I'm bored teach me about something you like

The video games you play almost certainly use GPUs or graphics processing units to generate the graphics but did you know they can also be used for hardcore scientific computation?

each GPU is bit like a CPU (standard computer processor) linked to thousands of ALUs or arithmetic logic units. The ALU is the CPUs calculator and by having so many the gpu can do many calculations at once. But it can only run one program at once which means every ALU has to perform the same calculation just on different numbers. The programmes GPUs run can’t do conditional jumps like a cpu can.

a cpu can check if two numbers are the same and if they are go to some other part of the program to do something difrent. But in a GPU some ALU would have to jump and some not. That’s not allowed, there is only one program running on the gpu so you can’t be in two different parts of the program at the same time.

this makes writing good code for GPUs hard. Instead of conditional jumps you have conditional saves. Where an output gets saved only if a condition is satisfied.

so if in a GPU program you want each ALU to calculate 1 of 2 alternative formulas every ALU has to compute both formulas and save the answer from the one you wanted in each case. 
avoiding these sections where you have to do redundant work is a big part of good GPU programing.

The video games you play almost certainly use GPUs or graphics processing units to generate the graphics but did you know they can also be used for hardcore scientific computation?

each GPU is bit like a CPU (standard computer processor) linked to thousands of ALUs or arithmetic logic units. The ALU is the CPUs calculator and by having so many the gpu can do many calculations at once. But it can only run one program at once which means every ALU has to perform the same calculation just on different numbers. The programmes GPUs run can’t do conditional jumps like a cpu can.

a cpu can check if two numbers are the same and if they are go to some other part of the program to do something difrent. But in a GPU some ALU would have to jump and some not. That’s not allowed, there is only one program running on the gpu so you can’t be in two different parts of the program at the same time.

this makes writing good code for GPUs hard. Instead of conditional jumps you have conditional saves. Where an output gets saved only if a condition is satisfied.

so if in a GPU program you want each ALU to calculate 1 of 2 alternative formulas every ALU has to compute both formulas and save the answer from the one you wanted in each case. 
avoiding these sections where you have to do redundant work is a big part of good GPU programing.

isnt this now dealt with thread based software where the different cores of processors process threads at same time