E L I T E
C H A N N E L PA R T N E R
simulation@rand.com | RandSIM.com | 888.483.0674
Test 1 - External Flow Over Passenger Vehicle
This is our first test with a simple car in tunnel simulation. The GPU was equivalent to about 25 CPU cores in this test.
In the graph on the right, we see a comparison of the speed for 100 iterations for various hardware. We see on
the far left that the CPU 32 SP, or single precision (32 cores), took less than one minute to perform 100 iterations.
On the other end of the graph, we see that CPU 4 DP, or double precision (4 cores)—a configuration many clients
today are utilizing—took more than four minutes. In the middle, we see 12 cores DP was about 2.5x faster than 32.
Single precision was slightly slower than the 32 CPU run, and the GPU 32 DP was about double the simulation
time. That is typical with a GPU solver.
In the graph on the left, we see the results of the vehicle drag simulations for the 1,000 iterations ran.
• The GPU solves, represented by the dotted lines, show minor variance between the single and
double precision runs in terms of accuracy. Since the single precision run is about twice as fast, we
recommend using the GPU solver in single precision.
• The CPU 32 DP Coupled Solver, represented by the green line, shows a completely different behavior.
It took significantly longer to run an iteration as a coupled solver, but fewer iterations were needed; it
was about 3.2 times slower but required 4 times fewer iterations.
Does Ansys Fluent with a Native GPU Solver Live Up to the Hype?
