## Stacked Bar Chart: Performance Comparison of CPU Only vs. CPU-GPU-NPU (Heterogeneous) ### Overview The image is a stacked bar chart comparing the performance of "CPU Only" and "CPU-GPU-NPU (Heterogeneous)" systems. The chart breaks down the total time into three components: "Compute Time", "Memory Transfer", and "Controller Overhead". The y-axis represents time in milliseconds (ms). ### Components/Axes * **X-axis:** Categorical axis with two categories: "CPU Only" and "CPU-GPU-NPU (Heterogeneous)". * **Y-axis:** Numerical axis labeled "Time (ms)" with a scale from 0.0 to 20.0, incrementing by 2.5. * **Legend:** Located in the top-right corner, the legend identifies the components of each stacked bar: * Dark Slate Gray: "Compute Time" * Cadet Blue: "Memory Transfer" * Light Coral: "Controller Overhead" ### Detailed Analysis * **CPU Only:** * Compute Time (Dark Slate Gray): Approximately 18.3 ms * Memory Transfer (Cadet Blue): Approximately 1.8 ms * Controller Overhead (Light Coral): Approximately 0.6 ms * Total Time: 20.7 ms (indicated above the bar) * **CPU-GPU-NPU (Heterogeneous):** * Compute Time (Dark Slate Gray): Approximately 7.3 ms * Memory Transfer (Cadet Blue): Approximately 0.9 ms * Controller Overhead (Light Coral): Approximately 0.4 ms * Total Time: 8.6 ms (indicated above the bar) ### Key Observations * The "CPU-GPU-NPU (Heterogeneous)" system significantly reduces the total time compared to the "CPU Only" system. * "Compute Time" is the dominant factor in both systems, but it is drastically reduced in the "CPU-GPU-NPU (Heterogeneous)" system. * "Memory Transfer" and "Controller Overhead" are relatively small components in both systems. ### Interpretation The chart demonstrates that using a heterogeneous system (CPU-GPU-NPU) leads to a substantial performance improvement compared to a CPU-only system. The primary driver of this improvement is the reduction in "Compute Time," suggesting that the GPU and NPU are effectively offloading and accelerating the computational workload. The "Memory Transfer" and "Controller Overhead" components are also reduced in the heterogeneous system, but their impact on the overall performance is less significant. The data suggests that for the specific workload represented in the chart, leveraging the parallel processing capabilities of GPUs and NPUs is highly beneficial.

## Stacked Bar Chart: Time Breakdown for Computational Scenarios ### Overview This image displays a stacked bar chart comparing the total time taken for two computational scenarios: "CPU Only" and "CPU-GPU-NPU (Heterogeneous)". The time is broken down into three components: "Compute Time", "Memory Transfer", and "Controller Overhead". The chart visually represents the significant reduction in total time when utilizing a heterogeneous computing approach. ### Components/Axes * **Chart Type**: Stacked Bar Chart * **Title**: Implicitly, the chart compares performance between two computational configurations. * **Y-axis Title**: "Time (ms)" * **Scale**: Linear, ranging from 0.0 to 20.0 ms, with major tick marks at 2.5 ms intervals (0.0, 2.5, 5.0, 7.5, 10.0, 12.5, 15.0, 17.5, 20.0). * **X-axis Labels**: * "CPU Only" * "CPU-GPU-NPU (Heterogeneous)" * **Legend**: Located in the top-right quadrant of the chart. * **Compute Time**: Represented by a dark grey color. * **Memory Transfer**: Represented by a teal/cyan color. * **Controller Overhead**: Represented by a coral/salmon color. * **Data Labels**: * "20.7ms" is displayed above the "CPU Only" bar. * "8.6ms" is displayed above the "CPU-GPU-NPU (Heterogeneous)" bar. ### Detailed Analysis The chart contains two stacked bars, each representing a computational scenario. **Bar 1: CPU Only** * **Total Time**: Approximately 20.7 ms (as indicated by the data label). * **Components (from bottom to top)**: * **Compute Time (Dark Grey)**: This component extends from 0.0 ms to approximately 17.7 ms. * *Estimated Value*: ~17.7 ms. * **Memory Transfer (Teal/Cyan)**: This component is stacked on top of Compute Time, extending from approximately 17.7 ms to approximately 20.0 ms. * *Estimated Value*: ~2.3 ms. * **Controller Overhead (Coral/Salmon)**: This component is stacked on top of Memory Transfer, extending from approximately 20.0 ms to approximately 20.7 ms. * *Estimated Value*: ~0.7 ms. **Bar 2: CPU-GPU-NPU (Heterogeneous)** * **Total Time**: Approximately 8.6 ms (as indicated by the data label). * **Components (from bottom to top)**: * **Compute Time (Dark Grey)**: This component extends from 0.0 ms to approximately 7.2 ms. * *Estimated Value*: ~7.2 ms. * **Memory Transfer (Teal/Cyan)**: This component is stacked on top of Compute Time, extending from approximately 7.2 ms to approximately 8.3 ms. * *Estimated Value*: ~1.1 ms. * **Controller Overhead (Coral/Salmon)**: This component is stacked on top of Memory Transfer, extending from approximately 8.3 ms to approximately 8.6 ms. * *Estimated Value*: ~0.3 ms. ### Key Observations * **Significant Time Reduction**: The heterogeneous "CPU-GPU-NPU" configuration results in a total time of approximately 8.6 ms, which is less than half of the time taken by the "CPU Only" configuration (approximately 20.7 ms). This represents a reduction of approximately 12.1 ms, or about a 58.5% improvement. * **Dominance of Compute Time**: In both scenarios, "Compute Time" is the largest contributor to the total time. However, its absolute value is significantly lower in the heterogeneous setup. * **Reduced Overhead**: Both "Memory Transfer" and "Controller Overhead" are also substantially reduced in the heterogeneous configuration compared to the "CPU Only" configuration. ### Interpretation This stacked bar chart clearly demonstrates the performance benefits of employing a heterogeneous computing architecture (CPU-GPU-NPU) over a CPU-only approach for the task represented. The data suggests that the heterogeneous setup is more efficient, leading to a dramatic reduction in overall execution time. The breakdown into "Compute Time", "Memory Transfer", and "Controller Overhead" provides insight into where these improvements are realized. The significant decrease in "Compute Time" indicates that the GPU and NPU are effectively offloading computational tasks from the CPU, leading to faster processing. Furthermore, the reduction in "Memory Transfer" and "Controller Overhead" suggests that the heterogeneous system is also more optimized in terms of data movement and inter-component communication, or that the overall workload is smaller due to faster computation. The data implies that for workloads that can be parallelized or accelerated by specialized hardware like GPUs and NPUs, adopting a heterogeneous architecture is a highly effective strategy for improving performance and reducing latency. The chart serves as a strong piece of evidence for the advantages of such systems.

\n ## Stacked Bar Chart: Performance Comparison - CPU Only vs. CPU-GPU-NPU ### Overview This is a stacked bar chart comparing the total execution time of a task performed on a system using only a CPU versus a heterogeneous system utilizing a CPU, GPU, and NPU. The chart breaks down the total time into three components: Compute Time, Memory Transfer, and Controller Overhead. The total time is displayed above each bar. ### Components/Axes * **X-axis:** Represents the system configuration. Categories are "CPU Only" and "CPU-GPU-NPU (Heterogeneous)". * **Y-axis:** Represents Time in milliseconds (ms), ranging from 0.0 to 22.5 ms, with increments of 2.5 ms. * **Legend (Top-Right):** * Compute Time (Dark Gray) * Memory Transfer (Light Teal) * Controller Overhead (Light Orange) * **Total Time Labels:** Displayed above each bar, indicating the total execution time for each configuration. ### Detailed Analysis The chart consists of two stacked bars, one for each system configuration. **CPU Only:** * Total Time: Approximately 20.7 ms. * Compute Time: Approximately 16.5 ms. This is the largest component of the total time. * Memory Transfer: Approximately 3.2 ms. * Controller Overhead: Approximately 1.0 ms. **CPU-GPU-NPU (Heterogeneous):** * Total Time: Approximately 8.6 ms. * Compute Time: Approximately 6.0 ms. * Memory Transfer: Approximately 1.7 ms. * Controller Overhead: Approximately 0.9 ms. ### Key Observations * The heterogeneous system (CPU-GPU-NPU) demonstrates a significant reduction in total execution time compared to the CPU-only system. The total time is reduced from approximately 20.7 ms to 8.6 ms, representing a decrease of approximately 58%. * Compute Time is the dominant factor in both configurations, but the reduction in Compute Time in the heterogeneous system is substantial. * Memory Transfer and Controller Overhead are relatively small components of the total time in both configurations, but they are also reduced in the heterogeneous system. ### Interpretation The data strongly suggests that utilizing a heterogeneous computing architecture (CPU-GPU-NPU) significantly improves performance for the task being measured. The substantial reduction in Compute Time indicates that the GPU and NPU are effectively offloading computational workload from the CPU. The reduction in Memory Transfer and Controller Overhead suggests that the heterogeneous system is also more efficient in data handling and resource management. The chart highlights the benefits of hardware acceleration and parallel processing. By leveraging the specialized capabilities of the GPU and NPU, the heterogeneous system can complete the task much faster than a CPU-only system. This is likely due to the GPU and NPU being optimized for specific types of computations, allowing them to perform these tasks more efficiently than a general-purpose CPU. The relatively small contribution of Memory Transfer and Controller Overhead suggests that these are not major bottlenecks in either configuration. However, even small improvements in these areas can contribute to the overall performance gain.

## Stacked Bar Chart: Performance Comparison of CPU-Only vs. Heterogeneous Computing ### Overview The image displays a stacked bar chart comparing the total execution time (in milliseconds) and its constituent components for two different computing architectures: a "CPU Only" system and a "CPU-GPU-NPU (Heterogeneous)" system. The chart visually demonstrates a significant performance improvement when using a heterogeneous computing approach. ### Components/Axes * **Chart Type:** Stacked Bar Chart. * **Y-Axis:** Labeled **"Time (ms)"**. The scale runs from 0.0 to 20.0, with major gridlines at intervals of 2.5 ms (0.0, 2.5, 5.0, 7.5, 10.0, 12.5, 15.0, 17.5, 20.0). * **X-Axis:** Contains two categorical bars: 1. **"CPU Only"** (left bar) 2. **"CPU-GPU-NPU (Heterogeneous)"** (right bar) * **Legend:** Positioned in the **top-right corner** of the chart area. It defines the three stacked components: * **Compute Time:** Represented by a **dark gray** color. * **Memory Transfer:** Represented by a **teal/green** color. * **Controller Overhead:** Represented by a **salmon/light red** color. * **Data Labels:** The total time for each bar is displayed directly above it: * Above "CPU Only": **20.7ms** * Above "CPU-GPU-NPU": **8.6ms** ### Detailed Analysis **1. CPU Only Bar (Total: 20.7ms)** * **Compute Time (Dark Gray):** This is the largest component. The segment starts at 0.0 ms and extends to approximately **18.3 ms**. * **Memory Transfer (Teal):** This segment sits atop the Compute Time. It starts at ~18.3 ms and ends at approximately **20.1 ms**, indicating a duration of about **1.8 ms**. * **Controller Overhead (Salmon):** This is the topmost segment. It starts at ~20.1 ms and ends at the labeled total of **20.7 ms**, indicating a duration of about **0.6 ms**. **2. CPU-GPU-NPU (Heterogeneous) Bar (Total: 8.6ms)** * **Compute Time (Dark Gray):** Again the largest component, but significantly reduced. The segment starts at 0.0 ms and extends to approximately **7.4 ms**. * **Memory Transfer (Teal):** This segment sits atop the Compute Time. It starts at ~7.4 ms and ends at approximately **8.2 ms**, indicating a duration of about **0.8 ms**. * **Controller Overhead (Salmon):** The topmost segment. It starts at ~8.2 ms and ends at the labeled total of **8.6 ms**, indicating a duration of about **0.4 ms**. ### Key Observations * **Total Time Reduction:** The heterogeneous system reduces the total execution time by **~58.5%** (from 20.7ms to 8.6ms). * **Compute Time Dominance:** In both architectures, "Compute Time" is the dominant cost, accounting for the vast majority of the total time. * **Proportional Changes:** While all components decrease in absolute time, the reduction in "Compute Time" is the most dramatic, dropping by over 10ms. "Memory Transfer" time is roughly halved, and "Controller Overhead" sees a modest reduction. * **Visual Trend:** The "CPU Only" bar is more than twice the height of the "CPU-GPU-NPU" bar, providing a clear visual cue of the performance advantage. ### Interpretation This chart provides strong empirical evidence for the efficacy of heterogeneous computing (combining CPU, GPU, and NPU) for the measured workload. The data suggests that offloading computational tasks to specialized processors (GPU/NPU) drastically reduces the primary bottleneck—"Compute Time." The associated reductions in "Memory Transfer" and "Controller Overhead" time, while smaller in absolute terms, indicate a more efficient overall system orchestration. The outlier is not a data point but the architecture itself: the heterogeneous approach is the clear solution for minimizing latency in this context. The chart effectively communicates that the performance gain is not from optimizing a single component but from a systemic redesign of how the computation is scheduled and executed across different processing units.

## Stacked Bar Chart: CPU vs Heterogeneous (CPU-GPU-NPU) Performance Comparison ### Overview The chart compares computational performance between two scenarios: "CPU Only" and "CPU-GPU-NPU (Heterogeneous)" across three components: Compute Time, Memory Transfer, and Controller Overhead. Total execution times are 20.7ms (CPU Only) and 8.6ms (Heterogeneous). ### Components/Axes - **X-axis**: Categories labeled "CPU Only" (left) and "CPU-GPU-NPU (Heterogeneous)" (right) - **Y-axis**: Time in milliseconds (0–20ms, increments of 2.5ms) - **Legend**: - Dark blue = Compute Time - Teal = Memory Transfer - Orange = Controller Overhead - **Title**: Positioned at the top center, displaying total times (20.7ms/8.6ms) ### Detailed Analysis 1. **CPU Only (20.7ms total)**: - **Compute Time**: ~18.0ms (dark blue, 87% of total) - **Memory Transfer**: ~2.0ms (teal, 9.7%) - **Controller Overhead**: ~0.7ms (orange, 3.4%) 2. **CPU-GPU-NPU (8.6ms total)**: - **Compute Time**: ~7.5ms (dark blue, 87% of total) - **Memory Transfer**: ~0.8ms (teal, 9.3%) - **Controller Overhead**: ~0.3ms (orange, 3.5%) ### Key Observations - Compute Time dominates both scenarios but decreases by **58%** in the heterogeneous case (18.0ms → 7.5ms). - Memory Transfer and Controller Overhead remain relatively stable as proportions but decrease absolutely (2.0ms → 0.8ms; 0.7ms → 0.3ms). - Heterogeneous configuration reduces total time by **58%** (20.7ms → 8.6ms). ### Interpretation The data demonstrates that offloading computation to GPU/NPU significantly improves performance while maintaining similar overhead proportions. The **Compute Time reduction** (58%) is the primary driver of efficiency gains, suggesting that heterogeneous architectures effectively parallelize workloads. Despite increased system complexity, Memory Transfer and Controller Overhead remain minor contributors (<10% each), indicating that the benefits of parallel processing outweigh the associated costs. This aligns with principles of Amdahl's Law, where speedup is limited by sequential portions of the task.