AVC Labs AI video 1280x720 MP4 to 1980x1080

The provided documents offer a comprehensive look into an **AMD Ryzen system's configuration and its performance during AMD Amuse video-to-video locomotion model test runs**, along with detailed hardware diagnostics.

**System Configuration and Software Environment:**

* **Processor:** The system is powered by an **AMD Ryzen 7 255 model with Radeon 780M Graphics**, featuring 8 cores and 16 threads. It operates at a core speed of 2814.5 MHz and has a Thermal Design Power (TDP) limit of 54.0 Watts.
* **Memory:** It is equipped with a significant **96 GB of fast DDR5 RAM**, configured across 4 x 32-bit channels, with a memory frequency of 2786.1 MHz (effective 5572.2 MHz, max bandwidth DDR5-5600). The RAM modules are manufactured by Micron Technology.
* **Storage:** The system includes a **0.50 TB fast M.2 NVMe SSD** (specifically a CT480E100SSD8 model with 447.1 GB capacity), which uses a PCI-E 4x link at 16.0 GT/s.
* **Power Supply:** The system's power limits are noted due to a 120-watt power supply, with the CPU having a maximum power draw of 53 watts.
* **Operating System & Software:** The system runs **Windows 11 Home (x64), Version 24H2**, with Developer mode active, SUDO enabled, and relevant software like SDK, AMD Adrenalin, WSL, Python V3.13, and Miniconda installed.
* **Performance Tweaks:** While there are no official AMD Adrenalin overclock settings or AMD Ryzen Master series available for this specific model, **performance can be boosted through BIOS tweaks**.
* **Background Activity:** The test runs indicate that **background desktop multi-tasking impacts results**.

**AMD Amuse Test Run Details:**

The AMD Amuse software leverages two engines:
* **AMD Engine 0** handles prompt processing.
* **AMD Engine 1** is responsible for preparing the pipeline and input frames.

Each test run in the provided log follows a consistent structure, detailing:
* The Amuse model used (e.g., Amuse 3.0.7 Video-to-video Locomotion model, Locomotion ToonYou, Locomotion Samaritan3D, Realistic Vision, CyberRealistic).
* Whether a prompt was used.
* Clip title, dimensions, codec, and quality settings.
* AMD Amuse HQ setting (High Quality, either On or Off).
* Locomotion output resolution.
* Input and output Frame FPS.
* Number of clip frames.
* **AMD Amuse Guidance Scale**: Generally set to 2, indicating a **"low guide to prompt"**.
* **Strength**: Typically set to 0.50, representing a **"mix between AI and original video"**. Other values like 0.20, 0.30, 0.40, and 0.70 were also tested.
* **Overlap**: Predominantly set to 7, meaning **"High overlap less background AI and more action AI"**. Overlap 5 and 6 were also used in some tests.
* **Inference Steps**: Default is 8 unless otherwise specified. Steps varied significantly in tests, from 11 to 9000 steps.

**Key Test Results and Observations:**

* **ToonYou Model:**
* Generally yielded **high-quality results**, with several tests achieving 9/10, 9.5/10, and even **10/10 for quality**.
* Showed good performance in background and people detail, often scoring 8.5/10 to 9.6/10.
* Longer clips and higher frame counts significantly increased duration and compute time. For example, a 6-minute "Lost in Space" clip (9000 frames) with 9000 steps led to the **RAM peaking at 50GB**. A shorter 5026-frame clip also peaked at **39GB RAM Use**.
* Some runs were cancelled due to "excessive time".
* **CyberRealistic Model:**
* Achieved **excellent quality (9/10 for Quality, Background, and People)** in several tests, particularly with a Strength of 0.20 and 11 inference steps.
* Had relatively shorter durations and compute times compared to ToonYou for similar clip lengths and frames.
* One test run "failed" without specific quality or performance metrics.
* **Realistic Vision Model:**
* Results varied widely. One test showed **poor quality (2/10, hopeless background and people)** with Strength 0.50 and 102 steps.
* Improved significantly to **9/10 quality** with lower Strength settings (0.20 and 0.30) and fewer steps (34 or 154).
* Attempting "ghibli style" with this model resulted in very low quality for background and people (0/10 and 2/10 respectively).
* **Locomotion Samaritan3D Model:**
* Produced "unusable" quality (7/10) in one test without a prompt and 102 steps.
* With prompts like "cartoon," quality improved to "good" (7/10), with "very good" background results.
* **General Performance Metrics (from GPU-Z and Task Manager screenshots):**
* During active use, the **GPU Load was consistently at 99%**, indicating high utilization.
* **GPU Power Draw ranged from 39.0 to 41.0 W**.
* **Dedicated GPU Memory Used was 1935 MB**, and Dynamic Memory Used was 5652 MB. The GPU memory usage on the Amuse application itself showed **1.9/2.0 GB used**, suggesting near-full dedicated VRAM utilization.
* System Memory Used was around **30-31 GB out of 96 GB total RAM**, although it peaked at 50GB in some long runs.
* CPU usage was observed between 40-60%.
* Temperatures were stable, with SOC Temperature around 56-57°C and CPU Temperature around 60-61°C. DIMM core temperatures were also recorded between 62-68°C.

In summary, the AMD Ryzen 7 255 system, with its 8-core CPU, 96GB DDR5 RAM, and Radeon 780M Graphics, appears to be a robust platform for video-to-video processing with AMD Amuse. The test runs demonstrate varying results depending on the model, prompt, and specific parameters like Strength, Overlap, and Inference Steps. The system can handle demanding workloads, sometimes pushing RAM usage to high levels, but overall shows good performance, especially with models like ToonYou and CyberRealistic when tuned appropriately.