F2 Benchmark Suite

• CPU-FPGA communication throughput measurement
• FPGA operations-per-second testing
• Memory bandwidth simulation and analysis
• Parallel FPGA workload performance evaluation
• Real-time system resource monitoring
• Automated benchmark execution with timestamped results
• JSON-formatted performance reports

Overview

The F2 Benchmark Suite is a specialized testing framework designed to evaluate the performance characteristics of AWS EC2 F2 instances equipped with AMD Virtex UltraScale+ HBM FPGAs. This suite provides comprehensive metrics on CPU-FPGA communication efficiency, computational throughput, memory bandwidth, and parallel processing capabilities.

Tech Stack

Core Technologies

• Python 3.6+
• Shell scripting
• NumPy for numerical computation
• psutil for system monitoring

Target Hardware

• AWS EC2 F2 instances (f2.48xlarge)
• AMD Virtex UltraScale+ HBM VU47P FPGAs
• Up to 8 FPGAs per instance

System Specifications

• 192 vCPU cores
• 2 TiB system RAM
• 16GB HBM per FPGA (460 GiB/s bandwidth)
• 28 TOPS INT8 computational capacity per FPGA

Key Features

Performance Benchmarking

Comprehensive testing suite measuring critical FPGA performance metrics:

# Example: FPGA throughput benchmark
def benchmark_fpga_operations():
    operations = 0
    start_time = time.time()
 
    # Simulate FPGA computations
    for _ in range(num_iterations):
        data = np.random.rand(matrix_size, matrix_size)
        result = np.dot(data, data.T)
        operations += matrix_size ** 3
 
    elapsed = time.time() - start_time
    gops = (operations / elapsed) / 1e9
 
    return {
        "throughput_gops": gops,
        "elapsed_seconds": elapsed
    }

CPU-FPGA Communication Testing

Measures PCIe bandwidth and data transfer efficiency between host CPU and FPGA accelerators, targeting 7+ GB/s throughput rates.

Memory Bandwidth Analysis

Evaluates High Bandwidth Memory (HBM) performance, simulating data movement patterns to assess the 460 GiB/s bandwidth capabilities per FPGA.

Parallel Workload Execution

Tests multi-FPGA configurations to measure aggregate performance when running distributed computations across all 8 available FPGAs.

Automated Monitoring

Real-time tracking of system resources including CPU usage, memory utilization, and thermal characteristics during benchmark execution.

Benchmark Metrics

Expected Performance Targets:

• Peak memory throughput: 460 GiB/s per FPGA
• PCIe communication: 7+ GB/s
• Aggregate operations: 10+ GOPS across all FPGAs
• Total system throughput: 28 TOPS INT8 per FPGA

Architecture

The benchmark suite consists of three main components working in concert:

1. Primary Benchmark Engine (f2_fpga_benchmark.py): Core testing logic and metric collection
2. Throughput Monitor (fpga_throughput_monitor.py): Continuous performance tracking utility
3. Orchestration Script (run_f2_benchmark.sh): Automated environment setup and execution

# Run complete benchmark suite
./run_f2_benchmark.sh
 
# Results stored with timestamp
# Output: benchmark_results_YYYY-MM-DD_HH-MM-SS.json

Use Cases

This benchmark suite is ideal for:

• Hardware acceleration research and development
• FPGA performance evaluation before production deployment
• Comparing GPU vs FPGA computational efficiency
• Optimizing Zero Knowledge Proof systems on FPGAs
• Academic research on hardware-accelerated cryptography
• Cloud FPGA performance validation

Research Integration

This benchmark suite directly supports research into hardware acceleration for Zero Knowledge Proof systems, providing the performance data needed to compare FPGA implementations against traditional GPU-based approaches. The detailed metrics inform architectural decisions for cryptographic workload optimization.