内存带宽、延迟性能测试
在内存方面,杰和GCR2524MP-RF服务器配置了128GB内存。通过SiSoftware Sandra 2011中提供了内存带宽、内存延迟和缓存性能的测试,来考量杰和GCR2524MP-RF服务器的在应用内存控制器技术下内存性能:
SiSoftware Sandra Lite 2011内存测试 | 内存带宽测试(越高越好) | Aggregate Memory Performance | 12.22GB/s |
Int Buff'd AVX 128 Memory Bandwidth | 12.17GB/s | ||
Float Buff'd FMA 256 Memory Bandwidth | 12.27GB/s | ||
内存延时测试(随机) | Memory(Random Access) Latency(越低越好) | 94.7ns | |
Speed Factor(越低越好) | 60.80 | ||
Internal Data Cache | 3clocks | ||
L2 On-board Cache | 10clocks | ||
L3 On-board Cache | 47clocks | ||
内存延时测试(线性) | Memory(Linear Access) Latency(越低越好) | 7.5ns | |
Speed Factor(越低越好) | 4.90 | ||
Internal Data Cache | 3clocks | ||
L2 On-board Cache | 10clocks | ||
L3 On-board Cache | 10clocks | ||
缓存及内存测试 | Cache/Memory Bandwidth(越高越好) | 206GB/s | |
Speed Factor(越低越好) | 155.00 | ||
Internal Data Cache | 1.32TB/s | ||
L2 On-board Cache | 601.73GB/s | ||
L3 On-board Cache | 217.25GB/s |
至强E5-4620 V4处理器将位架构升级为Broaswell,同时使用原有的v3插槽,继续使用DDR4内存,制造工艺从v3的22nm大步跨进了14nm。从测试结果上来看,尽管只配备128GB内存,杰和GCR2524MP-RF服务器也表现出了一定的性能提升。