The problem

Can workload coordination lower memory interference without sacrificing useful overlap?

What current research shows

Prior DRAM research shows that request order, bank-level parallelism, row locality, fairness, and interference affect memory-system outcomes. That establishes the measurement problem, not a result for Valen Systems' architecture.

Where the evidence stops

This is a hypothesis. No public result on this page should be read as proof that workload coordination improves every workload, memory device, or system.

What Valen Systems is testing

Run equal-work A/B tests under reviewed conditions. Measure row-buffer behavior where available, bank conflicts, queue depth, bandwidth, remote traffic, latency, energy, temperature, and work per joule.

Sources

Mutlu and Moscibroda: parallelism-aware DRAM schedulingStaged memory scheduling for heterogeneous systemsLinux scheduler debugfs: NUMA balancing