XPerf and Akash Systems tie Diamond Cooling to auditable AI server gains

XPerf and Akash Systems said June 9 they are collaborating to measure the energy and thermal impact of Diamond Cooling on AMD Instinct MI350X AI servers. The companies say the rack-level validation can document efficiency gains, temperature reductions and throughput improvements before customers run production workloads.

Why it matters: - AI server buyers are under pressure to prove power efficiency, thermal headroom and workload performance before deployment. - The collaboration aims to turn cooling claims into rack-level evidence that customers can audit at delivery. - Akash Systems says that could translate into lower operating costs, fewer throttling issues and faster time to revenue.

What happened: - Akash Systems and XPerf announced a collaboration on June 9 to validate the energy and thermal impact of Diamond Cooling on AMD Instinct MI350X AI servers. - XPerf is using ClusterReady as the rack-level validation platform. - The companies say the setup provides documented proof of watt-per-token efficiency and tokens-per-second gains before production workloads run.

The details: - Akash Systems says its patented Diamond Cooling removes heat five times faster than copper. - The cooling technology is designed to work alongside air and liquid cooling, including heat sinks and cold plates. - On AMD Instinct MI350X servers, the companies say ClusterReady-validated Diamond Cooled systems can deliver up to 10°C lower GPU temperatures. - The companies say the systems can produce a 22% FLOPs-per-watt improvement in standard ambient data center temperatures. - The companies say they can improve token-per-second throughput by up to 15% in high-ambient environments. - The companies say the servers can sustain throttle-free performance at 120°F ambient inlet temperatures. - Akash Systems says the cooling gains can create up to $1 million in incremental cash flow per server over the system lifetime. - XPerf says ClusterReady exercises the full rack under AI workloads and surfaces system-level behavior that component testing can miss. - XPerf says the platform checks interconnect bandwidth, RCCL communication throughput and sustained-load thermal performance. - ClusterReady uses a three-stage pipeline: Pre-Flight Health Check, AI Workload Validation and Stress Testing. - XPerf says the process tests GPUs, memory, interconnects and network fabric under production-like conditions. - XPerf says its validation can cut full-rack testing from weeks to hours or days. - XPerf says each rack ships with a validation report that documents tokens-per-second output for inference and training workloads. - XPerf says the test suite can detect RCCL bandwidth degradation, thermal-induced throttling and interconnect misconfiguration. - XPerf says its algorithms also provide remediation and optimization guidance for detected issues. - The release cites prior XPerf work with MiTAC Computing on AMD MI325X racks, where the company said ClusterReady showed 20% higher performance than AMD’s published FLUX.1-dev benchmark results, 95.8% scaling efficiency from one node to four nodes and 33,695 tokens per second on Llama 2 70B single-node inference.

Between the lines: - The partnership is as much about proof as performance. - Rack-level validation gives Akash Systems a way to attach measured outcomes to its cooling claims instead of relying on component-level specs. - The release frames the biggest benefit as auditable evidence that customers can use to compare deployments and estimate returns. - The forward-looking language also signals that MI350X results may vary by configuration, software and workload.

What’s next: - ClusterReady is commercially available for supported environments. - Akash Systems and XPerf are positioning the workflow as a pre-deployment validation step for enterprise customers. - The companies say they intend to extend the collaboration to future GPU architectures as diamond cooling scales across the AI data center market. - Customers can request a ClusterReady validation demo through more information or learn more about Akash Systems at the company’s announcement.

The bottom line: - XPerf and Akash Systems are trying to make cooling performance measurable, auditable and tied to AI workload outcomes before a server ever enters production.