Why is a speed test for my bare metal GPU showing slower network speeds?

Public internet speed tests are designed for residential connections and are not optimized for data center environments. As a result, these tests often report slower speeds than what your bare metal GPU can actually achieve. Most public test servers are geographically distant, heavily shared, or rate-limited, and the network paths to them may cross congested internet exchanges.

Why Public Speed Tests Appear Slow

Several factors can cause artificially low results when testing from a data center:

• Suboptimal or congested routes: Traffic to public test servers may travel through internet exchanges or networks outside of DigitalOcean’s control.
• Rate-limited or overloaded test servers: Public speed test hosts frequently cap throughput or are tuned for consumer ISP ranges.
• Small file or payload sizes: Many tests use small transfers that do not sustain high throughput long enough to reflect actual speeds.
• Single-threaded tools: Commands like scp or basic curl or wget tests are limited by CPU or protocol overhead and cannot saturate a high-bandwidth link.

These limitations mean that public speed tests are not reliable indicators of your bare metal GPU’s NIC performance or DigitalOcean’s internal network capacity.

Recommended Ways to Measure Network Performance

Instead of public speed tests, use methods designed for data center-grade networking:

• Verify the receiving endpoint’s capacity. Throughput depends on both sides. For example, an external storage provider may throttle uploads.
• Run end-to-end tests with a tool like iperf. iperf can generate multiple parallel streams to accurately measure available throughput.
• Test realistic destinations. If your workload pulls or pushes data to platforms like AWS S3, Google Cloud Storage, or another DigitalOcean resource, test directly against those endpoints.
• Use sufficiently large files. Large, sustained transfers give the most reliable results. Smaller files often under-report throughput due to TCP warm-up time.

Testing within DigitalOcean

If you are moving data between DigitalOcean resources (other bare metal GPUs, Droplets, or Spaces buckets), test performance inside the same region first. This helps you determine whether the bottleneck is internal or external to DigitalOcean.

When to Contact Support

If you still experience slow or inconsistent speeds, contact support. For the fastest help, include the following:

1. A file transfer test from your bare metal GPU to your expected destination.
2. Details about the file size, transfer method (wget, curl, rsync, scp, etc.), and observed speeds.
3. Comparison results from testing against a second endpoint.
4. Any relevant logs, command outputs, or observations.

Providing this information helps us determine whether the issue is caused by a configuration problem, tool limitation, external provider bottleneck, or a route that we can investigate.

DigitalOcean can help diagnose performance issues within our network. However, we may not be able to troubleshoot bottlenecks caused by third-party endpoints or public speed test services.

Related Topics

Can I make BIOS-level changes to bare metal GPUs?

You can request BIOS-level changes to bare metal GPUs through a support ticket. Some changes require reprovisioning.