# SPEC CPU 2026 Rate

可通过[交互式图表](<https://jia.je/benchmark/viewer.html>)查看和筛选本文的测试数据。

## SPEC INT 2026 Rate-1

下面贴出自己测的数据（SPECint2026，Estimated，rate，base，1 copy），不保证满足 SPEC 的要求，仅供参考。

### 原始数据

#### Debian Forky

服务器平台（LTO + Jemalloc）：

- Loongson 3C6000D @ 2.1 GHz LA664（`-O3 -flto -ljemalloc`）: [1.74](<https://jia.je/benchmark/data-forky/int2026_rate1/Loongson_3C6000D_O3-flto-ljemalloc_001.txt>)

服务器平台（LTO）：

- Loongson 3C6000D @ 2.1 GHz LA664（`-O3 -flto`）: [1.66](<https://jia.je/benchmark/data-forky/int2026_rate1/Loongson_3C6000D_O3-flto_001.txt>)

服务器平台：

- Loongson 3C6000D @ 2.1 GHz LA664（`-O3`）: [1.61](<https://jia.je/benchmark/data-forky/int2026_rate1/Loongson_3C6000D_O3_001.txt>)

#### Debian Trixie

桌面平台（`-march=native` + LTO + Jemalloc）：

- AMD Ryzen 7 5700X @ 4.65 GHz Zen 3（`-O3 -march=native -flto -ljemalloc`）: [3.97](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_Ryzen_7_5700X_O3-march=native-flto-ljemalloc_001.txt>)
- Apple M1 P-Core @ 3.2 GHz Firestorm（`-O3 -march=native -flto -ljemalloc`）: [4.31](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M1_P-Core_O3-march=native-flto-ljemalloc_001.txt>)
- Apple M2 P-Core @ 3.5 GHz Avalanche（`-O3 -march=native -flto -ljemalloc`）: [5.20](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M2_P-Core_O3-march=native-flto-ljemalloc_001.txt>)
- Huawei Kirin X90 VM P-Core @ 2.3 GHz（`-O3 -march=native -flto -ljemalloc`）: [2.82](<https://jia.je/benchmark/data-trixie/int2026_rate1/Huawei_Kirin_X90_VM_P-Core_O3-march=native-flto-ljemalloc_001.txt>)
- Intel Core i7-13700K P-Core @ 5.4 GHz Raptor Cove（`-O3 -march=native -flto -ljemalloc`）: [5.56](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i7-13700K_P-Core_O3-march=native-flto-ljemalloc_001.txt>)
- Intel Core i9-12900KS E-Core @ 4.1 GHz Gracemont（`-O3 -march=native -flto -ljemalloc`）: [3.00](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_E-Core_O3-march=native-flto-ljemalloc_001.txt>)
- Intel Core i9-12900KS P-Core @ 5.5 GHz Golden Cove（`-O3 -march=native -flto -ljemalloc`）: [5.60](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_P-Core_O3-march=native-flto-ljemalloc_001.txt>)
- Intel Core i9-14900K P-Core @ 6.0 GHz Raptor Cove（`-O3 -march=native -flto -ljemalloc`）: [6.26](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_P-Core_O3-march=native-flto-ljemalloc_001.txt>)

桌面平台（LTO + Jemalloc）：

- AMD Ryzen 7 5700X @ 4.65 GHz Zen 3（`-O3 -flto -ljemalloc`）: [3.87](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_Ryzen_7_5700X_O3-flto-ljemalloc_001.txt>)
- Apple M1 P-Core @ 3.2 GHz Firestorm（`-O3 -flto -ljemalloc`）: [4.31](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M1_P-Core_O3-flto-ljemalloc_001.txt>)
- Apple M2 P-Core @ 3.5 GHz Avalanche（`-O3 -flto -ljemalloc`）: [5.10](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M2_P-Core_O3-flto-ljemalloc_001.txt>)
- Huawei Kirin X90 VM P-Core @ 2.3 GHz（`-O3 -flto -ljemalloc`）: [2.80](<https://jia.je/benchmark/data-trixie/int2026_rate1/Huawei_Kirin_X90_VM_P-Core_O3-flto-ljemalloc_001.txt>)
- Intel Core i5-1135G7 @ 4.2 GHz Willow Cove（`-O3 -flto -ljemalloc`）: [3.35](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i5-1135G7_O3-flto-ljemalloc_001.txt>)
- Intel Core i7-13700K P-Core @ 5.4 GHz Raptor Cove（`-O3 -flto -ljemalloc`）: [5.36](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i7-13700K_P-Core_O3-flto-ljemalloc_001.txt>)
- Intel Core i9-10980XE @ 4.7 GHz Cascade Lake（`-O3 -flto -ljemalloc`）: [3.17](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-10980XE_O3-flto-ljemalloc_001.txt>)
- Intel Core i9-12900KS E-Core @ 4.1 GHz Gracemont（`-O3 -flto -ljemalloc`）: [2.97](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_E-Core_O3-flto-ljemalloc_001.txt>)
- Intel Core i9-12900KS P-Core @ 5.5 GHz Golden Cove（`-O3 -flto -ljemalloc`）: [5.43](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_P-Core_O3-flto-ljemalloc_001.txt>)
- Intel Core i9-14900K E-Core @ 4.4 GHz Gracemont（`-O3 -flto -ljemalloc`）: [3.28](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_E-Core_O3-flto-ljemalloc_001.txt>)
- Intel Core i9-14900K P-Core @ 6.0 GHz Raptor Cove（`-O3 -flto -ljemalloc`）: [6.03](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_P-Core_O3-flto-ljemalloc_001.txt>)
- Intel Xeon w9-3595X @ 4.5 GHz Golden Cove（`-O3 -flto -ljemalloc`）: [4.45](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_w9-3595X_O3-flto-ljemalloc_001.txt>)

桌面平台（LTO）：

- AMD Ryzen 7 5700X @ 4.65 GHz Zen 3（`-O3 -flto`）: [3.66](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_Ryzen_7_5700X_O3-flto_001.txt>)
- Apple M1 P-Core @ 3.2 GHz Firestorm（`-O3 -flto`）: [4.10](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M1_P-Core_O3-flto_001.txt>)
- Apple M2 P-Core @ 3.5 GHz Avalanche（`-O3 -flto`）: [4.91](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M2_P-Core_O3-flto_001.txt>)
- Huawei Kirin X90 VM P-Core @ 2.3 GHz（`-O3 -flto`）: [2.63](<https://jia.je/benchmark/data-trixie/int2026_rate1/Huawei_Kirin_X90_VM_P-Core_O3-flto_001.txt>)
- Intel Core i5-1135G7 @ 4.2 GHz Willow Cove（`-O3 -flto`）: [3.14](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i5-1135G7_O3-flto_001.txt>)
- Intel Core i7-13700K P-Core @ 5.4 GHz Raptor Cove（`-O3 -flto`）: [5.07](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i7-13700K_P-Core_O3-flto_001.txt>)
- Intel Core i9-10980XE @ 4.7 GHz Cascade Lake（`-O3 -flto`）: [2.97](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-10980XE_O3-flto_001.txt>)
- Intel Core i9-12900KS E-Core @ 4.1 GHz Gracemont（`-O3 -flto`）: [2.79](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_E-Core_O3-flto_001.txt>)
- Intel Core i9-12900KS P-Core @ 5.5 GHz Golden Cove（`-O3 -flto`）: [5.11](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_P-Core_O3-flto_001.txt>)
- Intel Core i9-14900K E-Core @ 4.4 GHz Gracemont（`-O3 -flto`）: [3.10](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_E-Core_O3-flto_001.txt>)
- Intel Core i9-14900K P-Core @ 6.0 GHz Raptor Cove（`-O3 -flto`）: [5.71](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_P-Core_O3-flto_001.txt>)
- Intel Xeon w9-3595X @ 4.5 GHz Golden Cove（`-O3 -flto`）: [4.17](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_w9-3595X_O3-flto_001.txt>)

桌面平台：

- AMD Ryzen 7 5700X @ 4.65 GHz Zen 3（`-O3`）: [3.53](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_Ryzen_7_5700X_O3_001.txt>)
- Apple M1 P-Core @ 3.2 GHz Firestorm（`-O3`）: [4.02](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M1_P-Core_O3_001.txt>)
- Apple M2 P-Core @ 3.5 GHz Avalanche（`-O3`）: [4.76](<https://jia.je/benchmark/data-trixie/int2026_rate1/Apple_M2_P-Core_O3_001.txt>)
- Huawei Kirin X90 VM P-Core @ 2.3 GHz（`-O3`）: [2.54](<https://jia.je/benchmark/data-trixie/int2026_rate1/Huawei_Kirin_X90_VM_P-Core_O3_001.txt>)
- Intel Core i5-1135G7 @ 4.2 GHz Willow Cove（`-O3`）: [3.08](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i5-1135G7_O3_001.txt>)
- Intel Core i7-13700K P-Core @ 5.4 GHz Raptor Cove（`-O3`）: [4.96](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i7-13700K_P-Core_O3_001.txt>)
- Intel Core i9-10980XE @ 4.7 GHz Cascade Lake（`-O3`）: [2.90](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-10980XE_O3_001.txt>)
- Intel Core i9-12900KS E-Core @ 4.1 GHz Gracemont（`-O3`）: [2.73](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_E-Core_O3_001.txt>)
- Intel Core i9-12900KS P-Core @ 5.5 GHz Golden Cove（`-O3`）: [5.00](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-12900KS_P-Core_O3_001.txt>)
- Intel Core i9-14900K E-Core @ 4.4 GHz Gracemont（`-O3`）: [3.05](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_E-Core_O3_001.txt>)
- Intel Core i9-14900K P-Core @ 6.0 GHz Raptor Cove（`-O3`）: [5.59](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Core_i9-14900K_P-Core_O3_001.txt>)
- Intel Xeon w9-3595X @ 4.5 GHz Golden Cove（`-O3`）: [4.09](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_w9-3595X_O3_001.txt>)

服务器平台（`-march=native` + LTO + Jemalloc）：

- AMD EPYC 7551 @ 2.5 GHz Zen 1（`-O3 -march=native -flto -ljemalloc`）: [1.52](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7551_O3-march=native-flto-ljemalloc_001.txt>)
- AMD EPYC 7742 @ 3.4 GHz Zen 2（`-O3 -march=native -flto -ljemalloc`）: [2.50](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7742_O3-march=native-flto-ljemalloc_001.txt>)
- AWS Graviton 5 @ 3.3 GHz Neoverse V3（`-O3 -march=native -flto -ljemalloc`）: [4.67](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_5_O3-march=native-flto-ljemalloc_001.txt>)
- Intel Xeon E5-2680 v4 @ 3.3 GHz Broadwell（`-O3 -march=native -flto -ljemalloc`）: [2.17](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_E5-2680_v4_O3-march=native-flto-ljemalloc_001.txt>)
- Kunpeng 920 @ 2.6 GHz TaiShan V110（`-O3 -march=native -flto -ljemalloc`）: [1.59](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_O3-march=native-flto-ljemalloc_001.txt>)
- Kunpeng 920 HuaweiCloud kc2 @ 2.9 GHz（`-O3 -march=native -flto -ljemalloc`）: [2.68](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_HuaweiCloud_kc2_O3-march=native-flto-ljemalloc_001.txt>)

服务器平台（LTO + Jemalloc）：

- AMD EPYC 7551 @ 2.5 GHz Zen 1（`-O3 -flto -ljemalloc`）: [1.50](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7551_O3-flto-ljemalloc_001.txt>)
- AMD EPYC 7742 @ 3.4 GHz Zen 2（`-O3 -flto -ljemalloc`）: [2.41](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7742_O3-flto-ljemalloc_001.txt>)
- AMD EPYC 9T25 @ 4.1 GHz Zen 5（`-O3 -flto -ljemalloc`）: [4.34](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_9T25_O3-flto-ljemalloc_001.txt>)
- AWS Graviton 4 @ 2.8 GHz Neoverse V2（`-O3 -flto -ljemalloc`）: [3.48](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_4_O3-flto-ljemalloc_001.txt>)
- AWS Graviton 5 @ 3.3 GHz Neoverse V3（`-O3 -flto -ljemalloc`）: [4.56](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_5_O3-flto-ljemalloc_001.txt>)
- Hygon C86 4G 7447V @ 3.0 GHz Hygon-4G（`-O3 -flto -ljemalloc`）: [2.11](<https://jia.je/benchmark/data-trixie/int2026_rate1/Hygon_C86_4G_7447V_O3-flto-ljemalloc_001.txt>)
- Intel Xeon E5-2680 v4 @ 3.3 GHz Broadwell（`-O3 -flto -ljemalloc`）: [2.14](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_E5-2680_v4_O3-flto-ljemalloc_001.txt>)
- Intel Xeon Gold 6430 @ 2.6 GHz Golden Cove（`-O3 -flto -ljemalloc`）: [2.50](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_Gold_6430_O3-flto-ljemalloc_001.txt>)
- Intel Xeon Platinum 8358P @ 3.4 GHz Sunny Cove（`-O3 -flto -ljemalloc`）: [2.65](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_Platinum_8358P_O3-flto-ljemalloc_001.txt>)
- Kunpeng 920 @ 2.6 GHz TaiShan V110（`-O3 -flto -ljemalloc`）: [1.59](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_O3-flto-ljemalloc_001.txt>)
- Kunpeng 920 HuaweiCloud kc2 @ 2.9 GHz（`-O3 -flto -ljemalloc`）: [2.65](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_HuaweiCloud_kc2_O3-flto-ljemalloc_001.txt>)

服务器平台（LTO）：

- AMD EPYC 7551 @ 2.5 GHz Zen 1（`-O3 -flto`）: [1.42](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7551_O3-flto_001.txt>)
- AMD EPYC 7742 @ 3.4 GHz Zen 2（`-O3 -flto`）: [2.28](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7742_O3-flto_001.txt>)
- AMD EPYC 9T25 @ 4.1 GHz Zen 5（`-O3 -flto`）: [4.13](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_9T25_O3-flto_001.txt>)
- AWS Graviton 4 @ 2.8 GHz Neoverse V2（`-O3 -flto`）: [3.22](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_4_O3-flto_001.txt>)
- AWS Graviton 5 @ 3.3 GHz Neoverse V3（`-O3 -flto`）: [4.27](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_5_O3-flto_001.txt>)
- Hygon C86 4G 7447V @ 3.0 GHz Hygon-4G（`-O3 -flto`）: [1.99](<https://jia.je/benchmark/data-trixie/int2026_rate1/Hygon_C86_4G_7447V_O3-flto_001.txt>)
- Intel Xeon E5-2680 v4 @ 3.3 GHz Broadwell（`-O3 -flto`）: [2.06](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_E5-2680_v4_O3-flto_001.txt>)
- Intel Xeon Gold 6430 @ 2.6 GHz Golden Cove（`-O3 -flto`）: [2.37](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_Gold_6430_O3-flto_001.txt>)
- Intel Xeon Platinum 8358P @ 3.4 GHz Sunny Cove（`-O3 -flto`）: [2.59](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_Platinum_8358P_O3-flto_001.txt>)
- Kunpeng 920 @ 2.6 GHz TaiShan V110（`-O3 -flto`）: [1.51](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_O3-flto_001.txt>)
- Kunpeng 920 HuaweiCloud kc2 @ 2.9 GHz（`-O3 -flto`）: [2.51](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_HuaweiCloud_kc2_O3-flto_001.txt>)

服务器平台：

- AMD EPYC 7551 @ 2.5 GHz Zen 1（`-O3`）: [1.37](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7551_O3_001.txt>)
- AMD EPYC 7742 @ 3.4 GHz Zen 2（`-O3`）: [2.19](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_7742_O3_001.txt>)
- AMD EPYC 9T25 @ 4.1 GHz Zen 5（`-O3`）: [3.99](<https://jia.je/benchmark/data-trixie/int2026_rate1/AMD_EPYC_9T25_O3_001.txt>)
- AWS Graviton 4 @ 2.8 GHz Neoverse V2（`-O3`）: [3.15](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_4_O3_001.txt>)
- AWS Graviton 5 @ 3.3 GHz Neoverse V3（`-O3`）: [4.21](<https://jia.je/benchmark/data-trixie/int2026_rate1/AWS_Graviton_5_O3_001.txt>)
- Hygon C86 4G 7447V @ 3.0 GHz Hygon-4G（`-O3`）: [1.93](<https://jia.je/benchmark/data-trixie/int2026_rate1/Hygon_C86_4G_7447V_O3_001.txt>)
- Intel Xeon E5-2680 v4 @ 3.3 GHz Broadwell（`-O3`）: [2.01](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_E5-2680_v4_O3_001.txt>)
- Intel Xeon Gold 6430 @ 2.6 GHz Golden Cove（`-O3`）: [2.32](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_Gold_6430_O3_001.txt>)
- Intel Xeon Platinum 8358P @ 3.4 GHz Sunny Cove（`-O3`）: [2.54](<https://jia.je/benchmark/data-trixie/int2026_rate1/Intel_Xeon_Platinum_8358P_O3_001.txt>)
- Kunpeng 920 @ 2.6 GHz TaiShan V110（`-O3`）: [1.46](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_O3_001.txt>)
- Kunpeng 920 HuaweiCloud kc2 @ 2.9 GHz（`-O3`）: [2.42](<https://jia.je/benchmark/data-trixie/int2026_rate1/Kunpeng_920_HuaweiCloud_kc2_O3_001.txt>)

#### 备注

1. 对于除了苹果以外的 ARM64 核心，内核的 branch-misses 计数器考虑了 speculative 而不只是 retired，因此数字会偏高，此时要用 r22 计数替代。

### 分支预测器比较

#### Debian Trixie

x86 平台的分支预测准确率（Average）由高到低（`-O3`）：

1. Golden Cove(Intel 12900KS P-Core)/Raptor Cove(Intel 13700K P-Core/Intel 14900K P-Core): MPKI=2.13
2. Gracemont(Intel 12900KS E-Core/Intel 14900K E-Core): MPKI=2.44
3. Cascade Lake(Intel 10980XE): MPKI=2.70
4. Broadwell(Intel E5-2680 v4): MPKI=2.94
5. Zen 1(AMD 7551): MPKI=3.73

ARM64 平台的分支预测准确率（Average）由高到低（`-O3`）：

1. Neoverse V3(AWS Graviton 5): MPKI=1.83
2. Avalanche(Apple M2 P-Core): MPKI=2.04
3. Firestorm(Apple M1 P-Core): MPKI=2.80
4. TSV110(Hisilicon Kunpeng 920): MPKI=4.07

### 网上的数据

来自 SPEC 官网：

| Machine | Score | Compilation Flags |
| --- | --- | --- |
| [Ampere eMAG 8180](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260210-00003.html>) | 1.00 | \-O3 -ffast-math -mcpu=native -flto=16 -ljemalloc |
| [AMD Ryzen AI 9 HX 370](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260210-00018.html>) | 5.00 | \-O3 -ffast-math -march=native -flto=full |
| [NVIDIA GB10](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260210-00020.html>) | 5.97 | \-fuse-ld=lld -O3 -ffast-math -mcpu=native -flto=thin -fomit-frame-pointer |
| [Apple M5 Pro](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260422-00243.html>) | 7.64 | \-O3 |

另见 [David Huang](<https://benchview.hjc.im>)、[Chips and Cheese](<https://chipsandcheese.com/p/evaluating-spec-cpu2026>)。

### 多版本 GCC 和 LLVM 性能比较

在 Intel i9-12900KS @ 5.5 GHz 上用 -O3 测试几种编译器组合的 SPEC INT 2026 Rate-1 性能：

| Benchmark | GCC 16 | GCC 15 | GCC 14 | LLVM 22 | LLVM 21 | LLVM 20 | LLVM 19 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 706.stockfish\_r | 8.37 | **8.45** | 6.13 | 6.12 | 6.11 | 6.07 | 5.82 |
| 707.ntest\_r | 4.37 | 4.38 | 4.03 | **4.52** | 4.49 | 4.49 | 4.49 |
| 708.sqlite\_r | **4.53** | 4.35 | 4.52 | 4.33 | 4.26 | 4.30 | 4.18 |
| 710.omnetpp\_r | **5.47** | 5.43 | 5.34 | 4.98 | 4.99 | 4.97 | 4.96 |
| 714.cpython\_r | 6.53 | **6.62** | 6.40 | 5.74 | 5.78 | 5.72 | 5.72 |
| 721.gcc\_r | **5.53** | 5.52 | 5.43 | 5.31 | 5.30 | 5.28 | 5.28 |
| 723.llvm\_r | **4.18** | 4.14 | 4.09 | 4.05 | 4.05 | 4.03 | 4.01 |
| 727.cppcheck\_r | 3.80 | **3.93** | 3.83 | 3.78 | 3.90 | 3.89 | 3.71 |
| 729.abc\_r | 4.54 | **4.57** | 4.51 | 4.45 | 4.43 | 4.41 | 4.40 |
| 734.vpr\_r | **5.02** | 5.01 | 4.90 | 4.85 | 4.86 | 4.83 | 4.71 |
| 735.gem5\_r | **5.52** | 5.33 | 5.24 | 5.27 | 5.25 | 5.14 | 5.18 |
| 750.sealcrypto\_r | 4.86 | 4.88 | 4.82 | **10.2** | 4.99 | 4.98 | 4.93 |
| 753.ns3\_r | 6.56 | **6.59** | 6.38 | 6.30 | 6.36 | 6.30 | 6.25 |
| 777.zstd\_r | **4.69** | 4.42 | 4.41 | 4.30 | 4.27 | 4.29 | 4.26 |
| geomean | **5.17** | 5.14 | 4.94 | 5.13 | 4.88 | 4.86 | 4.80 |

完整数据：

- [GCC 14.2.0](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_GCC_14.txt>)
- [GCC 15.2.0](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_GCC_15.txt>)
- [GCC 16.1.0](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_GCC_16.txt>)
- [LLVM 19.1.7](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_LLVM_19.txt>)
- [LLVM 20.1.8](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_LLVM_20.txt>)
- [LLVM 21.1.8](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_LLVM_21.txt>)
- [LLVM 22.1.6](<https://jia.je/benchmark/data-trixie/others/SPEC_INT_2026_Intel_i9-12900KS_O3_LLVM_22.txt>)

## SPEC FP 2026 Rate-1

下面贴出自己测的数据（SPECfp2026，Estimated，rate，base，1 copy），不保证满足 SPEC 的要求，仅供参考。

### 原始数据

#### Debian Trixie

桌面平台（`-march=native`）：

- Apple M1 P-Core @ 3.2 GHz Firestorm（`-O3 -march=native`）: [5.97](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Apple_M1_P-Core_O3-march=native_001.txt>)
- Apple M2 P-Core @ 3.5 GHz Avalanche（`-O3 -march=native`）: [6.73](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Apple_M2_P-Core_O3-march=native_001.txt>)
- Huawei Kirin X90 VM P-Core @ 2.3 GHz（`-O3 -march=native`）: [3.95](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Huawei_Kirin_X90_VM_P-Core_O3-march=native_001.txt>)
- Intel Core i5-1135G7 @ 4.2 GHz Willow Cove（`-O3 -march=native`）: [5.22](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i5-1135G7_O3-march=native_001.txt>)
- Intel Core i7-13700K P-Core @ 5.4 GHz Raptor Cove（`-O3 -march=native`）: [7.96](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i7-13700K_P-Core_O3-march=native_001.txt>)
- Intel Core i9-10980XE @ 4.7 GHz (AVX-512 @ 4.0 GHz) Cascade Lake（`-O3 -march=native`）: [4.22](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-10980XE_O3-march=native_001.txt>)
- Intel Core i9-12900KS E-Core @ 4.1 GHz Gracemont（`-O3 -march=native`）: [3.99](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-12900KS_E-Core_O3-march=native_001.txt>)
- Intel Core i9-12900KS P-Core @ 5.5 GHz Golden Cove（`-O3 -march=native`）: [8.27](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-12900KS_P-Core_O3-march=native_001.txt>)
- Intel Core i9-14900K E-Core @ 4.4 GHz Gracemont（`-O3 -march=native`）: [4.19](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-14900K_E-Core_O3-march=native_001.txt>)
- Intel Core i9-14900K P-Core @ 6.0 GHz Raptor Cove（`-O3 -march=native`）: [9.36](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-14900K_P-Core_O3-march=native_001.txt>)
- Intel Xeon w9-3595X @ 4.5 GHz Golden Cove（`-O3 -march=native`）: [6.34](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_w9-3595X_O3-march=native_001.txt>)

桌面平台：

- Apple M1 P-Core @ 3.2 GHz Firestorm（`-O3`）: [5.80](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Apple_M1_P-Core_O3_001.txt>)
- Apple M2 P-Core @ 3.5 GHz Avalanche（`-O3`）: [6.38](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Apple_M2_P-Core_O3_001.txt>)
- Huawei Kirin X90 VM P-Core @ 2.3 GHz（`-O3`）: [3.76](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Huawei_Kirin_X90_VM_P-Core_O3_001.txt>)
- Intel Core i5-1135G7 @ 4.2 GHz Willow Cove（`-O3`）: [4.11](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i5-1135G7_O3_001.txt>)
- Intel Core i7-13700K P-Core @ 5.4 GHz Raptor Cove（`-O3`）: [6.32](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i7-13700K_P-Core_O3_001.txt>)
- Intel Core i9-10980XE @ 4.7 GHz Cascade Lake（`-O3`）: [3.55](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-10980XE_O3_001.txt>)
- Intel Core i9-12900KS E-Core @ 4.1 GHz Gracemont（`-O3`）: [3.47](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-12900KS_E-Core_O3_001.txt>)
- Intel Core i9-12900KS P-Core @ 5.5 GHz Golden Cove（`-O3`）: [6.70](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-12900KS_P-Core_O3_001.txt>)
- Intel Core i9-14900K E-Core @ 4.4 GHz Gracemont（`-O3`）: [3.66](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-14900K_E-Core_O3_001.txt>)
- Intel Core i9-14900K P-Core @ 6.0 GHz Raptor Cove（`-O3`）: [7.44](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Core_i9-14900K_P-Core_O3_001.txt>)
- Intel Xeon w9-3595X @ 4.5 GHz Golden Cove（`-O3`）: [5.01](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_w9-3595X_O3_001.txt>)

服务器平台（`-march=native`）：

- AMD EPYC 7551 @ 2.5 GHz Zen 1（`-O3 -march=native`）: [2.04](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AMD_EPYC_7551_O3-march=native_001.txt>)
- AMD EPYC 7742 @ 3.4 GHz Zen 2（`-O3 -march=native`）: [3.88](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AMD_EPYC_7742_O3-march=native_001.txt>)
- AMD EPYC 9T25 @ 4.1 GHz Zen 5（`-O3 -march=native`）: [7.59](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AMD_EPYC_9T25_O3-march=native_001.txt>)
- AWS Graviton 4 @ 2.8 GHz Neoverse V2（`-O3 -march=native`）: [4.79](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AWS_Graviton_4_O3-march=native_001.txt>)
- AWS Graviton 5 @ 3.3 GHz Neoverse V3（`-O3 -march=native`）: [6.10](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AWS_Graviton_5_O3-march=native_001.txt>)
- Hygon C86 4G 7447V @ 3.0 GHz Hygon-4G（`-O3 -march=native`）: [2.93](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Hygon_C86_4G_7447V_O3-march=native_001.txt>)
- Intel Xeon E5-2680 v4 @ 3.3 GHz Broadwell（`-O3 -march=native`）: [2.74](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_E5-2680_v4_O3-march=native_001.txt>)
- Intel Xeon Gold 6430 @ 2.6 GHz Golden Cove（`-O3 -march=native`）: [3.86](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_Gold_6430_O3-march=native_001.txt>)
- Intel Xeon Platinum 8358P @ 3.4 GHz Sunny Cove（`-O3 -march=native`）: [4.16](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_Platinum_8358P_O3-march=native_001.txt>)
- Kunpeng 920 @ 2.6 GHz TaiShan V110（`-O3 -march=native`）: [1.62](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Kunpeng_920_O3-march=native_001.txt>)
- Kunpeng 920 HuaweiCloud kc2 @ 2.9 GHz（`-O3 -march=native`）: [3.69](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Kunpeng_920_HuaweiCloud_kc2_O3-march=native_001.txt>)

服务器平台：

- AMD EPYC 7551 @ 2.5 GHz Zen 1（`-O3`）: [1.86](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AMD_EPYC_7551_O3_001.txt>)
- AMD EPYC 7742 @ 3.4 GHz Zen 2（`-O3`）: [3.25](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AMD_EPYC_7742_O3_001.txt>)
- AMD EPYC 9T25 @ 4.1 GHz Zen 5（`-O3`）: [5.63](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AMD_EPYC_9T25_O3_001.txt>)
- AWS Graviton 4 @ 2.8 GHz Neoverse V2（`-O3`）: [4.50](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AWS_Graviton_4_O3_001.txt>)
- AWS Graviton 5 @ 3.3 GHz Neoverse V3（`-O3`）: [5.67](<https://jia.je/benchmark/data-trixie/fp2026_rate1/AWS_Graviton_5_O3_001.txt>)
- Hygon C86 4G 7447V @ 3.0 GHz Hygon-4G（`-O3`）: [2.21](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Hygon_C86_4G_7447V_O3_001.txt>)
- Intel Xeon E5-2680 v4 @ 3.3 GHz Broadwell（`-O3`）: [2.43](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_E5-2680_v4_O3_001.txt>)
- Intel Xeon Gold 6430 @ 2.6 GHz Golden Cove（`-O3`）: [3.01](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_Gold_6430_O3_001.txt>)
- Intel Xeon Platinum 8358P @ 3.4 GHz Sunny Cove（`-O3`）: [3.22](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Intel_Xeon_Platinum_8358P_O3_001.txt>)
- Kunpeng 920 @ 2.6 GHz TaiShan V110（`-O3`）: [1.54](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Kunpeng_920_O3_001.txt>)
- Kunpeng 920 HuaweiCloud kc2 @ 2.9 GHz（`-O3`）: [3.78](<https://jia.je/benchmark/data-trixie/fp2026_rate1/Kunpeng_920_HuaweiCloud_kc2_O3_001.txt>)

#### 备注

1. 对于除了苹果以外的 ARM64 核心，内核的 branch-misses 计数器考虑了 speculative 而不只是 retired，因此数字会偏高，此时要用 r22 计数替代。

### 网上的数据

来自 SPEC 官网：

| Machine | Score | Compilation Flags |
| --- | --- | --- |
| [Ampere eMAG 8180](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260210-00001.html>) | 1.00 | \-O3 -ffast-math -mcpu=native -flto=16 -ljemalloc |
| [AMD Ryzen AI 9 HX 370](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260210-00017.html>) | 8.59 | \-O3 -ffast-math -march=native -flto=full |
| [NVIDIA GB10](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260210-00019.html>) | 9.70 | \-fuse-ld=lld -O3 -ffast-math -mcpu=native -flto=thin -fomit-frame-pointer |
| [Apple M5 Pro](<https://www.spec.org/cpu2026/results/res2026q2/cpu2026-20260422-00245.html>) | 10.9 | \-O3 |

另见 [David Huang](<https://benchview.hjc.im>)、[Chips and Cheese](<https://chipsandcheese.com/p/evaluating-spec-cpu2026>)。

### 多版本 GCC 和 LLVM 性能比较

在 Intel i9-12900KS @ 5.5 GHz 上用 -O3 测试几种编译器组合的 SPEC FP 2026 Rate-1 性能：

| Benchmark | GCC 16 | GCC 15 | GCC 14 | LLVM 22 | LLVM 21 | LLVM 20 |
| --- | --- | --- | --- | --- | --- | --- |
| 709.cactus\_r | 6.77 | 7.52 | 7.62 | 8.28 | **8.29** | 8.16 |
| 722.palm\_r | 7.79 | 7.81 | 7.88 | **9.20** | 9.00 | 8.44 |
| 731.astcenc\_r | 4.77 | 4.76 | 4.60 | 5.96 | 5.84 | **6.16** |
| 736.ocio\_r | 6.20 | 6.25 | 6.26 | 6.81 | **6.87** | 6.84 |
| 737.gmsh\_r | **4.55** | 4.53 | 4.50 | 4.40 | 4.38 | 4.40 |
| 748.flightdm\_r | **7.76** | 7.54 | 7.42 | 7.05 | 7.09 | 7.11 |
| 749.fotonik3d\_r | 6.81 | 6.82 | 6.82 | 6.93 | **6.99** | 6.76 |
| 765.roms\_r | 7.50 | 7.39 | 7.40 | **7.52** | 7.51 | 7.38 |
| 766.femflow\_r | 9.92 | 7.97 | 8.16 | **12.1** | **12.1** | **12.1** |
| 767.nest\_r | 7.17 | **8.60** | 8.50 | 8.15 | 8.18 | 8.15 |
| 772.marian\_r | **10.3** | **10.3** | 6.58 | 4.78 | 4.78 | 4.78 |
| 782.lbm\_r | **4.81** | **4.81** | 4.80 | 4.49 | 4.49 | 4.47 |
| geomean | 6.81 | **6.83** | 6.56 | 6.84 | **6.83** | 6.79 |

完整数据：

- [GCC 14.2.0](<https://jia.je/benchmark/data-trixie/others/SPEC_FP_2026_Intel_i9-12900KS_O3_GCC_14.txt>)
- [GCC 15.2.0](<https://jia.je/benchmark/data-trixie/others/SPEC_FP_2026_Intel_i9-12900KS_O3_GCC_15.txt>)
- [GCC 16.1.0](<https://jia.je/benchmark/data-trixie/others/SPEC_FP_2026_Intel_i9-12900KS_O3_GCC_16.txt>)
- [LLVM 20.1.8](<https://jia.je/benchmark/data-trixie/others/SPEC_FP_2026_Intel_i9-12900KS_O3_LLVM_20.txt>)
- [LLVM 21.1.8](<https://jia.je/benchmark/data-trixie/others/SPEC_FP_2026_Intel_i9-12900KS_O3_LLVM_21.txt>)
- [LLVM 22.1.6](<https://jia.je/benchmark/data-trixie/others/SPEC_FP_2026_Intel_i9-12900KS_O3_LLVM_22.txt>)