我正在测试一台配备 Xeon 6130 CPU 的新服务器，并尝试将其与配备 E5-2660v4 的旧服务器进行比较。以下是我刚刚在每个服务器上运行的测试：
在 6130 上：
# stress-ng --matrix 40 -t 30s --times --perf --tz --metrics-brief
stress-ng: info: [12734] dispatching hogs: 40 matrix
stress-ng: info: [12734] cache allocate: default cache size: 22528K
stress-ng: info: [12734] successful run completed in 30.66s
stress-ng: info: [12734] stressor bogo ops real time usr time sys time bogo ops/s bogo ops/s
stress-ng: info: [12734] (secs) (secs) (secs) (real time) (usr+sys time)
stress-ng: info: [12734] matrix 2057213 30.02 1187.60 1.55 68534.05 1729.99
stress-ng: info: [12734] matrix:
stress-ng: info: [12734] 2,418,977,858,680 CPU Cycles 78.88 B/sec
stress-ng: info: [12734] 3,616,346,898,160 Instructions 0.12 T/sec (1.495 instr. per cycle)
stress-ng: info: [12734] 216,338,520 Cache References 7.05 M/sec
stress-ng: info: [12734] 30,689,400 Cache Misses 1.00 M/sec (14.19%)
stress-ng: info: [12734] 457,662,770,480 Branch Instructions 14.92 B/sec
stress-ng: info: [12734] 3,541,729,200 Branch Misses 0.12 B/sec ( 0.77%)
stress-ng: info: [12734] 29,326,176,200 Bus Cycles 0.96 B/sec
stress-ng: info: [12734] 2,463,405,104,320 Total Cycles 80.33 B/sec
stress-ng: info: [12734] 6,240 Page Faults Minor 203.49 sec
stress-ng: info: [12734] 0 Page Faults Major 0.00 sec
stress-ng: info: [12734] 103,360 Context Switches 3.37 K/sec
stress-ng: info: [12734] 1,280 CPU Migrations 41.74 sec
stress-ng: info: [12734] 0 Alignment Faults 0.00 sec
stress-ng: info: [12734] 2,360 System Call Enter 76.96 sec
stress-ng: info: [12734] 2,320 System Call Exit 75.66 sec
stress-ng: info: [12734] 3,480 TLB Flushes 113.49 sec
stress-ng: info: [12734] 920 Kmalloc 30.00 sec
stress-ng: info: [12734] 120 Kmalloc Node 3.91 sec
stress-ng: info: [12734] 840 Kfree 27.39 sec
stress-ng: info: [12734] 200 Kmem Cache Alloc 6.52 sec
stress-ng: info: [12734] 120 Kmem Cache Alloc Node 3.91 sec
stress-ng: info: [12734] 4,400 Kmem Cache Free 143.49 sec
stress-ng: info: [12734] 4,080 MM Page Alloc 133.05 sec
stress-ng: info: [12734] 2,160 MM Page Free 70.44 sec
stress-ng: info: [12734] 993,760 RCU Utilization 32.41 K/sec
stress-ng: info: [12734] 3,400 Sched Migrate Task 110.88 sec
stress-ng: info: [12734] 0 Sched Move NUMA 0.00 sec
stress-ng: info: [12734] 112,120 Sched Wakeup 3.66 K/sec
stress-ng: info: [12734] 0 Signal Generate 0.00 sec
stress-ng: info: [12734] 40 Signal Deliver 1.30 sec
stress-ng: info: [12734] 6,960 IRQ Entry 226.97 sec
stress-ng: info: [12734] 6,960 IRQ Exit 226.97 sec
stress-ng: info: [12734] 301,680 Soft IRQ Entry 9.84 K/sec
stress-ng: info: [12734] 301,680 Soft IRQ Exit 9.84 K/sec
stress-ng: info: [12734] 0 Writeback Dirty Inode 0.00 sec
stress-ng: info: [12734] 0 Writeback Dirty Page 0.00 sec
stress-ng: info: [12734] matrix:
stress-ng: info: [12734] x86_pkg_temp 53.35 °C
stress-ng: info: [12734] x86_pkg_temp 57.09 °C
stress-ng: info: [12734] for a 30.66s run time:
stress-ng: info: [12734] 1962.54s available CPU time
stress-ng: info: [12734] 1187.83s user time ( 60.53%)
stress-ng: info: [12734] 1.83s system time ( 0.09%)
stress-ng: info: [12734] 1189.66s total time ( 60.62%)
stress-ng: info: [12734] load average: 34.39 30.68 28.86
在 E5-2660v4 上：
# stress-ng --matrix 40 -t 30s --times --perf --tz --metrics-brief
stress-ng: info: [130239] dispatching hogs: 40 matrix
stress-ng: info: [130239] cache allocate: default cache size: 35840K
stress-ng: info: [130239] successful run completed in 30.45s
stress-ng: info: [130239] stressor bogo ops real time usr time sys time bogo ops/s bogo ops/s
stress-ng: info: [130239] (secs) (secs) (secs) (real time) (usr+sys time)
stress-ng: info: [130239] matrix 2996186 30.01 1196.39 0.32 99836.74 2503.69
stress-ng: info: [130239] matrix:
stress-ng: info: [130239] 2,857,631,562,680 CPU Cycles 93.86 B/sec
stress-ng: info: [130239] 4,989,313,222,360 Instructions 0.16 T/sec (1.746 instr. per cycle)
stress-ng: info: [130239] 2,706,075,120 Cache References 88.88 M/sec
stress-ng: info: [130239] 6,707,600 Cache Misses 0.22 M/sec ( 0.25%)
stress-ng: info: [130239] 630,738,079,720 Branch Instructions 20.72 B/sec
stress-ng: info: [130239] 4,830,368,040 Branch Misses 0.16 B/sec ( 0.77%)
stress-ng: info: [130239] 119,665,830,840 Bus Cycles 3.93 B/sec
stress-ng: info: [130239] 2,393,336,537,400 Total Cycles 78.61 B/sec
stress-ng: info: [130239] 22,440 Page Faults Minor 737.03 sec
stress-ng: info: [130239] 0 Page Faults Major 0.00 sec
stress-ng: info: [130239] 7,640 Context Switches 250.93 sec
stress-ng: info: [130239] 2,720 CPU Migrations 89.34 sec
stress-ng: info: [130239] 0 Alignment Faults 0.00 sec
stress-ng: info: [130239] 22,440 Page Faults User 737.03 sec
stress-ng: info: [130239] 0 Page Faults Kernel 0.00 sec
stress-ng: info: [130239] 2,440 System Call Enter 80.14 sec
stress-ng: info: [130239] 2,400 System Call Exit 78.83 sec
stress-ng: info: [130239] 1,920 TLB Flushes 63.06 sec
stress-ng: info: [130239] 40 Kmalloc 1.31 sec
stress-ng: info: [130239] 0 Kmalloc Node 0.00 sec
stress-ng: info: [130239] 120 Kfree 3.94 sec
stress-ng: info: [130239] 0 Kmem Cache Alloc 0.00 sec
stress-ng: info: [130239] 0 Kmem Cache Alloc Node 0.00 sec
stress-ng: info: [130239] 2,480 Kmem Cache Free 81.45 sec
stress-ng: info: [130239] 1,920 MM Page Alloc 63.06 sec
stress-ng: info: [130239] 0 MM Page Free 0.00 sec
stress-ng: info: [130239] 915,040 RCU Utilization 30.05 K/sec
stress-ng: info: [130239] 2,200 Sched Migrate Task 72.26 sec
stress-ng: info: [130239] 0 Sched Move NUMA 0.00 sec
stress-ng: info: [130239] 8,000 Sched Wakeup 262.76 sec
stress-ng: info: [130239] 0 Signal Generate 0.00 sec
stress-ng: info: [130239] 40 Signal Deliver 1.31 sec
stress-ng: info: [130239] 0 IRQ Entry 0.00 sec
stress-ng: info: [130239] 0 IRQ Exit 0.00 sec
stress-ng: info: [130239] 460,840 Soft IRQ Entry 15.14 K/sec
stress-ng: info: [130239] 460,840 Soft IRQ Exit 15.14 K/sec
stress-ng: info: [130239] 0 Writeback Dirty Inode 0.00 sec
stress-ng: info: [130239] 0 Writeback Dirty Page 0.00 sec
stress-ng: info: [130239] matrix:
stress-ng: info: [130239] x86_pkg_temp 53.85 °C
stress-ng: info: [130239] x86_pkg_temp 52.91 °C
stress-ng: info: [130239] for a 30.45s run time:
stress-ng: info: [130239] 1705.01s available CPU time
stress-ng: info: [130239] 1196.57s user time ( 70.18%)
stress-ng: info: [130239] 0.48s system time ( 0.03%)
stress-ng: info: [130239] 1197.05s total time ( 70.21%)
stress-ng: info: [130239] load average: 19.53 19.17 14.00
Skylake CPU 的得分为 68k bogo ops/s。较旧的 Broadwell CPU 的得分为 ~100k bogo ops/s。跨 CPU 比较这些值是否公平？6130 的 L3 缓存大约是 E5-2660v4 的 2/3。这就是为什么它的得分为 2660v4 的 ~70% 的原因吗？
答案1
简而言之，不是。
在其他条件相同的情况下，如果是较新、功耗更大、内核更多的处理器 (6130)，您的分数会更高。作为参考，在 PassMark 中，Xeon E5-2660 v4 得分为 17786 分，而 Xeon 6130 Gold 得分为 20222 分，如下所示：https://www.cpubenchmark.net/compare/Intel-Xeon-E5-2660-v4-vs-Intel-Xeon-Gold-6130/2881vs3126。
您永远不应该相信其他基准测试；相反，您应该创建自己的一套基准测试来将硬件配置与现实世界的情况进行比较，因为我怀疑您是否整天使用这些基准测试来执行压力测试。