I am thinking of assembling this system:
- AMD CPU(A8-3870 APU which has Radeon HD 6550D inside: 400 stream processors:xxx GFLOPS) nearly 110$
- AMD Graphics card: HD 7750 (512 stream processors:819 GFLOPS peak performance) nearly 170$
- an appropriate ram(1600MHz bus) and mainboard
Can i achieve 819+xxx GFLOPS peak-performance mentioned in official sites with using OpenCL and similar programs?
Can i use all 912 cores with OpenCL/Jocl and is it important to add cpu-cores to the pot(4 of them(of course 2 of them will be used for feeding gpu))?
C++ or Java, which one has the most yielding libraries for using multiple gpu’s or apu’s present on computer?
What happens if i cancel both apu and gpu and buy a single Nvidia GTX-660? This wins?(229$ -1800GFLOPS)(with a simple 4-core cpu of cheapest without apu)
I am not trying to do a VS question. I need to know what could be better for scientific computing(%75 of the time) and gaming(%25 of the time) because i have a low budget. With “scientific calculations” i mean fluid dynamics+solidstate physics simulating. With games i mean those have openCL and PhysX.
Can you give a very very minimal simple example of OpenCL code using multiple GPUs ?
Thank you.
This is the peak performance. One definition of peak perform is; A manufacturers guarantee not exceed this rating.
You can achieve this number most likely, but not doing something useful. What you can achieve for your specific requirement depends greatly on what it is. You might expect to get 0.1% to 10% of this value in reality.
I would use whatever you are most comfortable with. You can call the GPU from either, but the language you use is C-like so it doesn’t matter what the “host” language is.
Impossible to say, but there is a good chance whatever you choose will be okay.
There are lots of example on the web, but you really need to focus on what you will be using the system for.