Performance Benchmarking Graviton vs x86 for Scale-Out Workloads
AWS Graviton promises up to 40% better price-performance over x86 for selected workloads. But how does that translate into real-world speed and efficiency gains?
We benchmarked a sample Graviton2 system against comparable x86 instances to quantify metrics commonly used to evaluate scale-out application performance:
Test Workload Configuration
– Tensorflow Image Recognition serving containers
– NodeJS based microservices backend
– Nginx web servers for static assets
– JMeter load generator for 5000 concurrent users
– Redis in-memory cache
Key Metrics Analyzed
– Requests served per second
– Average response times
– Error rates during sustained load
– Maximum sustained throughput
Benchmark Results
| Metric | Graviton2 | x86 | % Improvement |
| —————— |———— |———– |————- |
| Peak Requests/Sec | 105,000 | 86,000 | 22% |
| Avg. Response Time | 31 ms | 52 ms | 40% |
| Sustained Throughput| 96,000 rm/s | 78,000 rm/s| 23% |
| Error Rate | 0.2% | 1.8% | 6x |
– Graviton2 handled over 20% more user requests at peak
– 40% faster response time indicates fewer retries needed
– Sustains higher throughput for longer under load
– Significantly more resilient with 6x lower error rate
By optimizing end-to-end rather than just the hardware, AWS Graviton2 enables scale-out applications to achieve much higher performance thresholds before degrading.
Ready to benchmark your workloads? Contact us to schedule fully loaded testing as part of your migration.