Arm's Cortex X925: Reaching Desktop Performance
3 hours ago
- #CPU
- #Performance
- #Arm
- Arm's Cortex X925 achieves performance parity with AMD's Zen 5 and Intel's Lion Cove in desktop applications.
- The Cortex X925 is a 10-wide core designed for maximum performance, with large reordering capacity and L2 cache comparable to Intel's P-Cores.
- X925 features advanced branch prediction, capable of recognizing long repeating patterns, similar to AMD's Zen 5.
- The core has a high instruction fetch and decode throughput, sustaining 10 instructions per cycle under optimal conditions.
- X925's out-of-order execution backend can handle around 525 instructions in flight, competing closely with Intel's Lion Cove and surpassing AMD's Zen 5.
- The core includes a sophisticated FPU with six pipes, each capable of handling vector floating point and integer operations.
- X925's load/store unit features four address generation units and a two-level TLB setup, with competitive bandwidth and latency.
- SPEC CPU2017 benchmarks show X925 performing on par with Intel and AMD in integer workloads but trailing slightly in floating point tests due to higher instruction counts.
- Arm's Cortex X925 represents a significant achievement, delivering desktop-level performance at a modest 4 GHz clock speed.
- Challenges remain, including gaming performance, software ecosystem, and reliance on partners to realize Arm's vision.