## Chart: Frametime vs. Shader Complexity for Different Rendering Techniques ### Overview The image presents three line charts comparing the frametime performance of CPU and GPU rendering for two scenes ("Fairy Forest" and "Buddha") under different shader complexities. The charts illustrate the performance of three rendering techniques: FreePipe-style Aggressive Stage Fusion, Spatial Binning, and Binning with Stage Fusion. The x-axis represents shader complexity (number of lights), and the y-axis represents frametime relative to a baseline. ### Components/Axes **Common Elements:** * **X-axis:** Shader Complexity (Number of Lights). The x-axis is logarithmic, with markers at 1, 10, 100, and 1000. * **Y-axis:** Frametime (Relative to Baseline). The y-axis scale varies between the three charts. * **Legend:** Located at the top of each chart. * Blue: CPU Fairy Forest * Purple: CPU Buddha * Red: GPU Fairy Forest * Green: GPU Buddha **Chart (a): FreePipe-style Aggressive Stage Fusion** * Y-axis ranges from 0 to 7. **Chart (b): Spatial Binning** * Y-axis ranges from 0 to 1.4. **Chart (c): Binning with Stage Fusion** * Y-axis ranges from 0 to 6. Y-axis label is "Frametime (Relative to all LoadBalance Schedules)". ### Detailed Analysis **Chart (a): FreePipe-style Aggressive Stage Fusion** * **CPU Fairy Forest (Blue):** The frametime increases steadily with shader complexity. * At 1 light: ~0.6 * At 10 lights: ~1.3 * At 100 lights: ~2.5 * At 1000 lights: ~3.0 * **CPU Buddha (Purple):** The frametime remains relatively constant with increasing shader complexity. * At 1 light: ~0.5 * At 10 lights: ~0.6 * At 100 lights: ~0.8 * At 1000 lights: ~0.9 * **GPU Fairy Forest (Red):** The frametime remains very low until a sharp increase at 1000 lights. * At 1 light: ~0.1 * At 10 lights: ~0.3 * At 100 lights: ~0.4 * At 1000 lights: ~6.5 * **GPU Buddha (Green):** The frametime increases slowly until a sharp increase at 1000 lights. * At 1 light: ~1.3 * At 10 lights: ~1.1 * At 100 lights: ~1.3 * At 1000 lights: ~2.0 **Chart (b): Spatial Binning** * **CPU Fairy Forest (Blue):** The frametime increases with shader complexity, but plateaus. * At 1 light: ~0.45 * At 10 lights: ~0.6 * At 100 lights: ~0.9 * At 1000 lights: ~0.95 * **CPU Buddha (Purple):** The frametime remains relatively constant with increasing shader complexity. * At 1 light: ~0.78 * At 10 lights: ~0.78 * At 100 lights: ~0.85 * At 1000 lights: ~0.9 * **GPU Fairy Forest (Red):** The frametime remains very low and constant. * At 1 light: ~0.08 * At 10 lights: ~0.08 * At 100 lights: ~0.1 * At 1000 lights: ~0.1 * **GPU Buddha (Green):** The frametime decreases slightly with shader complexity, then plateaus. * At 1 light: ~1.25 * At 10 lights: ~1.05 * At 100 lights: ~1.0 * At 1000 lights: ~0.95 **Chart (c): Binning with Stage Fusion** * **CPU Fairy Forest (Blue):** The frametime remains relatively constant. * At 1 light: ~0.5 * At 10 lights: ~0.6 * At 100 lights: ~0.7 * At 1000 lights: ~0.7 * **CPU Buddha (Purple):** The frametime remains relatively constant. * At 1 light: ~0.7 * At 10 lights: ~0.7 * At 100 lights: ~0.7 * At 1000 lights: ~0.7 * **GPU Fairy Forest (Red):** The frametime remains low until a sharp increase at 1000 lights. * At 1 light: ~0.4 * At 10 lights: ~0.3 * At 100 lights: ~0.3 * At 1000 lights: ~5.5 * **GPU Buddha (Green):** The frametime increases slowly until a sharp increase at 1000 lights. * At 1 light: ~1.0 * At 10 lights: ~1.5 * At 100 lights: ~2.0 * At 1000 lights: ~2.0 ### Key Observations * GPU rendering of the "Fairy Forest" scene (Red line) experiences a significant performance drop (increased frametime) at high shader complexity (1000 lights) in the "FreePipe-style Aggressive Stage Fusion" and "Binning with Stage Fusion" techniques. * Spatial Binning (Chart b) generally provides more stable frametimes across all shader complexities for both CPU and GPU rendering. * CPU rendering of the "Fairy Forest" scene (Blue line) shows a steady increase in frametime with increasing shader complexity in the "FreePipe-style Aggressive Stage Fusion" technique. * CPU rendering of the "Buddha" scene (Purple line) is relatively stable across all shader complexities and rendering techniques. * GPU rendering of the "Buddha" scene (Green line) shows a sharp increase in frametime with increasing shader complexity in the "FreePipe-style Aggressive Stage Fusion" and "Binning with Stage Fusion" techniques. ### Interpretation The charts demonstrate the impact of shader complexity on frametime performance for different rendering techniques and hardware (CPU vs. GPU). The "Fairy Forest" scene appears to be more sensitive to shader complexity, particularly when rendered on the GPU using "FreePipe-style Aggressive Stage Fusion" and "Binning with Stage Fusion". Spatial Binning seems to offer a more stable performance profile, suggesting it is more resilient to increasing shader complexity. The "Buddha" scene, when rendered on the CPU, exhibits relatively consistent performance across all techniques and shader complexities. The sharp increase in frametime for GPU rendering at 1000 lights in charts (a) and (c) suggests a potential bottleneck or limitation in the GPU's ability to handle extremely complex shaders with these techniques. These results could inform decisions about which rendering technique to use based on the target hardware and the complexity of the scene.

## Charts/Graphs: Frametime Performance Across Different Rendering Techniques and Hardware ### Overview This image presents three line charts, labeled (a), (b), and (c), each illustrating the "Frametime" performance relative to a baseline or other schedules, as "Shader Complexity (Number of Lights)" increases. The charts compare the performance of "CPU Fairy Forest", "CPU Buddha", "GPU Fairy Forest", and "GPU Buddha" across three different rendering techniques: "FreePipe-style Aggressive Stage Fusion", "Spatial Binning", and "Binning with Stage Fusion". The x-axis for all charts is logarithmic, representing shader complexity, while the y-axis is linear, representing frametime. ### Components/Axes The image consists of three distinct sub-charts arranged horizontally. Each sub-chart shares a common x-axis label and legend, but has a unique y-axis label and scale. **Common X-axis:** - **Title:** "Shader Complexity (Number of Lights)" - **Scale:** Logarithmic, ranging from 1 to 1000. - **Markers:** 1, 10, 100, 1000. **Common Legend (positioned in the top-right corner of each sub-chart):** - **CPU Fairy Forest:** Blue square marker, blue line. - **CPU Buddha:** Purple diamond marker, purple line. - **GPU Fairy Forest:** Red circle marker, red line. - **GPU Buddha:** Green triangle marker, green line. **Chart (a) Specific Y-axis:** - **Title:** "Frametime (Relative to Baseline)" - **Scale:** Linear, ranging from 0 to 7. - **Markers:** 0, 1, 2, 3, 4, 5, 6, 7. **Chart (b) Specific Y-axis:** - **Title:** "Frametime (Relative to Baseline)" - **Scale:** Linear, ranging from 0 to 1.4. - **Markers:** 0, 0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4. **Chart (c) Specific Y-axis:** - **Title:** "Frametime (Relative to all LoadBalance Schedules)" - **Scale:** Linear, ranging from 0 to 6. - **Markers:** 0, 1, 2, 3, 4, 5, 6. ### Detailed Analysis #### (a) FreePipe-style Aggressive Stage Fusion This chart, located in the bottom-left, shows frametime relative to baseline using "FreePipe-style Aggressive Stage Fusion". - **CPU Fairy Forest (Blue Square):** The frametime starts at approximately 0.5 at 1 light, increases to about 1.3 at 10 lights, then to 2.6 at 100 lights, and finally reaches approximately 3.0 at 1000 lights. The trend is a steady increase, with a slight acceleration between 10 and 100 lights, then a slower increase. - **CPU Buddha (Purple Diamond):** The frametime starts at approximately 0.3 at 1 light, increases to about 0.5 at 10 lights, then to 0.8 at 100 lights, and stabilizes around 0.9 at 1000 lights. The trend shows a moderate increase followed by a flattening. - **GPU Fairy Forest (Red Circle):** The frametime starts very low, at approximately 0.2 at 1 light, remains low at about 0.3 at 10 lights, then increases to 0.8 at 100 lights, and sharply rises to approximately 5.5 at 1000 lights. The trend is relatively flat at low complexity, followed by a dramatic exponential increase at high complexity. - **GPU Buddha (Green Triangle):** The frametime starts at approximately 1.3 at 1 light, decreases to about 0.9 at 10 lights, then increases to 1.5 at 100 lights, and sharply rises to approximately 3.8 at 1000 lights. The trend shows an initial decrease, followed by a sharp increase at higher complexities. #### (b) Spatial Binning This chart, located in the bottom-center, shows frametime relative to baseline using "Spatial Binning". - **CPU Fairy Forest (Blue Square):** The frametime starts at approximately 0.45 at 1 light, increases to about 0.58 at 10 lights, then to 0.85 at 100 lights, and reaches approximately 0.95 at 1000 lights. The trend is a consistent, moderate increase. - **CPU Buddha (Purple Diamond):** The frametime starts at approximately 0.78 at 1 light, remains around 0.78 at 10 lights, then increases slightly to 0.85 at 100 lights, and reaches approximately 0.95 at 1000 lights. The trend is largely flat at lower complexities, with a slight increase at higher complexities. - **GPU Fairy Forest (Red Circle):** The frametime starts very low, at approximately 0.08 at 1 light, remains around 0.08 at 10 lights and 100 lights, and slightly increases to approximately 0.1 at 1000 lights. The trend is remarkably flat and consistently very low across all complexities. - **GPU Buddha (Green Triangle):** The frametime starts at approximately 1.25 at 1 light, decreases to about 1.0 at 10 lights, remains around 1.0 at 100 lights, and slightly decreases to approximately 0.95 at 1000 lights. The trend shows an initial decrease, then flattens out. #### (c) Binning with Stage Fusion This chart, located in the bottom-right, shows frametime relative to all LoadBalance Schedules using "Binning with Stage Fusion". - **CPU Fairy Forest (Blue Square):** The frametime starts at approximately 0.5 at 1 light, increases to about 0.6 at 10 lights, then to 0.7 at 100 lights, and remains around 0.7 at 1000 lights. The trend shows a slight increase followed by a flattening. - **CPU Buddha (Purple Diamond):** The frametime starts at approximately 0.6 at 1 light, increases to about 0.7 at 10 lights, then to 0.8 at 100 lights, and remains around 0.8 at 1000 lights. The trend shows a slight increase followed by a flattening. - **GPU Fairy Forest (Red Circle):** The frametime starts at approximately 0.4 at 1 light, increases to about 0.5 at 10 lights, then to 0.8 at 100 lights, and sharply rises to approximately 5.0 at 1000 lights. The trend is relatively flat at low complexity, followed by a dramatic exponential increase at high complexity. - **GPU Buddha (Green Triangle):** The frametime starts at approximately 0.9 at 1 light, increases to about 1.2 at 10 lights, then to 1.5 at 100 lights, and sharply rises to approximately 4.0 at 1000 lights. The trend shows a steady increase at lower complexities, followed by a sharp exponential increase at high complexity. ### Key Observations - **GPU Fairy Forest Performance:** In both "FreePipe-style Aggressive Stage Fusion" (a) and "Binning with Stage Fusion" (c), GPU Fairy Forest exhibits a dramatic increase in frametime at high shader complexities (1000 lights), becoming the worst performer. However, in "Spatial Binning" (b), GPU Fairy Forest consistently maintains the lowest frametime across all complexities. - **CPU Performance Stability:** CPU-based methods (CPU Fairy Forest and CPU Buddha) generally show more stable and predictable frametime increases compared to GPU methods, especially at higher complexities, in charts (a) and (c). Their frametime curves tend to flatten or increase linearly rather than exponentially. - **Spatial Binning Effectiveness:** "Spatial Binning" (b) appears to be highly effective for GPU Fairy Forest, keeping its frametime very low. It also shows that CPU Buddha and CPU Fairy Forest have frametimes below 1.0 (relative to baseline) for all complexities, suggesting good performance. GPU Buddha in (b) also performs well, staying below 1.25. - **Impact of Stage Fusion:** The addition of "Stage Fusion" (a) and (c) seems to negatively impact GPU performance at high complexities, leading to significant frametime spikes for both GPU Fairy Forest and GPU Buddha. - **CPU Buddha Consistency:** CPU Buddha generally maintains a relatively low and stable frametime across all three techniques and complexities, often performing better than CPU Fairy Forest, especially at higher complexities in (a) and (c). ### Interpretation The data suggests that the choice of rendering technique significantly impacts performance, particularly for GPU-based rendering as shader complexity increases. 1. **Spatial Binning (b) is highly optimized for GPU Fairy Forest:** The "Spatial Binning" technique demonstrates exceptional efficiency for "GPU Fairy Forest", maintaining a frametime close to zero relative to the baseline, even with 1000 lights. This indicates that spatial binning effectively manages the workload for this specific GPU configuration and scene type, preventing the performance degradation seen in other techniques. CPU methods also perform well under spatial binning, staying below the baseline. 2. **Stage Fusion introduces scalability challenges for GPUs:** Both "FreePipe-style Aggressive Stage Fusion" (a) and "Binning with Stage Fusion" (c) show that when stage fusion is involved, GPU performance (especially "GPU Fairy Forest" and "GPU Buddha") degrades significantly and non-linearly at higher shader complexities. This suggests that the overhead or architectural limitations related to stage fusion become a bottleneck for GPUs when processing a large number of lights. The exponential rise in frametime indicates a potential resource exhaustion or a non-optimal scaling of the fusion process on GPUs. 3. **CPUs handle stage fusion complexity more gracefully:** In contrast to GPUs, CPU-based rendering ("CPU Fairy Forest" and "CPU Buddha") with stage fusion (a and c) shows a more controlled increase in frametime. While frametime does increase, it tends to flatten out or increase linearly, suggesting that CPUs might be better equipped to handle the specific type of workload introduced by stage fusion, or that their performance degradation model is different and less severe than GPUs in these scenarios. "CPU Buddha" consistently shows better or comparable performance to "CPU Fairy Forest" across all scenarios, indicating its potential efficiency. 4. **Baseline and Relative Performance:** The y-axis labels are crucial. In (a) and (b), "Relative to Baseline" implies a comparison to a standard or initial performance. In (c), "Relative to all LoadBalance Schedules" suggests a comparison against a set of load-balancing techniques, implying that the values represent how well "Binning with Stage Fusion" performs compared to other load-balancing approaches. Values below 1.0 are generally desirable, indicating better performance than the baseline/other schedules. In summary, for scenarios with high shader complexity, "Spatial Binning" appears to be the superior technique, especially for GPU rendering. However, when "Stage Fusion" is employed, CPU-based rendering, particularly "CPU Buddha", demonstrates more robust and scalable performance compared to GPU-based methods, which suffer significant performance drops. This highlights a trade-off between different rendering techniques and hardware platforms depending on the specific demands of the scene (shader complexity) and the chosen optimization strategies.

\n ## Line Charts: Performance Comparison of Rendering Techniques ### Overview The image presents three line charts comparing the performance of different rendering techniques (CPU Fairy Forest, CPU Buddha, GPU Fairy Forest, GPU Buddha) across varying levels of shader complexity, measured by the number of lights. Each chart focuses on a different optimization strategy: (a) FreePipe-style Aggressive Stage Fusion, (b) Spatial Binning, and (c) Binning with Stage Fusion. The performance metric varies across the charts: Frametime (relative to baseline) for (a) and (b), and LoadBalance (relative to all LoadBalance schedules) for (c). ### Components/Axes Each chart shares the following components: * **X-axis:** Shader Complexity (Number of Lights) - Scale: 1, 10, 100, 1000. * **Y-axis:** Varies per chart. * (a): Frametime (Relative to Baseline) - Scale: 0 to 7, increments of 1. * (b): Frametime (Relative to Baseline) - Scale: 0 to 1.4, increments of 0.2. * (c): LoadBalance (Relative to all LoadBalance schedules) - Scale: 0 to 6, increments of 1. * **Legend:** Located at the top-right of each chart. * CPU Fairy Forest (Blue Line) * CPU Buddha (Green Line) * GPU Fairy Forest (Red Line) * GPU Buddha (Yellow Line) * **Chart Titles:** Located below the x-axis. * (a): FreePipe-style Aggressive Stage Fusion * (b): Spatial Binning * (c): Binning with Stage Fusion ### Detailed Analysis or Content Details **Chart (a): FreePipe-style Aggressive Stage Fusion** * **CPU Fairy Forest (Blue):** Starts at approximately 0.1, increases steadily to around 0.6 at 100 lights, and then jumps sharply to approximately 2.5 at 1000 lights. * **CPU Buddha (Green):** Starts at approximately 0.2, increases to around 0.8 at 100 lights, and then rises sharply to approximately 6.5 at 1000 lights. * **GPU Fairy Forest (Red):** Remains relatively flat around 0.1-0.2 across all shader complexities. * **GPU Buddha (Yellow):** Starts at approximately 0.1, increases to around 0.3 at 100 lights, and then rises to approximately 1.5 at 1000 lights. **Chart (b): Spatial Binning** * **CPU Fairy Forest (Blue):** Starts at approximately 0.8, decreases slightly to around 0.7 at 10 lights, and then remains relatively stable around 0.8-0.9 up to 1000 lights. * **CPU Buddha (Green):** Starts at approximately 0.9, decreases to around 0.7 at 10 lights, and then increases to approximately 1.1 at 1000 lights. * **GPU Fairy Forest (Red):** Remains consistently low, around 0.1-0.2 across all shader complexities. * **GPU Buddha (Yellow):** Starts at approximately 0.6, decreases to around 0.4 at 10 lights, and then remains relatively stable around 0.5-0.6 up to 1000 lights. **Chart (c): Binning with Stage Fusion** * **CPU Fairy Forest (Blue):** Remains relatively flat around 0.8-1.0 across all shader complexities. * **CPU Buddha (Green):** Starts at approximately 0.9, decreases to around 0.7 at 10 lights, and then increases to approximately 1.5 at 1000 lights. * **GPU Fairy Forest (Red):** Starts at approximately 0.2, decreases to around 0.1 at 10 lights, and then remains relatively stable around 0.1-0.2 up to 1000 lights. * **GPU Buddha (Yellow):** Starts at approximately 1.0, decreases to around 0.8 at 10 lights, and then rises sharply to approximately 5.5 at 1000 lights. ### Key Observations * **GPU consistently outperforms CPU:** Across all three charts, the GPU-based rendering techniques (Fairy Forest and Buddha) generally exhibit lower frametimes and/or better load balance compared to their CPU counterparts, especially at higher shader complexities. * **Shader complexity impact:** In chart (a) and (c), both CPU-based techniques show a significant performance degradation (increased frametime/load balance) as shader complexity increases. * **Spatial Binning (b) stabilizes CPU performance:** Chart (b) demonstrates that Spatial Binning helps to stabilize the performance of CPU-based techniques, preventing the sharp increase in frametime observed in chart (a) and (c). * **GPU Buddha outlier in (c):** GPU Buddha shows a dramatic increase in LoadBalance at 1000 lights in chart (c), suggesting a potential bottleneck or inefficiency at high shader complexity. ### Interpretation These charts compare the scalability of different rendering techniques under increasing shader complexity. The consistent outperformance of GPU-based methods highlights the benefits of GPU acceleration for this workload. The effectiveness of Spatial Binning in mitigating performance degradation on the CPU suggests that this optimization technique can improve scalability. The outlier behavior of GPU Buddha in chart (c) warrants further investigation to identify the root cause of the performance bottleneck at high shader complexity. The data suggests that the choice of rendering technique and optimization strategy significantly impacts performance, and that careful consideration must be given to the target hardware and expected shader complexity. The FreePipe-style Aggressive Stage Fusion appears to be the least effective optimization for CPU rendering, while Spatial Binning offers a substantial improvement. Binning with Stage Fusion provides a good balance for GPU rendering, but the outlier behavior of GPU Buddha at high complexity needs to be addressed.

## Line Charts: Performance Comparison of Three Rendering Techniques ### Overview The image displays three line charts arranged horizontally, comparing the performance (framerate relative to a baseline) of two 3D models ("Fairy Forest" and "Buddha") rendered on CPU and GPU across three different rendering optimization techniques. The charts plot performance against increasing shader complexity (number of lights). ### Components/Axes * **Common X-Axis (All Charts):** "Shader Complexity (Number of Lights)". It is a logarithmic scale with major tick marks at 1, 10, 100, and 1000. * **Common Legend (All Charts):** Located in the top-left corner of each chart. * Blue Square Line: `CPU Fairy Forest` * Red Circle Line: `GPU Fairy Forest` * Purple Diamond Line: `CPU Buddha` * Green Triangle Line: `GPU Buddha` * **Chart-Specific Y-Axes:** * **Chart (a):** "Frametime (Relative to Baseline)". Scale ranges from 0 to 7. * **Chart (b):** "Frametime (Relative to Baseline)". Scale ranges from 0 to 1.4. * **Chart (c):** "Frametime (Relative to all LoadBalance Schedules)". Scale ranges from 0 to 6. ### Detailed Analysis #### Chart (a): FreePipe-style Aggressive Stage Fusion * **Trend Verification:** * `CPU Fairy Forest` (Blue): Slopes steadily upward. * `GPU Fairy Forest` (Red): Slopes gently upward until 100 lights, then spikes sharply upward. * `CPU Buddha` (Purple): Remains nearly flat, with a very slight upward trend. * `GPU Buddha` (Green): Slopes gently upward until 100 lights, then increases more sharply. * **Data Points (Approximate):** * **At 1 Light:** CPU Fairy Forest ~0.5, GPU Fairy Forest ~0.2, CPU Buddha ~0.2, GPU Buddha ~1.3. * **At 10 Lights:** CPU Fairy Forest ~1.3, GPU Fairy Forest ~0.3, CPU Buddha ~0.4, GPU Buddha ~0.7. * **At 100 Lights:** CPU Fairy Forest ~2.6, GPU Fairy Forest ~0.7, CPU Buddha ~0.8, GPU Buddha ~1.0. * **At 1000 Lights:** CPU Fairy Forest ~3.0, GPU Fairy Forest ~6.0, CPU Buddha ~0.9, GPU Buddha ~4.0. #### Chart (b): Spatial Binning * **Trend Verification:** * `CPU Fairy Forest` (Blue): Slopes steadily upward. * `GPU Fairy Forest` (Red): Remains very low and nearly flat. * `CPU Buddha` (Purple): Remains nearly flat. * `GPU Buddha` (Green): Slopes gently downward. * **Data Points (Approximate):** * **At 1 Light:** CPU Fairy Forest ~0.45, GPU Fairy Forest ~0.05, CPU Buddha ~0.8, GPU Buddha ~1.25. * **At 10 Lights:** CPU Fairy Forest ~0.6, GPU Fairy Forest ~0.05, CPU Buddha ~0.8, GPU Buddha ~1.0. * **At 100 Lights:** CPU Fairy Forest ~0.85, GPU Fairy Forest ~0.05, CPU Buddha ~0.85, GPU Buddha ~1.0. * **At 1000 Lights:** CPU Fairy Forest ~1.0, GPU Fairy Forest ~0.1, CPU Buddha ~1.0, GPU Buddha ~0.9. #### Chart (c): Binning with Stage Fusion * **Trend Verification:** * `CPU Fairy Forest` (Blue): Slopes gently upward. * `GPU Fairy Forest` (Red): Slopes gently upward until 100 lights, then spikes sharply upward. * `CPU Buddha` (Purple): Remains nearly flat. * `GPU Buddha` (Green): Slopes gently upward until 100 lights, then increases more sharply. * **Data Points (Approximate):** * **At 1 Light:** CPU Fairy Forest ~0.6, GPU Fairy Forest ~0.5, CPU Buddha ~0.8, GPU Buddha ~0.9. * **At 10 Lights:** CPU Fairy Forest ~0.7, GPU Fairy Forest ~0.6, CPU Buddha ~0.9, GPU Buddha ~1.3. * **At 100 Lights:** CPU Fairy Forest ~0.9, GPU Fairy Forest ~1.1, CPU Buddha ~1.0, GPU Buddha ~1.1. * **At 1000 Lights:** CPU Fairy Forest ~1.0, GPU Fairy Forest ~5.0, CPU Buddha ~1.0, GPU Buddha ~4.0. ### Key Observations 1. **Performance Cliff for GPU:** In techniques (a) and (c), the GPU performance for both models degrades severely (framerate spikes) when shader complexity reaches 1000 lights. This cliff is absent in technique (b). 2. **CPU Stability:** CPU performance for the "Buddha" model is remarkably stable across all techniques and complexities. CPU performance for "Fairy Forest" degrades linearly with complexity in all cases. 3. **Technique (b) Anomaly:** Spatial Binning (b) shows a unique trend where GPU performance for "Buddha" actually *improves* slightly as complexity increases, and GPU performance for "Fairy Forest" remains excellent and flat. 4. **Model Sensitivity:** The "Fairy Forest" model generally shows greater performance sensitivity (steeper slopes) to increased complexity than the "Buddha" model, especially on the GPU. ### Interpretation The data demonstrates a critical performance trade-off in rendering pipeline design. **FreePipe-style Aggressive Stage Fusion (a)** and **Binning with Stage Fusion (c)** offer good performance at low-to-moderate light counts but suffer from a catastrophic performance breakdown at very high complexity (1000 lights), particularly on the GPU. This suggests these methods may have scalability issues or encounter a resource bottleneck (e.g., register pressure, memory bandwidth) under extreme load. In contrast, **Spatial Binning (b)** exhibits excellent and stable scalability. It maintains low frametimes for GPU workloads even at 1000 lights, indicating it effectively manages the increased computational load. The slight performance *improvement* for the GPU Buddha model is anomalous and could suggest that the binning strategy becomes more efficient with certain scene characteristics at higher densities, or it may be within measurement noise. The consistent linear degradation of CPU Fairy Forest performance across all techniques implies that CPU-bound work for that model scales directly with light count, unaffected by the GPU-focused optimizations. The stability of CPU Buddha suggests its workload is less dependent on the number of lights. This analysis highlights that the choice of rendering technique must be guided by the expected scene complexity (light count) and the target hardware's performance profile.

## Line Graphs: Rendering Technique Performance vs Shader Complexity ### Overview Three line graphs compare the performance of CPU/GPU rendering techniques (Fairy Forest and Buddha) across increasing shader complexity (number of lights). Each chart represents a different optimization strategy: (a) FreePipe-style Aggressive Stage Fusion, (b) Spatial Binning, and (c) Binning with Stage Fusion. Performance is measured as "Frametime Relative to Baseline" on the y-axis, with shader complexity (1–1000 lights) on the x-axis. --- ### Components/Axes 1. **X-Axis**: Shader Complexity (Number of Lights) - Logarithmic scale: 1, 10, 100, 1000 2. **Y-Axis**: Frametime Relative to Baseline - Varies by chart: - (a) 0–7 - (b) 0–1.4 - (c) 0–6 3. **Legends**: - **CPU Fairy Forest** (blue squares) - **CPU Buddha** (purple diamonds) - **GPU Fairy Forest** (red circles) - **GPU Buddha** (green triangles) - Positioned in the top-left corner of each chart. --- ### Detailed Analysis #### Chart (a): FreePipe-style Aggressive Stage Fusion - **Trends**: - **CPU Buddha** (purple): Gradual increase from ~0.2 to ~1.0. - **CPU Fairy Forest** (blue): Steeper rise from ~0.5 to ~3.0. - **GPU Buddha** (green): Starts at ~1.2, dips to ~0.8 at 10 lights, then rises sharply to ~4.0 at 1000 lights. - **GPU Fairy Forest** (red): Sharp increase from ~0.1 to ~6.5 at 1000 lights. - **Key Data Points**: - At 1 light: All methods ~0.1–0.2. - At 100 lights: GPU Buddha ~1.5, GPU Fairy Forest ~2.0. - At 1000 lights: GPU Fairy Forest peaks at ~6.5 (highest), GPU Buddha ~4.0. #### Chart (b): Spatial Binning - **Trends**: - **CPU Buddha** (purple): Flat (~0.8). - **CPU Fairy Forest** (blue): Moderate rise from ~0.4 to ~0.9. - **GPU Buddha** (green): Slight decline from ~1.2 to ~0.9. - **GPU Fairy Forest** (red): Stable (~0.1–0.2). - **Key Data Points**: - At 1 light: GPU Buddha ~1.2 (highest). - At 1000 lights: CPU Buddha ~0.9, GPU Buddha ~0.9. #### Chart (c): Binning with Stage Fusion - **Trends**: - **CPU Buddha** (purple): Flat (~1.0). - **CPU Fairy Forest** (blue): Slight rise from ~0.8 to ~1.0. - **GPU Buddha** (green): Sharp increase from ~1.0 to ~5.0 at 1000 lights. - **GPU Fairy Forest** (red): Moderate rise from ~0.3 to ~5.0 at 1000 lights. - **Key Data Points**: - At 1 light: All methods ~0.3–1.0. - At 1000 lights: GPU Buddha ~5.0 (highest), GPU Fairy Forest ~5.0. --- ### Key Observations 1. **GPU Performance Degradation**: - GPU Buddha (green) shows significant performance drops in (a) and (c) as shader complexity increases, suggesting inefficiency in handling complex lighting. 2. **CPU vs GPU**: - CPU methods (blue/purple) generally outperform GPUs in (b) but lag behind in (a) and (c). 3. **Fairy Forest vs Buddha**: - GPU Fairy Forest (red) consistently outperforms GPU Buddha (green) in (a) and (c), but underperforms in (b). 4. **Spatial Binning Stability**: - Chart (b) shows the least performance variation across methods, with GPU Buddha maintaining near-baseline performance. --- ### Interpretation The data highlights trade-offs between rendering techniques and hardware: - **FreePipe-style Fusion (a)** and **Binning with Stage Fusion (c)** amplify GPU inefficiencies at high complexity, with GPU Buddha suffering the most. - **Spatial Binning (b)** mitigates performance degradation, making GPU Buddha competitive. - CPU methods (Fairy Forest/Buddha) scale better in (b) but struggle with aggressive optimizations in (a) and (c). - The sharp spikes in GPU performance (e.g., GPU Fairy Forest in (a)) suggest potential bottlenecks in shader compilation or memory access at extreme complexity. This analysis underscores the importance of optimization strategy (e.g., spatial binning) in balancing GPU/CPU performance across varying workloads.