name stringclasses 9
values | timestamp timestamp[s]date 2026-06-20 00:00:00 2026-06-20 00:00:00 | task stringclasses 3
values | model stringclasses 3
values | precision stringclasses 3
values | target_device stringclasses 1
value | draft_device stringclasses 3
values | num_assistant_tokens_seed int64 1 5 ⌀ | methodology stringclasses 1
value | environment dict | measurements unknown |
|---|---|---|---|---|---|---|---|---|---|---|
standalone_draft_cpu | 2026-06-20T00:00:00 | standalone_draft_throughput | Qwen3-0.6B | INT4 symmetric (--sym, group-size 128) | null | CPU | null | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 106.8,
"throughput_tok_s_std": 0.7000000000000001,
"tpot_ms": 9.4,
"construct_s": 1.3,
"num_generated_tokens": 128
} |
standalone_draft_gpu | 2026-06-20T00:00:00 | standalone_draft_throughput | Qwen3-0.6B | INT4 symmetric (--sym, group-size 128) | null | GPU | null | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 150,
"throughput_tok_s_std": 0.9,
"tpot_ms": 6.7,
"construct_s": 3.5,
"num_generated_tokens": 128
} |
standalone_draft_npu | 2026-06-20T00:00:00 | standalone_draft_throughput | Qwen3-0.6B | INT4 symmetric (--sym, group-size 128) | null | NPU | null | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 29.2,
"throughput_tok_s_std": 0.1,
"tpot_ms": 34.3,
"construct_s": 13.7,
"num_generated_tokens": 128
} |
gpu_only | 2026-06-20T00:00:00 | baseline_no_draft | Qwen3-14B | INT4 asymmetric | GPU | null | null | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 12.198,
"throughput_tok_s_std": 0.14400000000000002,
"speedup_vs_gpu_only": 1,
"ttft_ms": 250.7,
"tpot_ms": 82,
"main_model_tpot_ms": null,
"draft_model_tpot_ms": null,
"acceptance_rate": null,
"construct_compile_s": 7.8,
"num_generated_tokens": 128
} |
gpu_cpu_static3 | 2026-06-20T00:00:00 | speculative_decoding | Qwen3-14B (target) + Qwen3-0.6B (draft) | target INT4 asymmetric / draft INT4 symmetric | GPU | CPU | 3 | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 16.46,
"throughput_tok_s_std": 0.487,
"speedup_vs_gpu_only": 1.349,
"ttft_ms": 255.2,
"tpot_ms": 60.8,
"main_model_tpot_ms": 53.3,
"draft_model_tpot_ms": 10.6,
"acceptance_rate": 0.4,
"construct_compile_s": 9.3,
"num_generated_tokens": 128
} |
gpu_cpu_static5 | 2026-06-20T00:00:00 | speculative_decoding | Qwen3-14B (target) + Qwen3-0.6B (draft) | target INT4 asymmetric / draft INT4 symmetric | GPU | CPU | 5 | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 14.697,
"throughput_tok_s_std": 0.874,
"speedup_vs_gpu_only": 1.205,
"ttft_ms": 259.5,
"tpot_ms": 68.2,
"main_model_tpot_ms": 57.3,
"draft_model_tpot_ms": 11.6,
"acceptance_rate": 0.316,
"construct_compile_s": 9.4,
"num_generated_tokens": 128
} |
gpu_npu_static1 | 2026-06-20T00:00:00 | speculative_decoding | Qwen3-14B (target) + Qwen3-0.6B (draft) | target INT4 asymmetric / draft INT4 symmetric | GPU | NPU | 1 | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 9.013,
"throughput_tok_s_std": 0.047,
"speedup_vs_gpu_only": 0.739,
"ttft_ms": 252.5,
"tpot_ms": 110.9,
"main_model_tpot_ms": 54.6,
"draft_model_tpot_ms": 60.9,
"acceptance_rate": 0.598,
"construct_compile_s": 22.2,
"num_generated_tokens": 128
} |
gpu_npu_static3 | 2026-06-20T00:00:00 | speculative_decoding | Qwen3-14B (target) + Qwen3-0.6B (draft) | target INT4 asymmetric / draft INT4 symmetric | GPU | NPU | 3 | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 7.054,
"throughput_tok_s_std": 0.031,
"speedup_vs_gpu_only": 0.578,
"ttft_ms": 250.4,
"tpot_ms": 141.8,
"main_model_tpot_ms": 62.9,
"draft_model_tpot_ms": 49.4,
"acceptance_rate": 0.354,
"construct_compile_s": 21.8,
"num_generated_tokens": 128
} |
gpu_npu_static5 | 2026-06-20T00:00:00 | speculative_decoding | Qwen3-14B (target) + Qwen3-0.6B (draft) | target INT4 asymmetric / draft INT4 symmetric | GPU | NPU | 5 | Greedy, ignore_eos=True, fixed 128-token budget, 1 warmup + 3 measured reps, each config in a fresh process. num_assistant_tokens is the stateful pipeline's adaptive seed (not a fixed per-round count). acceptance_rate = accepted / draft-proposed (GenAI SDMetrics definition). Throughput/TTFT/TPOT from PerfMetrics; per-m... | {
"cpu": "Intel Core Ultra 7 258V (Lunar Lake)",
"gpu": "Intel Arc 140V (Xe2, 16 GB)",
"npu": "Intel AI Boost (NPU)",
"os": "Windows 11 Pro build 26200 (25H2)",
"npu_driver": "32.0.100.4778",
"gpu_driver": "32.0.101.8826",
"openvino_version": "2026.1.0",
"openvino_genai_version": "2026.1.0.0",
"not_me... | {
"throughput_tok_s": 6.709,
"throughput_tok_s_std": 0.053,
"speedup_vs_gpu_only": 0.55,
"ttft_ms": 252.5,
"tpot_ms": 149.1,
"main_model_tpot_ms": 64.7,
"draft_model_tpot_ms": 47.6,
"acceptance_rate": 0.318,
"construct_compile_s": 22,
"num_generated_tokens": 128
} |
Speculative decoding draft-device characterization on Intel Lunar Lake (OpenVINO GenAI)
Reference performance data for heterogeneous speculative decoding with OpenVINO GenAI: a GPU target model (Qwen3-14B INT4) paired with a small draft model (Qwen3-0.6B INT4) run on the CPU vs the NPU vs the GPU, plus standalone draft-only throughput per device. Measured on a single Intel Core Ultra 7 258V (Lunar Lake) laptop.
Headline: putting the draft on the NPU is a net slowdown (0.55–0.74× of GPU-only throughput); the same draft on the CPU is a 1.35× speedup. Root cause: the NPU runs the small draft model ~3.6× slower per token than the CPU, and speculative decoding only pays off when the draft is cheap. This corroborates OpenVINO's documented guidance (CPU draft for a GPU/CPU target, or both models on the NPU) — it is reference characterization, not a bug report.
- Narrative write-up: https://github.com/openvinotoolkit/openvino/discussions/36484
- Related (closed) issue: https://github.com/openvinotoolkit/openvino/issues/34450
Hardware
| Component | Detail |
|---|---|
| CPU | Intel Core Ultra 7 258V (Lunar Lake) |
| GPU | Intel Arc 140V (Xe2, 16 GB) |
| NPU | Intel AI Boost |
| OS | Windows 11 Pro build 26200 (25H2) |
| NPU driver | 32.0.100.4778 |
| GPU driver | 32.0.101.8826 |
| OpenVINO | 2026.1.0 |
| OpenVINO GenAI | 2026.1.0.0 |
Models
- Target: Qwen3-14B, INT4 (asymmetric per-group — fine on GPU).
- Draft: Qwen3-0.6B, INT4 symmetric (
--sym --group-size 128 --ratio 1.0). Symmetric is required for the NPU (per-group asymmetric INT4 is unsupported there). Same draft used on every device. - Target and draft share the Qwen3 tokenizer (required for speculative decoding).
Methodology
Greedy, ignore_eos=True, fixed 128-token budget, 1 warm-up + 3 measured reps, each config in a
fresh process. num_assistant_tokens (NAT) is the stateful pipeline's adaptive seed, not a fixed
per-round proposal count. Metrics from PerfMetrics / SDPerModelsPerfMetrics;
acceptance_rate = accepted / draft-proposed. Construct/compile time is excluded from throughput.
Single prompt — acceptance numbers are illustrative; the throughput/per-token-latency result is the
robust, device-physics finding.
Results
Standalone draft-only (Qwen3-0.6B-sym, no speculative decoding)
| device | throughput | ms/token |
|---|---|---|
| CPU | 106.8 tok/s | 9.4 |
| GPU | 150.0 tok/s | 6.7 |
| NPU | 29.2 tok/s | 34.3 |
Speculative decoding (GPU target + draft)
| draft device | NAT seed | throughput | speedup | acceptance |
|---|---|---|---|---|
| none (GPU only) | — | 12.2 tok/s | 1.00× | — |
| CPU | 3 | 16.5 tok/s | 1.35× | 0.40 |
| CPU | 5 | 14.7 tok/s | 1.20× | 0.32 |
| NPU | 1 | 9.0 tok/s | 0.74× | 0.60 |
| NPU | 3 | 7.1 tok/s | 0.58× | 0.35 |
| NPU | 5 | 6.7 tok/s | 0.55× | 0.32 |
Files
npu_specdecode_perf.jsonl— one record per config (standalone + spec-decode); programmatic import.npu_specdecode_perf.csv— flattened (env_*/meas_*prefixes).
Not measured (explicit)
Single machine, single prompt, n=3; greedy only; default NPU prompt/response budgets; no co-resident memory cost; no subjective quality beyond a greedy token-identity check. Behavior may differ on OpenVINO 2026.2/master.
License
CC-BY-4.0 (adjust to your preference before publishing). Author: Blair DuCray-Oppat (GitHub @blairducrayoppat).
Citation
NPU vs CPU draft speculative-decoding characterization on Intel Lunar Lake (OpenVINO GenAI 2026.1), 2026. https://github.com/openvinotoolkit/openvino/discussions/36484
- Downloads last month
- 39