Find edge AI hardware
Task LLM generation · model-size class by model · batch per group · 3 platforms · edge hardware only · model_key "deepseekr17b"
Results are compared only within the same model. 1 model have results for this task.
How to read power: results are compared only within the same model. By efficiency ranks only results whose source publishes a power figure; By speed ranks every result. Measured = measured for this workload. Vendor-measured (typical/avg) = the vendor’s own measurement at typical or average load, not an independent test and not a maximum, so fps/W looks better than at max TDP. Chip, Module or Board TDP = the vendor’s published maximum, not measured. Typical power, not max TDP = only a typical figure is published, so fps/W looks better than at full load. Scope says what the watts cover: Accelerator only (host not included), Module, or Whole system. Max of range = top of a published range, so fps/W is a floor (≥). “— not published” = the source gives no power; we never estimate one.
Deepseek R1 7B 3 platforms · 0 with published power
| Rank | Platform | tok/s/W | tok/s | Power | Model · precision · batch | Grade · date | Source |
|---|---|---|---|---|---|---|---|
| – | NVIDIA Jetson AGX Thor | ≥0.318Module TDP · Module · floor | 41.32 tok/s | 130 WModuleModule TDP · max of range 40-130 W | Deepseek R1 7Bn/s on page · batch 1 | D on or before 2025-11-08 | NVIDIA Jetson BenchmarksJetPack 7.0, CUDA 13.0, TensorRT 10.13 (stated on page) |
| – | DC-ROMA Mainboard II (8-core RISC-V) (as tested by Jeff Geerling) | 0.126● Measured · Whole system | 4.90 tok/s | 38.90 WWhole system Measured | deepseek-r1:7bOllama default tag (4-bit unless tag says otherwise) · batch 1 | C 2025-10-10 (page last edited) | Jeff Geerling sbc-reviews (AI / LLM inference tables)ai-benchmarks README @ 2f56f19 |
| – | DC-ROMA RISC-V Mainboard II (as tested by Jeff Geerling) | 2.53 tok/s | — not published | deepseek-r1:7bOllama default tag (4-bit unless tag says otherwise) · batch 1 | C 2025-10-24 (page last edited) | Jeff Geerling ai-benchmarks issue results (comments)ai-benchmarks README @ 2f56f19 |
| Rank | Platform | tok/s/W | tok/s | Power | Model · precision · batch | Grade · date | Source |
|---|---|---|---|---|---|---|---|
| 1 | NVIDIA Jetson AGX Thor | ≥0.318Module TDP · Module · floor | 41.32 tok/s | 130 WModuleModule TDP · max of range 40-130 W | Deepseek R1 7Bn/s on page · batch 1 | D on or before 2025-11-08 | NVIDIA Jetson BenchmarksJetPack 7.0, CUDA 13.0, TensorRT 10.13 (stated on page) |
| 2 | DC-ROMA Mainboard II (8-core RISC-V) (as tested by Jeff Geerling) | 0.126● Measured · Whole system | 4.90 tok/s | 38.90 WWhole system Measured | deepseek-r1:7bOllama default tag (4-bit unless tag says otherwise) · batch 1 | C 2025-10-10 (page last edited) | Jeff Geerling sbc-reviews (AI / LLM inference tables)ai-benchmarks README @ 2f56f19 |
| 3 | DC-ROMA RISC-V Mainboard II (as tested by Jeff Geerling) | 2.53 tok/s | — not published | deepseek-r1:7bOllama default tag (4-bit unless tag says otherwise) · batch 1 | C 2025-10-24 (page last edited) | Jeff Geerling ai-benchmarks issue results (comments)ai-benchmarks README @ 2f56f19 |
Excluded from ranking (2)
Partial or suspected partial models, fps/W outliers and rows HWBench Data flags (rank_note) are kept out of the rankings. Rows with a published power figure that Data does not mark for the default fps/W ranking still appear under By speed. Every row stays in /find.json.
| Platform | Model | Result | Power | Left out of | Why | Source |
|---|---|---|---|---|---|---|
| NVIDIA Jetson AGX Thor | Deepseek R1 7B | 41.32 tok/s | 130 WModuleModule TDP · max of range 40-130 W | Efficiency ranking | HWBench Data has not verified tok/s/W rows for ranking yet | NVIDIA Jetson BenchmarksJetPack 7.0, CUDA 13.0, TensorRT 10.13 (stated on page) |
| DC-ROMA Mainboard II (8-core RISC-V) (as tested by Jeff Geerling) | deepseek-r1:7b | 4.90 tok/s | 38.90 WWhole system Measured | Efficiency ranking | HWBench Data has not verified tok/s/W rows for ranking yet | Jeff Geerling sbc-reviews (AI / LLM inference tables)ai-benchmarks README @ 2f56f19 |