CPU
QEMU Virtual CPU version 2.5+
BIOS pc-i440fx-10.1 CPU @ 2.0GHz
Memory
7.7Gi (5.9Gi avail)
Storage
96G NVMe (49G free)
Single-Thread
301.74 events/s
sysbench prime calc to 20,000
Completed in 10.0022s
Multi-Thread (4 cores)
1156.60 events/s
sysbench prime calc to 20,000
Completed in 10.0029s
Consistency (10 identical runs)
0.0001s stddev
Avg: 10.0013s | Min: 10.0006s | Max: 10.0027s
Low stddev = consistent = DEDICATED. Burstable instances degrade over repeated runs.
Why CPU consistency matters: This test ran the same computation 10 times. On dedicated hardware, every run takes the same time. On burstable instances (AWS t3, Azure B-series), runs 5-10 slow down as CPU credits deplete. The bar chart above should show nearly identical bars — that's proof of dedicated CPU with zero throttling.
Sequential Write
785 MB/s
dd 1GB direct I/O write
Sequential Read
909 MB/s
dd 1GB direct I/O read
Random 4K Read
73429 IOPS
fio randread, depth 64, 30s
Avg latency: 3468.7 µs
AWS EBS gp3: 3,000 IOPS free / 16,000 max (+$65/mo)
Random 4K Write
61724 IOPS
fio randwrite, depth 64, 30s
Avg latency: 4126.8 µs
AWS EBS gp3: 3,000 IOPS free / 16,000 max (+$65/mo)
Mixed Random R/W (70/30)
69441 IOPS
Read: 48552 + Write: 20889
Most realistic database I/O pattern
| Metric | 639Cloud (NVMe included) | AWS EBS gp3 (baseline) | AWS EBS gp3 (max provisioned) |
|---|
| Random 4K Read IOPS | 73429 | 3,000 | 16,000 (+$65/mo) |
| Random 4K Write IOPS | 61724 | 3,000 | 16,000 (+$65/mo) |
| Sequential Read | 909 MB/s | 125 MB/s | 1,000 MB/s (+$40/mo) |
| Storage Cost | Included in VM price | $0.08/GB/mo + IOPS | $0.08/GB + $0.005/IOPS + throughput |
Key takeaway: 639Cloud NVMe is local to the VM and included at no extra cost. AWS EBS charges separately for volume, provisioned IOPS, and throughput. To match these IOPS on AWS, you'd pay $65+/mo extra for storage alone — on top of the $140/mo compute cost.
Transactions Per Second
1958 TPS
pgbench, 10 clients, 4 threads, 60 seconds
Scale factor 20 (~300MB database)
Real database transactions — not synthetic I/O
Transaction Latency
5.106 ms avg
Per-transaction latency
10 concurrent client connections
Total Transactions
processed:
Completed in 60 seconds
Zero errors
What pgbench measures: Real PostgreSQL transactions (SELECT, UPDATE, INSERT) on real tables with real indexes. This is the most honest database benchmark — it shows exactly how your Postgres will perform on this VM with NVMe storage. The TPS number is what every DBA wants to see.
Requests Per Second
56.937208 req/s
Sustained throughput under load
11986 total requests in 3m30s
P95 Response Time
192.95 ms
95% of requests faster than this
Target was under 200ms — crushed it
P50: 3.69ms | P99: 289.91ms
Error Rate
0.00%
11986 requests, 100 concurrent users
Every request succeeded
| Operation | Avg Latency | What It Tests |
|---|
| Product Lookups | 3.612803 ms | Redis cache + PostgreSQL reads on NVMe |
| Search Queries | 4.650119 ms | Full-text PostgreSQL search |
| Order Writes | 5.149538 ms | Database INSERT on NVMe |
| CPU Benchmark | 14.11862 ms | Prime calculation — consistency under load |
What this test does: 100 concurrent virtual users hit a real application stack for 3.5 minutes — product catalog reads, full-text search, order creation, and CPU-intensive calculations. The mix is 35% reads, 20% search, 20% single lookups, 15% writes, 5% CPU benchmark, 5% aggregation. This is a realistic production traffic pattern, not a synthetic benchmark.
The P95 at 192.95ms under 100 concurrent users means: even during peak load, 95% of your users get a response in under 192.95 milliseconds. On a burstable instance, this number would degrade over the test duration as CPU credits deplete. On dedicated, it stays flat.
| Benchmark | Result | Why It Matters |
|---|
| CPU Consistency | 0.0001s stddev across 10 runs | Dedicated CPU — no throttling, no credit system |
| Random 4K Read IOPS | 73429 IOPS | vs AWS EBS gp3 baseline of 3,000 IOPS |
| Random 4K Write IOPS | 61724 IOPS | NVMe included — no extra storage charges |
| PostgreSQL TPS | 1958 TPS | Real database performance on NVMe |
| App P95 Latency | 192.95 ms | Full stack under 100 concurrent users |
| App Error Rate | 0.00% | 11986 requests, zero failures |
| VM Cost | $75/month | vs AWS m5.xlarge + gp3 = $160-240+/month |
This entire benchmark ran on a single 639Cloud Medium VM at $75/month.
Dedicated CPU (no throttling) + NVMe storage (included) + bandwidth (included) = predictable, high-performance infrastructure at roughly half the cost of equivalent AWS dedicated instances.
Powered by 100% renewable solar energy.
