Intel's Open-Source SVT-AV1 Video Encoder Ends May With Another Performance Boost
It's been very fascinating to watch the speed improvements of Intel's SVT-AV1 open-source AV1 video encoder since in February when being made aware of Intel's new SVT video projects. The SVT-AV1 project is ending out May with another step-up in performance for what is already one of the fastest CPU-based AV1 video encoders.
It was just in mid-May that SVT-AV1 0.5 was released while since then the Intel open-source developers have remained busy working on more improvements.
While looking at my daily LinuxBenchmarking.com performance tracker for these video encoders, I noticed this week was marked by yet another improvement.
Indeed when checking out their Git repository between 27 and 28 May for when my Phoronix Test Suite / Phoromatic powered benchmark tracker noticed the change, there's this commit noting performance changes as a result of pre-set tuning, but there's also been a lot of other commits too in recent days (due to power/thermal constraints, I don't have this tracker set to benchmark on a per-commit basis but rather just a timed per-day manner). As for the missing systems recently, those systems have been temporarily re-assigned to some other benchmarks but I should have them back doing daily video encoder benchmarks in the days ahead... With the video encoder test result page being ad-free and otherwise unfunded, it's mainly driven out of my own curiosity and testing PTS/Phoromatic/OpenBenchmarking.org software functionality, so for now, pardon the brief periods when a subset of the systems may be unavailable for one reason or another.
It's certainly a wild ride if expanding the daily coverage back to early March when I began running these daily benchmarks... I was motivated to start the daily tests after seeing the course of improvements over February, which were already measurable back then compared to when the SVT-AV1 code first became public. This AV1 encoder keeps getting much faster though over this time Intel's SVT-HEVC and SVT-VP9 encoders have remained largely flat.
As for its intended use, back in April when Intel formally announced the Scalable Video Technology work, it was also mentioned Netflix is involved in this effort as well and presumably will be powering their future AV1 streams. Likewise it will be interesting to see who else in the industry decides to make use of these SVT encoders in production, which are optimized for the likes of 2nd Gen Xeon Scalable (Cascade Lake) processors. Interestingly, SVT-AV1 performs even better when running on Clear Linux compared to say Ubuntu (as for that behavior, I'm told it likely is due to Clear's power management policies and other factors over say just the compiler flags). It will certainly be interesting to see where the SVT-AV1 performance is at by year's end or even in the next month or two.
It was just in mid-May that SVT-AV1 0.5 was released while since then the Intel open-source developers have remained busy working on more improvements.
While looking at my daily LinuxBenchmarking.com performance tracker for these video encoders, I noticed this week was marked by yet another improvement.
Indeed when checking out their Git repository between 27 and 28 May for when my Phoronix Test Suite / Phoromatic powered benchmark tracker noticed the change, there's this commit noting performance changes as a result of pre-set tuning, but there's also been a lot of other commits too in recent days (due to power/thermal constraints, I don't have this tracker set to benchmark on a per-commit basis but rather just a timed per-day manner). As for the missing systems recently, those systems have been temporarily re-assigned to some other benchmarks but I should have them back doing daily video encoder benchmarks in the days ahead... With the video encoder test result page being ad-free and otherwise unfunded, it's mainly driven out of my own curiosity and testing PTS/Phoromatic/OpenBenchmarking.org software functionality, so for now, pardon the brief periods when a subset of the systems may be unavailable for one reason or another.
It's certainly a wild ride if expanding the daily coverage back to early March when I began running these daily benchmarks... I was motivated to start the daily tests after seeing the course of improvements over February, which were already measurable back then compared to when the SVT-AV1 code first became public. This AV1 encoder keeps getting much faster though over this time Intel's SVT-HEVC and SVT-VP9 encoders have remained largely flat.
As for its intended use, back in April when Intel formally announced the Scalable Video Technology work, it was also mentioned Netflix is involved in this effort as well and presumably will be powering their future AV1 streams. Likewise it will be interesting to see who else in the industry decides to make use of these SVT encoders in production, which are optimized for the likes of 2nd Gen Xeon Scalable (Cascade Lake) processors. Interestingly, SVT-AV1 performs even better when running on Clear Linux compared to say Ubuntu (as for that behavior, I'm told it likely is due to Clear's power management policies and other factors over say just the compiler flags). It will certainly be interesting to see where the SVT-AV1 performance is at by year's end or even in the next month or two.
7 Comments