumor has it that the upcoming Ivy Bridge IGP will support OpenCL.
"The Ivy Bridge GPU adds support for OpenCL 1.1, DirectX 11 and OpenGL 3.1"
See http://www.anandtech.com/show/4830/intels-ivy-bridge-architecture-exposed/5 for more details.

As for why it's so difficult to mine on IGPs, it's because they weren't designed to compute. AMD and Nvidia specifically gave the hardware computational abilities. That's why you cannot mine with an old ATI 3870 or Geforce 7900GTX. They came out before APP and CUDA.

I could be wrong here but i think HD graphics is embedded into all of those chips.

You would be wrong. ;D

Im sorry if I was mis-informed in making my claim, but several "white-sheets" from Intel suggests it may be possible.

http://software.intel.com/en-us/articles/introduction-to-intel-opencl-tools/ (http://software.intel.com/en-us/articles/introduction-to-intel-opencl-tools/)
"There are multiple tools packaged under Intel GPA, for OpenCL we will only use Intel GPA monitor (to collect traces) and Intel GPA Platform Analyzer (to view traces).

Intel GPA Platform Analyzer aligns clocks across all cores to provide uniform timeline for CPU and GPU workloads and running tasks/threads."

Also another interesting find: http://software.intel.com/en-us/forums/showthread.php?t=79766 (http://software.intel.com/en-us/forums/showthread.php?t=79766)
Intel rep, says it might be possible in the future.

For now all that Intel has said is:
http://software.intel.com/en-us/articles/opencl-sdk-frequently-asked-questions/#9 (http://software.intel.com/en-us/articles/opencl-sdk-frequently-asked-questions/#9)
"11. Will Intel support OpenCL on its new platforms?
Intel will support OpenCL on future tools and platforms. Intel is evaluating when and where OpenCL support will intercept our products, but no announcement has been made."

Again sorry for my mistake, but hopefully OpenCl can be implemented in the near future with Ivy Bridge around the corner.

This.

OpenCL is currently only supported at the CPU level, in Intel CPUs. It will be supported in future on-dye GPUs of Ivy Bridge, but it isn't supported in Sandy Bridge GPUs atm.

Remember that "OpenCL" doesn't mean that it's GPU-only. You can even compile CUDA to run code on CPU, right now.

Here is my best guess:

1. CPU only = 1-2 mhash
2. Intel HD Graphics = 3-10 mhash
3. ATI video card = 200+  mhash

Until someone comes up with a miner that actually uses Intel HD... My laptop has a T4400, I think it has Intel 4 series graphics. I'll try your miner to see how many hashes I can do. Incidentally, suggest this to the guy making vanitygen.

As far as I know, the best use for intel HD is to use it for rendering desktop so the rest of GPUs in the rig are dedicated to mining. But that's only useful if you're using your rig for something other than mining as well.

Well, reviews are up for Ivy Bridge, but it looks like the OpenCL driver is still MIA. It might be awhile before we get any mining benchmarks for Ivy Bridge.

At what point would IB (and Haswell) make CPU mining relevant again? Or, more troubling, would it make botnets that much more dangerous? If IVB was good for 10MH/s and Haswell GT3 was good for 20MH/s, it could really add a new wrinkle to the casual mining market. Even though 10-20MH/s might not be a lot, when you're shipping 150M desktop PCs a year even a small percentage of users mining would make a large impact on the network.

Has anyone tried yet to run BrookGPU or some GLSL code on Intel HD hardware?

From:
http://software.intel.com/en-us/articles/vcsource-tools-opencl-sdk/

Wow.  Nice find.  So Intel is finally getting on board with GPU acceleration.  It likely is very academic but I wonder how much hashing power those crappy integrated GPU have.

Could be big... for botnets.

I bought a motherboard that supports using the built in Sandy Bridge GPU in parallel with added PCI-E GPU's....just in case I would ever want to use that extra GPU + monitor output.  Well, if someone can get it mining, let me know!  Even 25MH/s is a cool little boost.

I may try this, just got a new SBE board friday.

I think maybe if ckolivas is made aware of this SDK he could pump some code out in a week to get this tested.  Author of cgminer for those of you unaware.

cgminer is simply an engine for running open CL KERNELS.

With Intel OpenCL drivers you could simply use existing kernels however for optimal performance you would want an optimized kernel.

OK, then if it already works how do we pass options to cgminer to fire it up?

Try: cgminer -n

Should report any SDKs installed and any OpenCL compatible GPUs detected. Presumably the GPUs need drivers that advertise their OpenCL capability.

I will enable my Sandy Bridge GPU in the BIOS, install drivers, install Lucid Logix bullshit if I have to, install SDK, fiddle and then report if I can get stuff working.  Thanks.

Sandy-Bridge GPU is not OpenCL capable (unless I missed an announcement by Intel)

Ah shoot you are right!  This OpenCL SDK only supports Ivy Bridge integrated GPU's (HD 4000/2500).  Dammit!  Well, the SDK is detected in cgminer at least:

There is no data on Ivy Bridge OpenCL power yet? i know it should be weak, but it worth a look, Intel does miracles, when Quick Sync performance becomed public, no one could belive it....

QuickSync doesnt even use OpenCL or the gpu (as in shaders), its basically a built-in asic for videotranscoding. I find it very easy to believe they achieve that kind of performance.

That seems a bit optimistic IMO. I'm guessing it would be around 500khash.

LOL, you may be the first person to estimate even less than I would have.

It only has 16 stream processors, and the 5830 has 1600 stream processors.

My CPU has 64 stream processors (or can do 64 parallel computations) and I only get ~7Mhash, I'd be surprised if the Intel HD got over 5 Mhash.

Are Intel's "Execution Units" directly comparable to AMD's "Stream Processors"? I'd be more curious to see what the power consumption will be for the respective hash rate.

That's the million BTC question. You can look at the performance data from the 680 to see that a new CUDA core isn't comparable to at VLIW5 SP, so we likely won't know until someone tests it if they are comparable.
If one EU does equal one VLIW5 SP (and assuming they can run at the 1.15GHz boost rate all the time), you'd expect to see about 5MH/s out of HD4000. The 5830 only has 1120 SPs, btw.

We dont know!

Some reviews shows up that DirectCompute performance is well over than an AMD A8, its not the same than OpenCL performance, OK, but someone needs to try it out.

Just a peeve here - We had coders "tricking" GPUs into running semi-arbitrary code considerably faster than CPUs by writing it as custom shader routines, years before OpenCL hit the scene (Google for "GLSL" or "Cg").  So yes, the possibility does exist of mining on a completely non-CL-supporting platform.

That said, shaders (at least back then) had a fairly limited set of available operations they could do (mostly linear algebraic transforms on single or half precision floats, not bitwise ops on integers), and an incredibly small stack space (on the order of 256 bytes) - Good luck implementing SHA256 under those conditions.    ;D

Just found this...

http://www.chw.net/2012/04/llano-vs-ivy-bridge-duelo-en-opencl/

in spanish.

Did anyone get this working? I'd like to mine on my Intel hd 4000 cpu while waiting for bfl.

My ivy bridge hd 4000 should have opencl now

Intel 4000 can do OpenCL. In fact I'm mining on it right now ;) with 1150MHz you can get about 12MH/s with ~8W power consumption, unfortunately my GPU in that ultrabook is throttling so I get about 10MH/s with ~5W power consumption of the GPU.
Also, it only works under M$ Winsh!t Vista/7 since the OpenCL support for Linux was put on hold.

Hope it helps ;)

Tell me! I tried both 32 and 64 sdk and cgminer can't find cl devices :/

Hmm, I have small Win7 SP1 32bit partion only for mining. You need 9.17.10.2792 GPU driver from Intel and I currently use the 2013 BETA OpenCL SDK (http://software.intel.com/en-us/vcsource/tools/opencl-sdk-2013) (You need the same arch as your Win is eg 32bit OpenCL for 32bit Win), then I just downloaded the precompiled CGMiner (don't want to compile on Winblows :D BTW 2.11.3 it is) and started the app. Dunno why it doesn't work on your PC  ???

I'll keep trying. Wasn't sure whether arch needs to match os or app. I also have driver issues with Windows server 2012.

Who doesn't have issues with Windows? :D
Best would be to use Linux everywhere, but since the dev who worked on OpenSource OpenCL with Intel GPU (https://github.com/karolherbst/beignet) left the company there is only little hope...

BTW why not use the same arch for all? 64bit Win + 64bit SDK + 64bit app?

Got it working! My i7-3770 hd 4000 does 12mh/s. Thanks!

There is no cgminer x64 is there?

I forget how bad cpu mining is and can't really test it any more. But is 12mh/s much better than cpu mining? I miss mining and have been waiting for so long for bfl so this will be fun for a while.

Can anyone tell me cpu only mining speed of a i7-3770? Wanna make sure gpu mining is worthwhile and whether to try to add cpu mining. I don't think 12mh/s is bad for a "free" gpu though. Just not sure if it's any faster than cpu only.

check here

https://en.bitcoin.it/wiki/Mining_hardware_comparison#Intel

I think its 5 or 6 mhs. at what, ~70 watts?

vs 12 mh/s @ ~5 watts

Yeah amazing, mines using exactly 5.9w

If someone made us an optimized ocl kernel we could probably push a few more mhs :-)

That would be great IMHO. I know that this GPU is extremely low-end, but since I have free electricity and since I'm on that ultrabook almost all the time I like to use it :)

My home server is always on too. You're lucky you get the same performance ;-)

Any numbers on Haswell HD4600? it seems to wipe the floor with the 7660D on OpenCL performance...

I have one of these on my laptop and the Bitminter client mines on it, slowly  ;D:

Edit: Before anyone asks, no I don't mine on my laptop, I downloaded the client simply to see what it looked like and was surprised to see the Intel HD device available.

https://i.imgur.com/5AWBLqX.png

bitpop:  to answer some of your questions.. I have a similar chip and get about 12Mh/s on the Intel 4000 and 27Mh/s using all cores of the 3770, slightly over clocked.

If you want to play with it, I use minerd.exe, which I can't remember where I found, hehe.  I also use a mining_proxy locally to accept its work and translate it to stratum, since it only supports getwork.

So, basically, if you want the machine to do nothing else, that cpu/gpu can do about 39mh/s at considerable wattage (havent measured, because I don't mine with it, but maxing the cores of that chip is going to easily be over 100w.)

No way, the 3770 cpu is faster than a hd 400 gpu? I thought the worst gpu kills the best cpu. Oh well I stopped the hd 4000. I finally got some lancelots and didn't want to shorten the 3770s life.

Heh, now you know better... The worst AMD GPU is better than the best CPU. The same is not true for intel GPUs, or even Nvidia (to a lesser extent with the 210 for example). Intel GPUs are a token gesture with adequate graphics for most normal people; they are hopeless for any high performance application, including opencl.

So cpu mining isn't that bad
I was wasting hd 4000 because i couldn't cpu mine on the 3770 ;)
But then again the hd 4000 took 5 watts, not 75

+1
Folding@Home made a client for ATI 1900 cards wich ran on DirectX 9. Yup, science computing on DirectX 9.

Back then OpenCL didn't even exist, not even CAL (CAL is like CUDA for ATI, supported since ATI 2000 if i am right)  :D

My intel 3820-E is putting out 50 Kh/s, so yes, there are CPUs that can do better than some low end (< $100) GPUs.

Is it more price efficient? No.

mhs or khs? if it's khs then yeah any gpu would do better. I have a 5 year old nvidia that does 3 mhs

Can you please tell us how did youdo it?

Might want to read the dates on the thread before asking for 4 year old information