Fix cudaErrorPeerAccessUnsupported: P2P Not Available

Overview

cudaErrorPeerAccessUnsupported occurs when attempting P2P access between GPUs that don't support direct communication.

Error Messages

CUDA error: peer access unsupported
cudaErrorPeerAccessUnsupported

Common Causes

• •GPUs on different PCIe root complexes
• •No NVLink connection
• •Different GPU architectures
• •Virtualized environment blocking P2P

Solutions

int canAccess;
cudaDeviceCanAccessPeer(&canAccess, dev0, dev1);
if (canAccess) {
    cudaSetDevice(dev0);
    cudaDeviceEnablePeerAccess(dev1, 0);
}

cudaMemcpyPeer(d_dst, dstDev, d_src, srcDev, size);

Prevention Tips

• ✓Check cudaDeviceCanAccessPeer before enabling
• ✓Use NVLink-connected GPUs for best P2P
• ✓Fall back to cudaMemcpyPeer

Code Examples

Before (Problematic)

Not checking capability first.

cudaDeviceEnablePeerAccess(1, 0);  // May fail

After (Fixed)

Check before enabling.

int canAccess;
cudaDeviceCanAccessPeer(&canAccess, 0, 1);
if (canAccess) cudaDeviceEnablePeerAccess(1, 0);

Frequently Asked Questions

Related Errors

Need help debugging CUDA errors? Download RightNow AI for intelligent error analysis and optimization suggestions.