VVenC H.266 Encoder Rolls Out More ARM Optimizations For Nice Performance Gains
Fraunhofer HHI this week released a new version of VVenC, their open-source H.266 video encoder. Among the changes this release are more performance optimizations for ARM and I have run some comparison benchmarks using a NVIDIA GB10 SoC with the Dell Pro Max GB10.
This isn't the first release of VVenc to feature targeted ARM performance optimizations but simply the latest in continuing to strive for better ARM64 video encode performance. The VVenC 1.14 release brings "many" ARM SIMD optimizations around NEON and SVE as well as enabling Scalable Vector Extensions (SVE) by default for capable ARM processors.
The NVIDIA GB10 with its ten Arm Cortex-X925 and ten Arm Cortex-A725 cores do support SVE/SVE2. Thus a very interesting test target and thanks to Dell having provided the Dell Pro Max GB10 review sample it's possible to benchmark these latest VVenC improvements.
For today's testing I built and benchmarked the VVenC 1.13 and VVenC 1.14 H.266 encoders on the Dell Pro Max GB10 using the same software stack and default compiler/flags with simply comparing the version of VVenC being tested.
Around 35% higher performance for VVenC 1.14 with 4K content at the fast preset. And that's on top of VVenC 1.13 already having some ARM optimizations and the like - this isn't their first release trying to tune for AArch64 performance.
The "faster" preset with 4K content was still yielding around 7% better performance for this new release.
Even for 1080p content some nice VVenC 1.14 performance gains for H.266 video encoding on the Dell Pro Max GB10.
More VVenC benchmarks to come for those curious how the performance stacks up between AArch64 and x86_64 processors. Those wanting to try out this open-source H.266 encoder can find it on GitHub.
This isn't the first release of VVenc to feature targeted ARM performance optimizations but simply the latest in continuing to strive for better ARM64 video encode performance. The VVenC 1.14 release brings "many" ARM SIMD optimizations around NEON and SVE as well as enabling Scalable Vector Extensions (SVE) by default for capable ARM processors.
👁 NVIDIA GB10 with ARM SVE and SVE2
The NVIDIA GB10 with its ten Arm Cortex-X925 and ten Arm Cortex-A725 cores do support SVE/SVE2. Thus a very interesting test target and thanks to Dell having provided the Dell Pro Max GB10 review sample it's possible to benchmark these latest VVenC improvements.
For today's testing I built and benchmarked the VVenC 1.13 and VVenC 1.14 H.266 encoders on the Dell Pro Max GB10 using the same software stack and default compiler/flags with simply comparing the version of VVenC being tested.
Around 35% higher performance for VVenC 1.14 with 4K content at the fast preset. And that's on top of VVenC 1.13 already having some ARM optimizations and the like - this isn't their first release trying to tune for AArch64 performance.
The "faster" preset with 4K content was still yielding around 7% better performance for this new release.
Even for 1080p content some nice VVenC 1.14 performance gains for H.266 video encoding on the Dell Pro Max GB10.
More VVenC benchmarks to come for those curious how the performance stacks up between AArch64 and x86_64 processors. Those wanting to try out this open-source H.266 encoder can find it on GitHub.