EGU (Enhanced Graphical Upscaler) is a system designed to upscale images while maintaining controlled and predictable resolution output. Unlike a standard upscaler, EGU is a proprietary system that contains AI Models, Parameters and Advanced Systems.
The Pixel Processor is a crucial setting in an Enhanced Graphical Upscaler (EGU) that controls how much VRAM is allocated during upscaling operations. Adjusting this parameter can influence performance speed, system stability, and, in some cases, image quality.
The Pixel Processor controls the VRAM usage of the EGU in a single image pass:
Lower VRAM allocation: May decrease processing speed and slightly reduce image quality. This setting also helps prevent crashes caused by insufficient VRAM.
Higher VRAM allocation: Can increase processing speed and may improve image quality, but it increases the risk of crashes if the system cannot supply enough VRAM.
Users can adjust this setting in the LPS Editor or any software that supports EGU, allowing precise control over how the EGU performs.
Users can adjust this setting in the LPS Editor or any software that supports EGU, allowing fine-grained control over the upscaler’s behavior.
Q.T. (Quality Transistors) is a feature that optimizes an EGU’s internal model to enhance image quality. Unlike a traditional upscaler, Q.T. focuses on improving how the internal model handles details at the target resolution.
While both Magnification and Q.T. upscale and enhance image output, they function differently:
Magnification increases the image resolution based on the model’s scale factor. For example, it can transform a 512×512 image to 1024×1024 or higher. This process primarily enlarges the image while maintaining the existing level of detail.
Q.T. controls the EGU’s internal model to improve image quality. Unlike Magnification, Q.T. does not directly scale resolution linearly. Instead, it allows users to set a target resolution and adjust Q.T. to match or increase the model’s quality output for that resolution. The effect depends on the EGU’s capacity and is not a fixed proportional increase.
In summary, Magnification is resolution-driven, whereas Q.T. is quality-driven
Considerations:
Q.T. effectiveness depends on the EGU’s Q.T. capacity. Some EGU’s handle higher Q.T. better than others.
There is a trade-off between computational cost and quality; increasing Q.T. may require more processing time.
V.R.M. is a memory management system used to support heavy processing workloads, particularly during batch operations. At present, there are two available implementations:
Both versions have their performance influenced by the storage medium in use, such as SSD, HDD, or SATA-based storage. However, despite this shared dependency, eVRM is superior in terms of speed, while cVRM offers superior compression efficiency.
Although V.R.M. is designed to mitigate memory limitations, it still consumes GPU VRAM during processing. Therefore, sufficient and efficient VRAM capacity is required for optimal performance.
A common question is why V.R.M. is useful if VRAM usage still occurs. The answer lies in workload scalability rather than single-task acceleration.
Using V.R.M. instead of system RAM provides several advantages, including improved PC stability under heavy editor workloads. When an out-of-memory condition occurs while using V.R.M., the editor can detect the error and issue a refund. In contrast, if the editor crashes due to inefficient VRAM or RAM usage or other memory-related issues, the error cannot be properly captured, and a refund cannot be offered.
V.R.M. is most effective in scenarios involving single-to-large batch processing. While performance gains may be minimal during isolated operations, the advantages become significant during batch workloads.
Examples include:
Video enhancement processing
Large image batches
Extended processing sessions that would normally trigger VRAM out-of-memory errors
By leveraging V.R.M., these limitations are reduced or eliminated. This allows processing tasks to operate with minimal RAM and VRAM usage, while V.R.M. functions as an extension layer that maintains low memory consumption during processing. Due to this reduced memory footprint, higher output resolutions can be achieved, potentially reaching up to 8K, depending on the selected tier and the EGU max cap.
Increasing V.R.M. does not improve processing performance or increase benchmark scores. Instead, it extends the system’s capability to handle longer processing durations that utilize V.R.M.
V.R.M. does not outperform physical system RAM or GPU VRAM in raw speed. However, it is substantially faster than a traditional page file. This positions V.R.M. as an efficient intermediate memory layer for high-volume graphical processing.
L.S.S. (Light Shadow Simulation) is a technology integrated into LPS Zoom EGUs. It simulates light and shadow in images and videos, applying these simulations to enhance visual quality that elevates content beyond the original.
The output of L.S.S. depends on the EGU, the L.S.S. version, and the hardware in use, offering low to high levels of precision. Higher L.S.S. versions prioritize improved visual quality through more complex simulations and artifact reduction, which may increase processing cost and result in slower performance depending on system capability.
L.S.S. analyzes each image or video frame to determine the direction of light. It then processes how this light interacts with the image and simulates it. After the simulation, the resulting data is applied to your image.
Base version of L.S.S., compatible with most systems.
Can run alongside a standard upscaler or an EGU for enhanced output.
L.S.S. cannot be disabled by default, as it is hard-coded between the EGU and the Editor. If you encounter performance issues, you can enable Xelab AI in the settings. This will switch L.S.S. to the Xelab AI version, which may produce different results and increase in memory usage.
Please note that L.S.S. with Xelab AI can apply aggressive simulations. If the output is not as expected, you can disable Xelab AI during export; however, this may result in slower processing.
| Features | Enhanced Graphical Upscaler | AI Upscaler |
|---|---|---|
|
VRam Control |
✅ Controls VRAM usage to optimize EGU performance. |
❌ No VRAM control |
|
Set a resolution and adjust the quality |
✅ Quality tuning that improves output quality at a target resolution. |
❌ Fixed quality scaling tied only to resolution. |
|
Batch Processing |
✅ Built for large and continuous workloads with stable performance and fewer failures. |
❌ Designed for quick tasks but performance degrades on large or long running jobs. |
|
Cinamatic |
✅ Simulate realistic lighting and shadow effects in your image to achieve a cinematic appearance. Advanced versions of LSS include a ‘Stabilizer’ feature to prevent overshooting during the simulation. |
❌ Just base on your original image |
|
Image and Video High Resolution Output |
✅ It is designed to handle high resolution images, supporting output up to 8K resolution. Even with a minimum of 6 GB of VRAM, you can produce 8K images or videos without running out of memory by leveraging VRM. |
⚠️ Supports unlimited image resolutions using a scale-factor-based approach. Note that video processing may trigger out-of-memory errors if VRAM is insufficient. |
