Intel M6i Ice Lake vs. Graviton2 Amazon EC2 Benchmarks Amazon EC2 benchmarking for a future article. m6g.metal: Processor: ARMv8 Neoverse-N1 (64 Cores), Motherboard: Amazon EC2 m6g.metal v1.0, Memory: 252GB, Disk: 107GB Amazon Elastic Block Store, Network: Amazon Elastic OS: Ubuntu 20.04, Kernel: 5.4.0-1045-aws (aarch64), Vulkan: 1.0.2, Compiler: GCC 9.3.0, File-System: ext4 m5.24xlarge: Processor: 2 x Intel Xeon Platinum 8259CL (48 Cores / 96 Threads), Motherboard: Amazon EC2 m5.24xlarge (1.0 BIOS), Chipset: Intel 440FX 82441FX PMC, Memory: 374GB, Disk: 107GB Amazon Elastic Block Store, Network: Amazon Elastic OS: Ubuntu 20.04, Kernel: 5.4.0-1045-aws (x86_64), Vulkan: 1.0.2, Compiler: GCC 9.3.0, File-System: ext4, System Layer: KVM m6i.24xlarge: Processor: 2 x Intel Xeon Platinum 8375C (48 Cores / 96 Threads), Motherboard: Amazon EC2 m6i.24xlarge (1.0 BIOS), Chipset: Intel 440FX 82441FX PMC, Memory: 372GB, Disk: 107GB Amazon Elastic Block Store, Network: Amazon Elastic OS: Ubuntu 20.04, Kernel: 5.4.0-1045-aws (x86_64), Vulkan: 1.0.2, Compiler: GCC 9.3.0, File-System: ext4, System Layer: KVM m6i.32xlarge: Processor: 2 x Intel Xeon Platinum 8375C (64 Cores / 128 Threads), Motherboard: Amazon EC2 m6i.32xlarge (1.0 BIOS), Chipset: Intel 440FX 82441FX PMC, Memory: 496GB, Disk: 107GB Amazon Elastic Block Store, Network: Amazon Elastic OS: Ubuntu 20.04, Kernel: 5.4.0-1045-aws (x86_64), Vulkan: 1.0.2, Compiler: GCC 9.3.0, File-System: ext4, System Layer: KVM TNN 0.3 Target: CPU - Model: DenseNet ms < Lower Is Better m6g.metal .... 3288.83 |================================================ m5.24xlarge .. 3797.59 |======================================================= m6i.24xlarge . 3522.58 |=================================================== m6i.32xlarge . 3524.69 |=================================================== asmFish 2018-07-23 1024 Hash Memory, 26 Depth Nodes/second > Higher Is Better m6g.metal .... 104868482 |================================= m5.24xlarge .. 115160185 |==================================== m6i.24xlarge . 136656900 |=========================================== m6i.32xlarge . 169329043 |===================================================== High Performance Conjugate Gradient 3.1 GFLOP/s > Higher Is Better m6g.metal .... 21.46 |=============================== m5.24xlarge .. 26.89 |======================================= m6i.24xlarge . 37.22 |====================================================== m6i.32xlarge . 39.13 |========================================================= POV-Ray 3.7.0.7 Trace Time Seconds < Lower Is Better m6g.metal .... 57.439 |======================================================== m5.24xlarge .. 42.964 |========================================== m6i.24xlarge . 10.631 |========== m6i.32xlarge . 9.015 |========= miniFE 2.2 Problem Size: Small CG Mflops > Higher Is Better m6g.metal .... 23848.2 |======================================================= m5.24xlarge .. 14007.1 |================================ m6i.24xlarge . 19946.4 |============================================== m6i.32xlarge . 18797.6 |=========================================== NAS Parallel Benchmarks 3.4 Test / Class: EP.D Total Mop/s > Higher Is Better m6g.metal .... 2233.14 |============== m5.24xlarge .. 4875.47 |=============================== m6i.24xlarge . 6752.38 |========================================== m6i.32xlarge . 8765.66 |======================================================= Facebook RocksDB 6.22.1 Test: Random Read Op/s > Higher Is Better m6g.metal .... 270332614 |================================================ m5.24xlarge .. 194576074 |=================================== m6i.24xlarge . 231109408 |========================================= m6i.32xlarge . 298073130 |===================================================== Stockfish 13 Total Time Nodes Per Second > Higher Is Better m6g.metal .... 96657449 |============================== m5.24xlarge .. 105658561 |================================= m6i.24xlarge . 136790816 |=========================================== m6i.32xlarge . 169762583 |===================================================== NAS Parallel Benchmarks 3.4 Test / Class: BT.C Total Mop/s > Higher Is Better m6g.metal .... 24464.82 |====== m5.24xlarge .. 104533.15 |=========================== m6i.24xlarge . 136431.11 |==================================== m6i.32xlarge . 202455.31 |===================================================== Coremark 1.0 CoreMark Size 666 - Iterations Per Second Iterations/Sec > Higher Is Better m6g.metal .... 1236555.80 |============================== m5.24xlarge .. 1451630.52 |=================================== m6i.24xlarge . 1607068.54 |======================================= m6i.32xlarge . 2128843.09 |==================================================== TNN 0.3 Target: CPU - Model: MobileNet v2 ms < Lower Is Better m6g.metal .... 365.84 |================================================ m5.24xlarge .. 426.09 |======================================================== m6i.24xlarge . 350.38 |============================================== m6i.32xlarge . 349.17 |============================================== TNN 0.3 Target: CPU - Model: SqueezeNet v1.1 ms < Lower Is Better m6g.metal .... 341.32 |================================================ m5.24xlarge .. 394.51 |======================================================== m6i.24xlarge . 357.72 |=================================================== m6i.32xlarge . 357.69 |=================================================== NAS Parallel Benchmarks 3.4 Test / Class: FT.C Total Mop/s > Higher Is Better m6g.metal .... 21850.28 |=========== m5.24xlarge .. 50800.74 |========================== m6i.24xlarge . 70031.71 |==================================== m6i.32xlarge . 102661.18 |===================================================== Xcompact3d Incompact3d 2021-03-11 Input: input.i3d 193 Cells Per Direction Seconds < Lower Is Better m6g.metal .... 23.25 |========================================================= m5.24xlarge .. 21.47 |===================================================== m6i.24xlarge . 14.91 |===================================== m6i.32xlarge . 12.59 |=============================== Pennant 1.0.1 Test: sedovbig Hydro Cycle Time - Seconds < Lower Is Better m6g.metal .... 15.41 |=================================== m5.24xlarge .. 25.22 |========================================================= m6i.24xlarge . 17.25 |======================================= m6i.32xlarge . 15.14 |================================== m-queens 1.2 Time To Solve Seconds < Lower Is Better m6g.metal .... 19.43 |================================================== m5.24xlarge .. 22.34 |========================================================= m6i.24xlarge . 16.07 |========================================= m6i.32xlarge . 12.33 |=============================== LULESH 2.0.3 z/s > Higher Is Better m6g.metal .... 16867.37 |========================= m5.24xlarge .. 16272.59 |========================= m6i.24xlarge . 22519.12 |================================== m6i.32xlarge . 35739.82 |====================================================== Pennant 1.0.1 Test: leblancbig Hydro Cycle Time - Seconds < Lower Is Better m6g.metal .... 11.297260 |===================================================== m5.24xlarge .. 10.030260 |=============================================== m6i.24xlarge . 6.413928 |============================== m6i.32xlarge . 5.105541 |======================== NAS Parallel Benchmarks 3.4 Test / Class: CG.C Total Mop/s > Higher Is Better m6g.metal .... 13438.71 |=================== m5.24xlarge .. 30206.03 |========================================== m6i.24xlarge . 33146.76 |============================================== m6i.32xlarge . 38736.50 |====================================================== TNN 0.3 Target: CPU - Model: SqueezeNet v2 ms < Lower Is Better m6g.metal .... 105.07 |======================================================== m5.24xlarge .. 93.27 |================================================== m6i.24xlarge . 70.93 |====================================== m6i.32xlarge . 70.52 |====================================== NAS Parallel Benchmarks 3.4 Test / Class: EP.C Total Mop/s > Higher Is Better m6g.metal .... 2218.08 |=============== m5.24xlarge .. 4777.13 |================================ m6i.24xlarge . 6426.32 |============================================ m6i.32xlarge . 8107.02 |======================================================= N-Queens 1.0 Elapsed Time Seconds < Lower Is Better m6g.metal .... 3.761 |======================================================= m5.24xlarge .. 3.892 |========================================================= m6i.24xlarge . 3.144 |============================================== m6i.32xlarge . 2.312 |================================== Xcompact3d Incompact3d 2021-03-11 Input: input.i3d 129 Cells Per Direction Seconds < Lower Is Better m6g.metal .... 5.18380547 |==================================================== m5.24xlarge .. 4.76513163 |================================================ m6i.24xlarge . 3.49298970 |=================================== m6i.32xlarge . 2.95411030 |============================== NAS Parallel Benchmarks 3.4 Test / Class: MG.C Total Mop/s > Higher Is Better m6g.metal .... 25872.77 |============ m5.24xlarge .. 65732.22 |============================== m6i.24xlarge . 88248.73 |======================================== m6i.32xlarge . 117771.92 |===================================================== Geometric Mean Of All Test Results Result Composite - Intel M6i Ice Lake vs. Graviton2 Amazon EC2 Benchmarks Geometric Mean > Higher Is Better m6g.metal .... 671.58 |=========================== m5.24xlarge .. 821.74 |================================= m6i.24xlarge . 1117.98 |============================================= m6i.32xlarge . 1352.66 |======================================================= TNN 0.3 Performance / Cost - Target: CPU - Model: DenseNet ms x Dollar < Lower Is Better m6g.metal .... 516.35 |=================== m5.24xlarge .. 1298.78 |================================================ m6i.24xlarge . 1116.66 |========================================= m6i.32xlarge . 1487.42 |======================================================= asmFish 2018-07-23 Performance / Cost - 1024 Hash Memory, 26 Depth Nodes/second Per Dollar > Higher Is Better m6g.metal .... 1361928337.66 |================================================= m5.24xlarge .. 629290628.42 |======================= m6i.24xlarge . 962372535.21 |=================================== m6i.32xlarge . 834133216.75 |============================== High Performance Conjugate Gradient 3.1 Performance / Cost GFLOP/s Per Dollar > Higher Is Better m6g.metal .... 143.05 |========================================= m5.24xlarge .. 122.78 |=================================== m6i.24xlarge . 194.89 |======================================================== m6i.32xlarge . 147.67 |========================================== POV-Ray 3.7.0.7 Performance / Cost - Trace Time Seconds x Dollar < Lower Is Better m6g.metal .... 2.355 |====================================================== m5.24xlarge .. 2.492 |========================================================= m6i.24xlarge . 0.181 |==== m6i.32xlarge . 0.171 |==== miniFE 2.2 Performance / Cost - Problem Size: Small CG Mflops Per Dollar > Higher Is Better m6g.metal .... 1987350.00 |==================================================== m5.24xlarge .. 378570.27 |========== m6i.24xlarge . 738755.56 |=================== m6i.32xlarge . 522155.56 |============== NAS Parallel Benchmarks 3.4 Performance / Cost - Test / Class: EP.D Total Mop/s Per Dollar > Higher Is Better m6g.metal .... 53170.00 |========= m5.24xlarge .. 128301.84 |===================== m6i.24xlarge . 250088.15 |========================================= m6i.32xlarge . 324654.07 |===================================================== Facebook RocksDB 6.22.1 Performance / Cost - Test: Random Read Op/s Per Dollar > Higher Is Better m6g.metal .... 6593478390.24 |================================================= m5.24xlarge .. 2526962000.00 |=================== m6i.24xlarge . 3001420883.12 |====================== m6i.32xlarge . 2922285588.24 |====================== Stockfish 13 Performance / Cost - Total Time Nodes Per Second Per Dollar > Higher Is Better m6g.metal .... 6443829933.33 |================================================= m5.24xlarge .. 2855636783.78 |====================== m6i.24xlarge . 4716924689.66 |==================================== m6i.32xlarge . 4715627305.56 |==================================== NAS Parallel Benchmarks 3.4 Performance / Cost - Test / Class: BT.C Total Mop/s Per Dollar > Higher Is Better m6g.metal .... 302034.82 |== m5.24xlarge .. 2903698.61 |=================== m6i.24xlarge . 4872539.64 |================================= m6i.32xlarge . 7786742.69 |==================================================== Coremark 1.0 Performance / Cost - CoreMark Size 666 - Iterations Per Second Iterations/Sec Per Dollar > Higher Is Better m6g.metal .... 88325414.55 |=================================================== m5.24xlarge .. 36290762.98 |===================== m6i.24xlarge . 44640792.87 |========================== m6i.32xlarge . 44350897.80 |========================== TNN 0.3 Performance / Cost - Target: CPU - Model: MobileNet v2 ms x Dollar < Lower Is Better m6g.metal .... 6.585 |======================= m5.24xlarge .. 16.192 |======================================================== m6i.24xlarge . 11.212 |======================================= m6i.32xlarge . 15.014 |==================================================== TNN 0.3 Performance / Cost - Target: CPU - Model: SqueezeNet v1.1 ms x Dollar < Lower Is Better m6g.metal .... 5.461 |==================== m5.24xlarge .. 14.202 |==================================================== m6i.24xlarge . 11.447 |========================================== m6i.32xlarge . 15.381 |======================================================== NAS Parallel Benchmarks 3.4 Performance / Cost - Test / Class: FT.C Total Mop/s Per Dollar > Higher Is Better m6g.metal .... 1560734.29 |======= m5.24xlarge .. 4233395.00 |=================== m6i.24xlarge . 7781301.11 |=================================== m6i.32xlarge . 11406797.78 |=================================================== Xcompact3d Incompact3d 2021-03-11 Performance / Cost - Input: input.i3d 193 Cells Per Direction Seconds x Dollar < Lower Is Better m6g.metal .... 0.372 |=================================== m5.24xlarge .. 0.601 |========================================================= m6i.24xlarge . 0.283 |=========================== m6i.32xlarge . 0.277 |========================== Pennant 1.0.1 Performance / Cost - Test: sedovbig Hydro Cycle Time - Seconds x Dollar < Lower Is Better m6g.metal .... 0.170 |============ m5.24xlarge .. 0.832 |========================================================= m6i.24xlarge . 0.397 |=========================== m6i.32xlarge . 0.409 |============================ m-queens 1.2 Performance / Cost - Time To Solve Seconds x Dollar < Lower Is Better m6g.metal .... 0.253 |======================= m5.24xlarge .. 0.626 |========================================================= m6i.24xlarge . 0.321 |============================= m6i.32xlarge . 0.247 |====================== LULESH 2.0.3 Performance / Cost z/s Per Dollar > Higher Is Better m6g.metal .... 937076.11 |====================== m5.24xlarge .. 1084839.13 |========================= m6i.24xlarge . 2251911.50 |==================================================== m6i.32xlarge . 1786991.05 |========================================= Pennant 1.0.1 Performance / Cost - Test: leblancbig Hydro Cycle Time - Seconds x Dollar < Lower Is Better m6g.metal .... 0.090 |===================================== m5.24xlarge .. 0.140 |========================================================= m6i.24xlarge . 0.058 |======================== m6i.32xlarge . 0.051 |===================== NAS Parallel Benchmarks 3.4 Performance / Cost - Test / Class: CG.C Total Mop/s Per Dollar > Higher Is Better m6g.metal .... 1679838.75 |================ m5.24xlarge .. 5034338.33 |=============================================== m6i.24xlarge . 5524460.00 |==================================================== m6i.32xlarge . 5533785.71 |==================================================== TNN 0.3 Performance / Cost - Target: CPU - Model: SqueezeNet v2 ms x Dollar < Lower Is Better m6g.metal .... 0.525 |==================================== m5.24xlarge .. 0.839 |========================================================= m6i.24xlarge . 0.426 |============================= m6i.32xlarge . 0.635 |=========================================== NAS Parallel Benchmarks 3.4 Performance / Cost - Test / Class: EP.C Total Mop/s Per Dollar > Higher Is Better m6g.metal .... 739360.00 |============== m5.24xlarge .. 1592376.67 |=============================== m6i.24xlarge . 2142106.67 |========================================= m6i.32xlarge . 2702340.00 |==================================================== N-Queens 1.0 Performance / Cost - Elapsed Time Seconds x Dollar < Lower Is Better m6g.metal .... 0.011 |================================= m5.24xlarge .. 0.019 |========================================================= m6i.24xlarge . 0.013 |======================================= m6i.32xlarge . 0.007 |===================== Xcompact3d Incompact3d 2021-03-11 Performance / Cost - Input: input.i3d 129 Cells Per Direction Seconds x Dollar < Lower Is Better m6g.metal .... 0.021 |========================================= m5.24xlarge .. 0.029 |========================================================= m6i.24xlarge . 0.017 |================================= m6i.32xlarge . 0.015 |============================= NAS Parallel Benchmarks 3.4 Performance / Cost - Test / Class: MG.C Total Mop/s Per Dollar > Higher Is Better m6g.metal .... 5174554.00 |======= m5.24xlarge .. 16433055.00 |===================== m6i.24xlarge . 29416243.33 |====================================== m6i.32xlarge . 39257306.67 |===================================================