The GPU code seems to be relatively isolated to memory operations in gpuops.cu. Not altogether sure but a quick review suggests such mapping is supported by OpenCL. So one could rewrite that module and the whole thing would work without CUDA. Of course, in terms of compilers it is going to be a while before users can move away from nvcc for nvidia graphics card support.
However, as with the parent I'd really like to see a generic OpenCL vectorization kernel module. Strong suspicion that this work was directly or indirectly underwritten by Nvidia so I guess someone (Intel?) needs to step up and fund similar academic projects.
"KGPU is a project of the Flux Research Group at the University of Utah. It is supported by NVIDIA through a graduate fellowship awarded to Weibin Sun."
It really makes you wonder why AMD does not do the same thing; the hw is essentially free and how much could a fellowship cost. Does anyone know if research grants like this are a tax write-off?
I wonder if MSFT is subsidizing the cost? Either way it seems that market economics has now dictated that server operating systems are free as in beer.
I just created a MS free tier instance and found it was using Windows Datacenter edition. If I recall, with DC edition, if you buy a DC edition licence for a physical server and run only Hyper-V hypervisor role on it, you can have unlimited number of virtual servers on it for free. I'ts possible Amazon is using this licencing clause to their advantage.
Think about how they make money from Android while competiting with Windows Phone. They're clever enough to realise you can make money from multiple angles.
If they are, I'd see it the same way as funding the node port. The less temptations for developers to build on *nix, the better for Microsoft. Using Windows on EC2 means they're still using Windows, after all.
I'd imagine they'd prioritize lowering the barrier of creating .net webapps over Azure success. Azure is just one service but .net success is key part of their strategy.
The Tegra 3 is functionally a quad-core processor, but includes a fifth "companion" core. All cores are Cortex-A9s, but the companion core is manufactured with a special low power silicon process. This means it uses less power at low clock rates, but more at higher rates; hence it is limited to 500 MHz. There is also special logic to allow running state to be quickly transferred between the companion core and one of the normal cores. The goal is for a mobile phone or tablet to be able to power down all the normal cores and run on only the companion core, using comparatively little power, during standby mode or when otherwise using little CPU. According to Nvidia, this includes playing music or even video content.
To me this is the interesting feature, not the "more power than before", which was to be expected. Ironically, none of it is presented in the article nor videos.
I also find the "throw more power at it to get it smooth" argument quite appalling.
I think giving a brief explanation is useful. Maybe "Sorry! We can't generate images that large!" or something. That explains what's going on without going too deep on the max dimensions and whatnot.
Academia, of course, gives you entirely different salaries. For an Assistant Prof. job at a good university you are looking at ~65,000 +/- 10k depending on location. This of course is supposed to be for 9 months, but good luck telling your tenure review committee you didn't do any work during the summers. Word is lucrative part-time consulting gigs can be had if you are enterprising. It looks as if you are at about 1/2 of what an industrial lab will pay. But then if you are in the academy, supposedly you care more about freedom than money.
I've heard of offers as low as $48K for assistant professors in biological sciences. That's at a real PhD-granting research university, though admittedly not a particularly good one. Even lower for fields like English and History. Higher for hard sciences, and even higher for CS.
The CRA Taulbee Survey (http://www.cra.org/statistics/) has detailed info on CS faculty salaries in the US and Canada. 2005-2006 9-month salaries for a new PhD in CS with a tenure track position ranged from $70k to $99k with $82k median. Of course, you have to get one of those positions first. PhD production has been at record levels in North America for a few years now. It seems lots of people decided to go to grad school after the bubble burst. The survey has incredible detail including, for instance, how many positions opened up as a result of people dying!
