keyword spotting benchmarks on edge AI hardware
MLPerf Tiny, microcontroller class. Throughput in inferences per second; measured energy per inference is shown on each platform page. Not ranked against edge GPUs or accelerators.
33 published rows. Each table compares only results with the same model-size class and batch size. Search this workload · JSON.
Fastest in the best-covered like-for-like group (unknown, batch 1): Next Generation Snapdragon Mobile Platform MTP at 15396 inferences/s A 2025-09-17 (published). 1 groups, 17 platforms.
Show data — 1 like-for-like tables
unknown · batch 1 (17 platforms)
✓ Like-for-like: model-size class unknown, batch 1 (size or batch not stated by source — unverified). Model, precision and stack may still differ.
| Platform | Result | Model · precision · stack | Grade | Power | Source |
|---|---|---|---|---|---|
| Next Generation Snapdragon Mobile Platform MTP | 15396 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) PTQ · Qualcomm AI Stack · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| Snapdragon 8 Gen 3 Mobile HDK (Snapdragon 8 Gen 3 Mobile Platform) | 11705 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) PTQ · Qualcomm AI Stack · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| K1_BOARD_V1.0_HIGH_FREQ (K1 Octa-Core RISC-V) | 1735 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) Uniform symmetric quantization · · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| K1_BOARD_V1.0 (K1 Octa-Core RISC-V) | 717 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) Uniform symmetric quantization · · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| NDP9120 (N/A) | 675 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) Uniform scaling from pre-trained weights · N/A · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| EK-RA8D1 (R7FA8D1AHECBD) | 289 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-H7A3ZI-Q (STM32H7A3ZIT6Q) | 88.36 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v9.0.0 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-H7A3ZI (STM32H7A3ZIT6Q) | 68.08 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-U3C5ZIT6-Q | 60.49 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · STEDGE-AI Internal version · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| DISCO-F746NG (STM32F746NG) | 52.75 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-U575ZI-Q (STM32U575ZIT6Q) | 34.49 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v10.2 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| NUCLEO-L4R5ZI (STM32L4R5ZIT6U) | 23.16 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v9.0 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-L4R5ZI-P (STM32L4R5ZIT6P) | 23.03 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v10.2 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| NUCLEO-U385RG-Q (STM32U385RGT6Q) | 20.70 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v10.2 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| CY8CPROTO-062-4343W (PSoC 62 MCU) | 18.47 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| Renesas RH850/F1KM-S4 R7F701649 | 15.71 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-G0B1RE (STM32G0B1RE) | 2.11 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
Fastest in the best-covered like-for-like group (unknown, batch 1): Next Generation Snapdragon Mobile Platform MTP at 15396 inferences/s A 2025-09-17 (published). 1 groups, 17 platforms.
Show data — 1 like-for-like tables
unknown · batch 1 (17 platforms)
✓ Like-for-like: model-size class unknown, batch 1 (size or batch not stated by source — unverified). Model, precision and stack may still differ.
| Platform | Result | Model · precision · stack | Grade | Power | Source |
|---|---|---|---|---|---|
| Next Generation Snapdragon Mobile Platform MTP | 15396 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) PTQ · Qualcomm AI Stack · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| Snapdragon 8 Gen 3 Mobile HDK (Snapdragon 8 Gen 3 Mobile Platform) | 11705 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) PTQ · Qualcomm AI Stack · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| K1_BOARD_V1.0_HIGH_FREQ (K1 Octa-Core RISC-V) | 1735 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) Uniform symmetric quantization · · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| K1_BOARD_V1.0 (K1 Octa-Core RISC-V) | 717 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) Uniform symmetric quantization · · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| NDP9120 (N/A) | 675 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) Uniform scaling from pre-trained weights · N/A · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| EK-RA8D1 (R7FA8D1AHECBD) | 289 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-H7A3ZI-Q (STM32H7A3ZIT6Q) | 88.36 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v9.0.0 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-H7A3ZI (STM32H7A3ZIT6Q) | 68.08 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-U3C5ZIT6-Q | 60.49 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · STEDGE-AI Internal version · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| DISCO-F746NG (STM32F746NG) | 52.75 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-U575ZI-Q (STM32U575ZIT6Q) | 34.49 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v10.2 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| NUCLEO-L4R5ZI (STM32L4R5ZIT6U) | 23.16 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v9.0 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-L4R5ZI-P (STM32L4R5ZIT6P) | 23.03 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v10.2 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| NUCLEO-U385RG-Q (STM32U385RGT6Q) | 20.70 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) INT8 · X-CUBE-AI v10.2 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.3 (closed) 2025-09-17 (published) |
| CY8CPROTO-062-4343W (PSoC 62 MCU) | 18.47 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| Renesas RH850/F1KM-S4 R7F701649 | 15.71 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
| NUCLEO-G0B1RE (STM32G0B1RE) | 2.11 inferences/s throughput · b=1 | DS-CNN (MLPerf Tiny KWS) not stated · Bosch Hardware-Aware Lowering Engine (HALE) version 1.0.175 · unknown | A | unknown power n/a no power published | MLPerf Tiny v1.2 (closed) Older data (2024) |
How to read these numbers
Every number is labelled. Confidence: A MLPerf closed, audited · B documented third-party method · C community / review · D vendor-published, not independently verified · E estimate. Power basis: measured with the result · spec vendor spec/envelope · power n/a not published.
Machine-readable: /find.json (see llms.txt); full tables are in “Show data”. Confidence grades: A audited MLPerf closed · B documented third-party · C community · D vendor-published · E estimate. Unknown values are shown as unknown, never estimated. Generated 2026-10-16 from public sources.