Synthetic and Rendering Benchmarks (CPU)
Cinebench R20
Firestrike Physics
Vega 8 iGPU
RTX 2060 dGPU
Time Spy Physics
Vega 8 iGPU
RTX 2060 dGPU
Feel free to peruse all these charts, but I imagine most of you do not want to go through all 24, so here’s some of the takeaways.
Firstly, higher power profiles mean not only higher performance but also more retention for performance. Cinebench R20 multi threaded is an interesting test because it shows both the silent and performance profiles drop clock speed on the second run while the turbo profile actually only drops on the third run. Firestrike and Time Spy do not reflect this but I believe that is because the physics portion of the test is the last in the sequence, meaning the CPU is already worn out by the time it has to do the heavy lifting. It is nice to see that single threaded performance retention is equal between all profiles, too.
Secondly, there seems to be some strange behavior on Firestrike depending on the GPU configuration. The score is actually lower when using the Vega 8 iGPU instead of the RTX 2060 dGPU even though this should be a CPU bottlenecked test; perhaps the GPU has a greater role in the benchmark than I previously thought. Additionally, when using the 2060, performance actually went up run to run, not down, on the silent and performance profiles. I think this might be down to power management but I’m not quite sure what is changing between the first and second runs to allow the CPU to boost higher.
What surprised me the most is that the turbo profile actually had a noticeable uplift over the performance profile, though some of this might be due to the Firestrike and Time Spy physics tests being somewhat dependent on the GPU. Still, in the totally CPU bound Cinebench R20 multi threaded benchmark, it is clear that the turbo profile does have advantages for performance retention.
Synthetic and Rendering Benchmarks (GPU)
Firestrike Graphics
Vega 8 iGPU
RTX 2060 dGPU
Time Spy Graphics
Vega 8 iGPU
RTX 2060 dGPU
We already saw the ramifications of CPU performance and performance retention depending on the profile, but it’s more severe for the discrete GPU, the RTX 2060. In Firestrike, the silent profile actually does just fine on the first and second runs, but the third run reveals a large performance drop. In Time Spy, the 2060 was clearly throttling within the first run since performance levelled out in just the second run. There doesn’t seem to be a performance retention difference between performance and turbo, but turbo is a little faster.
One interesting thing is that the Vega 8 iGPU never throttled, indicating that the power profiles don’t actually impact the iGPU’s performance. Furthermore, when we look at third runs, the 2060 is only a hair faster than the Vega 8 iGPU in Firestrike and just ~40% faster in Time Spy. This really speaks to the performance AMD has delivered with mobile Vega, which should be great for titles that aren’t particularly demanding.
Synthetic and Rendering Benchmarks (Combined)
Firestrike Combined
Vega 8 iGPU
RTX 2060 dGPU
Firestrike Overall
Vega 8 iGPU
RTX 2060 dGPU
Time Spy Overall
Vega 8 iGPU
RTX 2060 dGPU
PCMark 10 Express
Vega 8 iGPU
RTX 2060 dGPU
The combined CPU and GPU tests really reveal the performance retention advantages of the performance and turbo profiles over the silent profile. Although the silent profile appears to be close in performance to the other two profiles, running another benchmark without waiting demonstrates how the performance of the silent profile drops. In both Firestrike and Time Spy, performance drops by more than half if you’re using the RTX 2060 dGPU. When using the Vega 8 iGPU, there really isn’t a performance retention issue on the silent profile, which is nice.