NUMA

• Shared memory
• Often made by physically linking two or more SMP
• One SMP can directly access memory of another SMP
• Not all processors have equal access time to all memories
• Memory access across link is slower
• Cache Coherence