Graphic cards and super computing

Post Reply New Topic

I wonder when scientist started using Graphic Cards for scientific calculation.

I read an article in a popular science magazine that one scientist came up with the idea after seeing her son happily playing in playstation, thought that as a game console or computer can simulate the of the the physics in shooters or sports games with such accuracy of, it can also simulate chemical, physical phenomena, and more.

Not until nVidia released a GPU. From there, it's blatantly obvious how powerful they are for performing the same algorithm on massively multiple data. I'm sure this was realized before the design even left the brainstorming phase.

The Stanford Folding@Home team would have the best answer for that... SETI@Home (I believe) and Bitcoin Mining also use the immense computational power of modern video cards to significantly increase the computation power. (Although at much higher wattage than a CPU will.)

nVidia has also taken it a step further in the last couple years to allow 3rd-party developers the ability to access the video card's power to improve the software with CUDA. (I believe AMD has a version of this as well but it's at a much smaller scale and not used as often or as effectively.) Some programs like Photoshop benefit from a fast video card's CUDA cores like extra processors. (Many diddlyboobs think more video RAM equals more speed - it doesn't, it's all the amount and speed of CUDA cores.)

Results certainly vary.

GPUs are very powerful for performing the same operations on lots of data in parallel, but the basic architecture they use that makes them that way was used in supercomputers long before the modern GPU appeared. They didn't invent that approach for GPUs and then apply it to scientific computing... it was actually the other way around.

there is alsoa similar API for AMD products.(as you noted) the scale is for any intents and purposes the same, nvidia is a bit easier to set up but amd allows for asysmetric multi gpu use(you have a 2 year old amd card and one of their newest beasts, well the old and new can function in crossfire together, plug and play)

as far as i understand the API uses the same pipelines used for physics calucaltions in games as well, where nvidia probably has more work put into the cuda core design.

i set up a system for a friend who is educating herself as a molecular biologist, she wanted to speed up her simulations to an acceptable level when she did them at home,
using 2 older amd cards and a new one actually gave her quite the boost, she went from 6-10 hours of simulation down to about 2-3.

CUDA is great very few knows about this that the word "CUDA" means "MIRACLES" in Polish

I lost my former NVIDIA GeForce 9800 GT because i allowed SETI@home use my GPU, at least i think is the main reason. fortunately i get new one NVIDIA GeForce GT 440 as replacement because i still have warranty the last two month
Now not using SETI@home regularly update video drivers and all working fine