You want green computing then buy an Abacus.

With rare earth mining, zero recycling and heat losses computing ain't green.

It is what it is.

Nvdia is at a disadvantage with gpu hpc. The mid range sale point of earlier designs will be replaced by the entry point for Sandy Bridge and Fusion.

The development costs have to increase due to that market loss.

AMD does not suffer that defficiency, they simply use the slightly oder design in the next product cycle refresh. The Top Radeon core will be a Fusion core within two tears. And they'll sell millions of them.

"Yet, IT Can be easily Done with New Drashek time Inverter"

That is the funniest thing I've heard all day!

Although Nvidia has a lead, I still do not count AMD out. The HPC market at the lower end is very much in growth. Since the prices have come down, developers might even be able to dabble in both and play against the strengths. Sometimes cornering the market is not necessary for all. It is important for Nvidia as the need new markets because they have been shut out of one of their bigest besides the GPU, the chipset.

There is a paper about GPU memory errors: "Haque IS and Pande VS. Hard Data on Soft Errors: A Large-Scale Assessment of Real-World Error Rates in GPGPU. In Proceedings of 10th IEEE/ACM International Conference on Cluster, Cloud, and Grid Computing (CCGrid 2010), pp 691-696."
It does point out a need for better memory error checking and memory error recovery procedures, because the GPU memory errors are more as a norm than exception. In that light it can be argued that having ECC is a good thing.

There is a tool or two for the GPU memory testing: https://simtk.org/home/memtest/

CUDA might not be open, or even a standard, but history tells us that such technicalities never stopped other languages from attaining widespread popularity. Being policed by IBM didn't stop Fortran from still being the numerical language, half a century after it first appeared. Even with Sun Microsystems' best efforts to create a cumbersome 'framework' and employ licensing peculiarities, Java's popularity has managed to surpass C. It has happened before and it's looking like history will repeat itself.
-Some may believe the above the grail of future prophetic wisdom - NOT- totally disregarding CUDA, the paragraph merely states the progression of languages from Fortran, C, to Java - each offering advantages over previous/ for specific tasks- If at all you believe language use defines the better language, the web site TIOBE.com keeps tabs of language usage..BTW CUDA which was introduced in '07, compared to OpenCl (08'), is not in the top 100- I recommend you check the use of OpenCL/GL in your cell phone.
asH

How many popular apllication use the ATI's Stream?

ONE ( ATI FoldingHome, at half the speed of any NVIDIA card ).

How many use CUDA, PhysX or NVIDIA's OpenCL? A lot ( Octane, Arion, Photoshop, xNormal, VRay, LuxRender, Badaboom, Nero, Roxio, PowerDVD, Lightworks, PantaRay, Just Cause 2, etc... ).

AMD's OpenCL SDK is completely immature.
AMD simply does not deliver for non-game apps. And, for games, their tessellator is ridicuously slow.

CUDA and NVIDIA are absolutely superior for GPGPU.

"says it's something AMD simply hasn't needed"
The point of ECC is that it is a customer requirement and not anything to do with the manufacturer.
"the need for ECC is mitigated by testing done in AMD's labs prior to shipping boards".
Sure, it's good to know AMD has eliminated faults caused by alpha particles from the process and packaging they are using.
ECC is needed in larger systems and fault tolerant embedded systems where the probability of flipped bits and consequences of failing devices are causing unacceptably high risk. The marketing people of the AMD should be perfectly aware of this.

Moore intresting Facts. Take test strip roll & mark 1/100th of inch for 100 miles 1,200 x 5280= 6 million x 100 =600 million strokes thru machine in 1 sec = 100 x (1x 528,000 x 3,600) = about 1/10th speed of light. Now WE Know How Fast Av Computer Is Stroking FIRE. About as fast as average Brain.

Since WE Know ONE IS Lonelyest Number. Getting Mary Is Like Stepping Back IN Time. About 800 Trillion Years saving Light Hydrogen that surrounds Milky Way. thats Long time to wait to get back here, to theINQ.

Yet, IT Can be easily Done with New Drashek time Inverter. Using Above Art as Command Center, titan V Rocket with Sodium Peroxide wound around Chromium Peroxide, Sodium Starts Long Journey, Then electric ring of firing pins on thruster nozzle Bumps explosion back into vortex & fuel area, AfterBurning Chromium in one gasp, NuClear Fire & small force holds whole together, while forward thrust approachs light speed instantly. See You Then.

"theoretically AMD could create a CUDA compliant driver that could run code on its GPUs"

You can also turn it around, what is stopping Nvidia to make is easy to convert to OpenCL so it work on both GPU's

I vote for OpenCL we have already enough standards that are under control of one company.

a ... a... a decent article on the Inq? Surely you jest?

Where are those Universities that teach you only CUDA and not both CUDA/OpenCL? Around here we spend equal time with both languages.

Nvidia misses OpenGL 4.0 promises

http://www.semiaccurate.com/2010/04/12/nvidia-misses-opengl-40-promises/

Nvidia fanbois

The truth is, AMD simply didn't take using GPUs for HPC seriously. Perhaps it thought that its Opteron chips could cut the mustard or maybe it was just a lack of vision, but either way it let Nvidia take the HPC lead.

AMD wasn’t able to compete in the HPC arena because the parts were not in place- their APU platform required a meld between hardware and widely accepted Open(X)platform- that didn’t happen until this year...do your research!!
asH

nicely written and summarizes all the companies mentalities.

last time i checked (a couple of months back), amd doesn't seem to that much interested in gpgpus. just look at their forum support for stream/opencl (there is hardly anyone from their supprot replying)

surely packetshader in place of packershader...