PotPlayer 8K playback benchmark: what GPU decoding actually buys you
PotPlayer can open an 8K file on almost any modern machine. Whether it plays that file at 60 frames per second is a different question, and the answer depends almost entirely on which decoder you let the player use. We took one 8K60 HEVC file, ran it through three decode paths in the current PotPlayer build, and logged CPU load, GPU utilisation, dropped frames and package power for ten minutes each.
The test file and the three paths
The source is a 12-minute 8K60 HEVC clip at 180 Mbit/s, 10-bit, in an MKV container with a single TrueHD track and no subtitles. It is deliberately hostile: high bitrate, heavy grain, and a scene at 04:12 where the camera pans across falling snow, which is the worst case for any motion-compensated decoder.
Path one is pure software decoding with all hardware options off. Path two uses DXVA2 with the default renderer. Path three uses D3D11 hardware decoding with the D3D11 renderer and HDR passthrough disabled, which is what most people will end up with after a clean install on Windows 11.
The machine is a Ryzen 5 5600 with 32 GB of DDR4-3200 and a GeForce RTX 3060 12 GB, driving a 4K60 panel. Windows 11 23H2, driver 566.36, PotPlayer 1.7.22769 — one build behind the current release, because that is what the machine was already running and the decode path has not changed since.
Every run starts from a cold player launch. We log with PresentMon at 250 ms intervals and take package power from HWiNFO64. Temperatures stayed under 68 °C on the CPU and 62 °C on the GPU, so no thermal throttling contaminates the numbers.
The numbers: 12 percent versus 78 percent
Software decoding averaged 78 percent total CPU load with all 12 threads engaged, peaked at 96 percent during the snow pan, and dropped 214 frames out of 9,000 — a 2.4 percent loss that shows up as visible stutter roughly every 40 seconds. Average frame time was 21.4 ms against a 16.7 ms budget.
DXVA2 averaged 27 percent CPU and 34 percent GPU, dropped 41 frames, and held 16.8 ms average frame time. D3D11 averaged 12 percent CPU and 41 percent GPU, dropped three frames, and held 16.71 ms. Package power told the same story from the other direction: 96 W on software, 74 W on DXVA2, 68 W on D3D11.
“The surprise is not that hardware decoding is faster,” said Tomas Berg, who ran the capture rig for this test. “It is that the gap between DXVA2 and D3D11 is still this visible on an 8K file. Forty-one dropped frames versus three is the difference between a clean pan and a hitch you notice.”
The GPU never came close to saturation. At 41 percent utilisation the RTX 3060 had headroom for a second 8K stream, which we confirmed by opening a 4K60 file in a second window: combined GPU load reached 63 percent and neither window dropped a frame.
What this means on slower hardware
We repeated the D3D11 run on a laptop with an 8th-generation Core i5 and integrated UHD 620 graphics. It played the same file at 8K but only at 41 fps average, dropping 1,880 frames; switching the renderer to D3D11 and enabling "wait for V-Sync" off brought that to 52 fps and 640 dropped frames. On that machine, 4K60 of the same clip was flawless.
“We tell people to try 8K only after 4K is clean,” said Aya Nakamura, who handles playback support for the build team. “Ninety percent of the stutter reports we get are a 4K file on a 1080p panel with the wrong renderer selected, not an 8K file on weak hardware.”
The practical settings are short. Enable hardware decoding for HEVC and AV1, set the video renderer to D3D11, keep the frame buffer at the default, and leave HDR tone mapping on automatic unless you have measured your panel. The statistics overlay, bound to Tab by default, shows dropped frames live, which is faster than guessing.
One caveat on the numbers above: they come from one file on one machine. A different 8K encode with a lower bitrate will behave better on software decoding, and a Ryzen 7 or a newer GPU will flatten the difference between DXVA2 and D3D11. The direction of the result, however, has been stable across every rig we have tested since the 1.7.22 line.
Switched from DXVA2 to D3D11 after reading this and the snow pan in my own copy stopped hitching. Same GPU, same driver. Did not expect that.
Good to see package power included. My laptop runs about 8 degrees cooler with hardware decoding on, which matters more to me than frame counts.
The UHD 620 section is realistic for once. Most benchmarks pretend everyone has a 4090. 52 fps at 8K on integrated graphics is honestly impressive.
Tab overlay tip saved me a lot of guessing. I had 300 dropped frames from a bad HDMI cable and assumed it was the player.
Would like to see AV1 in the same table. My 8K AV1 files behave differently from HEVC on a 3060, usually lower GPU load.
