Can AMD graphics cards such as RX 580 with 8 G VRAM be useful for FLUENT software?
What do you think about using NVIDIA graphics such as GTX 1650 cards in the same price range with less VRAM ( 4G VRAM) ?
Julian Vogel : ANSYS Fluent has changed the default in Release 2019.R3 from Segregated Solver to Coupled Solver. In my honest opinion, the Coupled Solver should be used for most application cases anyway. Only if computational speed in transient simulations is crucial, and if for that physics the inner iterations in a timestep are fast converging (e.g. because the timesteps are incredibly small anyway), than the Segregated Solver might have some benefit. An example are LES-like, SRS simulations with small timesteps, where the Coupled Solver would have a remarkable overhead.
The benefit of including GPU's in computations pays off, if the simulation is limited to a very limited number of nodes/workstations which have installed graphics cards, or if there are several graphics cards installed inside one workstation. If we talk about distributed parallel computing over many nodes/workstations, than the external communication is usually the bottleneck and the accelaration of the AMG solver on GPU is no longer of so large benefit.
Mohammad Reza Saffarian : The AMD graphics cards you mention are rather new and probably they are not supporting the NVIDIA's CUDA programming model. But CUDA is almost exclusively used so far for GPU programming. ANSYS Fluent has followed the CUDA programming model as well and I don't think that for the moment they are supporting other vendors graphics cards.
Regards,
Dr. Th. Frank.
yes but only for graphics display or 3D animated graphics
graphics cards can't affect the speed of computing abilities this is the work of processor
As you know, the “Compute Unified Device Architecture” (CUDA) is used as a parallel computing platform and programming model for Graphics Processing Units (GPUs), which reduces the complexity of programming.
Indeed, GPUs are stream processors – processors that can operate in parallel by running one kernel on many records in a stream at once (see GPGPU).
With the GPU acceleration, the code on multicore CPUs, so that currently the CPU+GPU code is about 20% faster than the pure CPU code on CFD applications.
It should be noted that the "CUDA" is only available for "Nvidia" graphic cards (of course, for now!)
Nvidia and Ansys developed a library to speed up the AMG solver by using the GPU. However, this is only used with the coupled solver at the moment, where it can be beneficial. It is not so much beneficial with the segregated solver, so it is not used by default. Also, some radiation models can benefit from the GPU. This information is from the nvidia site:
https://www.nvidia.com/en-us/data-center/gpu-accelerated-applications/ansys-fluent/
In my opinion, for many cases, you still can't make use of the GPU in Fluent.
As far as I know, GPU acceleration is available for a limited number of physical models in Fluent, mainly single phase flows.
Thanks Julian Vogel for the link. I confirm that for DO radiation modelling benefits can be obtained only for refined angular discretization.
Julian Vogel : ANSYS Fluent has changed the default in Release 2019.R3 from Segregated Solver to Coupled Solver. In my honest opinion, the Coupled Solver should be used for most application cases anyway. Only if computational speed in transient simulations is crucial, and if for that physics the inner iterations in a timestep are fast converging (e.g. because the timesteps are incredibly small anyway), than the Segregated Solver might have some benefit. An example are LES-like, SRS simulations with small timesteps, where the Coupled Solver would have a remarkable overhead.
The benefit of including GPU's in computations pays off, if the simulation is limited to a very limited number of nodes/workstations which have installed graphics cards, or if there are several graphics cards installed inside one workstation. If we talk about distributed parallel computing over many nodes/workstations, than the external communication is usually the bottleneck and the accelaration of the AMG solver on GPU is no longer of so large benefit.
Mohammad Reza Saffarian : The AMD graphics cards you mention are rather new and probably they are not supporting the NVIDIA's CUDA programming model. But CUDA is almost exclusively used so far for GPU programming. ANSYS Fluent has followed the CUDA programming model as well and I don't think that for the moment they are supporting other vendors graphics cards.
Regards,
Dr. Th. Frank.
- Badges
- Science topic