Ansys Fluent GPU Speed Test - Full Demo
Hi everybody, this is Mingyao from Ozen Engineering, Inc. In this video, I'm going to demonstrate the amazing performance of the new SSFluent GPU solver.
Demonstration Setup
I'll be opening up two instances of Fluent. I have several models prepared, and we'll start by running a simulation on 12 cores of a model that contains approximately 2.5 million cells. This is a simple model of flow over a cylinder.
- Model: Flow over a cylinder
- Cell Count: Approximately 2.5 million
- Mesh: Refined region in the middle with a boundary layer
I'm using a remote desktop for this demonstration, and I want to thank our partners at Dell and NVIDIA for providing the system I'm testing on, which features a great NVIDIA processor.
Performance Testing
We will compare the performance of the CPU and GPU solvers. The GPU solver utilizes an RTX A2000 with 12 GB of RAM and 26 processing units. I'll display the mesh to confirm that it's the same file for both solvers.
For the test:
- Run 200 iterations on both CPU and GPU solvers.
- Use a simple algorithm for both solvers due to GPU solver limitations.
The CPU solver will be given a head start. We'll initialize the case and start the calculations.
System Specifications
- CPU: 32-core AMD CPU
- GPU: 12 GB RTX A2000 from NVIDIA
The CPU is fully utilized, while the GPU operates at 64% utilization. The speedup is remarkable, with the GPU completing 200 iterations before the CPU finishes 5 iterations. This demonstrates a significant improvement, thanks to the collaboration between the ANSYS development team and NVIDIA.
Results and Observations
The GPU solver achieves approximately 2 iterations per second on a 2.5 million cell mesh. This is a great speedup for the physics that the GPU solver supports.
Let's take a look at the results:
- Contour display on two symmetry planes showing vortex shedding off the 3D cylinder.
- Velocity profile visualization with contour banding options.
This is a fantastic achievement with extremely fast processing speeds. The plan is to add more capabilities and features to ANSYS Fluent to further leverage GPU solutions across different physics.
Conclusion
Thank you for watching this quick video. I'm excited about this development. If you have any questions, feel free to contact us at info@ozeninc.com.
If you enjoyed this video, please subscribe to our channel or give us a thumbs up. Check out our other videos as well. Thanks and have a great day!
Hi everybody, this is Mingyao from OZEN Engineering and in this video I'm going to go through the amazing performance of the new SSFluent GPU solver. So I'm going to do a quick demonstration here and I'm going to open up two instances of Fluent.
I have a number of models set here and let's go with, let's start it out by running a simulation on let's say 12 cores of a model that's around two and a half million cells. Okay this is a simple model of a flow over a cylinder. Take a look, that's our cylinder. I can show the mesh as well.
To show you that for 12 cells we have a refined region right in the middle with a boundary layer and everything. So we're going to put this on to one side of our screen. I'm using a remote desktop here. This video is made possible thanks to our partners and friends at Dell and NVIDIA.
They provided this system I'm testing out on. It's a great system with a great NVIDIA processor. Okay we're going to go ahead and run this and get started. Okay so now the first thing I'm going to do is show you the performance and the performance of this system.
Okay so let's start with the results. So we're going to go ahead and start this and let's see how it works. Okay so I'm going to go ahead and start this and let's see how it works. Okay so this is the result. So we're going to go ahead and start this and let's see how it works.
So here we're going to go to our native GPU solver. You can see I'm using a RTX A2000 with 12 GB of RAM and 26 units for processing. We're actually going to read a different one. I don't want to read the same model, so I make copies of these. Let's open up this copy of that.
So here we're using the native GPU solver. I'm going to display the mesh just to show that it looks similar, to prove that it is the same file. And we're going to put it on the right hand side of the screen. There we go. Okay, so I'm going to let the results speak for themselves.
And I'm going to give the CPU solver a bit of a head start. We're going to run 200 iterations here. I should probably double check. One of the limitations of the GPU solver is that it is limited to a simple algorithm. So we're going to make that the same in both cases.
And I'm going to get rid of this so you can kind of see the speed of the simulation. Let's go to solution, run calculations, start calculating. I'm going to give the CPU solver a little bit of a head start here. We're going to initialize the case here. But really we're going to do the same thing.
So run calculation, 200 iterations. And we're off to the races. So you can see the GPU solver started already. And CPU here we're waiting for the initialization to finish. So we've got the CPU fight out and we're going to run this complex calculation. So number of iterations is on this side.
This is the use case against the CPU region. And it's the monumental mark. We're going to goane straight to here. So that's what we're going to call this operation on the CPU commercial viewer. So we're going to take ArchRTA olmuş, which allows us to see which passed to what's far out there.
That's been pretty filing. So what's default. So again, this model has about 2.5 million nodes in the model. It's fully hex mesh. And let's take a look at the system specs we have here. So the system we're using, this is again courtesy of our friends at Dell and NVIDIA. We have a 32-core AMD CPU.
And our graphics card is a 12 GB RTX A2000 from NVIDIA. And you can see the CPU is fully cranked up. My GPU is working at 64% of utilization. The speedup is amazing. So we're going to get to probably 200 iterations before my CPU simulation finishes 5 or so. Maybe 3. There you go.
All done on the GPU and CPU is just getting ramped up. This is really a phenomenal improvement. Amazing job by the ANSYS development team working with NVIDIA to produce such a fast solver.
You can see that we're able to crank through about 2 iterations per second on a 2.5 million cell mesh in a very economically affordable GPU processor. So for the physics that the GPU is doing, it's going to be a lot faster. But the GPU solver supports. This is a great speedup.
Let's take a look at some results here. So let's do a contour. We'll do a new contour on the two symmetry planes. Let's display that. Okay. You can see the vortex. It's going to be a lot faster. So let's do a little bit of vortex shedding off of our 3D cylinder here.
We can look at the velocity profile. You can see the... We can do contour banded if we want to. So really fantastic achievement. Super fast speed. It's extremely fast. We can see the 12 cores.
As we go, the plan is to add a lot more capability and features into Ansys Fluent to even better take advantage of GPU solutions in all of the different physics. That's Fluent support. So thank you very much for watching this quick video. I'm super excited by this development.
And if you have any questions, you're welcome to contact us at OZInk.com. If you like videos like this, please subscribe to our channel or give us a thumbs up. And you can check out our other videos as well. So if you have any questions, you're welcome to contact us at OZInk.com.
If you like videos like this, please subscribe to our channel or give us a thumbs up for this video. Thanks and have a great day!
