# Inference Throughput on Qwen3 Query Projection Module NVIDIA A40 Source: [Inference Throughput on Qwen3 Query Projection Module NVIDIA A40]() Collection index: [llms.txt]() Data updated: 2026-03-23T11:04:17.575Z Evaluation of the inference throughput performance for the Qwen3 Query Projection Module running on an NVIDIA A40 GPU, measured in thousands of tokens per second. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Inference Throughput | | Dataset (including any stated split) | Qwen3 Query Projection Module NVIDIA A40 | | Metric | Throughput (k tokens/sec) | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 30 of 30 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Throughput (k tokens/sec)](): Higher is better; unit: SCALAR. ## Recorded results | Method | Throughput (k tokens/sec) | Result date | Report paper | | --- | --- | --- | --- | | TTQ (r = 0) | 80.63 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 0) | 78.68 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (awq\_gemm) | 75.76 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (marlin\_gemm) | 73.98 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (marlin\_gemm) | 73.7 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (marlin\_gemm) | 73.32 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (awq\_gemm) | 71.6 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 0) | 69.11 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 16) | 66.67 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 16) | 59.08 | Not recorded | 69c0c9020636e5bf4afc3b59 | | FP16 | 57.58 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (awq\_gemm) | 54.61 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 16) | 49.67 | Not recorded | 69c0c9020636e5bf4afc3b59 | | FP16 | 41.29 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (marlin\_gemm) | 38.91 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (marlin\_gemm) | 37.52 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 0) | 36.52 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (awq\_gemm) | 32.68 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (marlin\_gemm) | 30.32 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 16) | 30.32 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (awq\_gemm) | 25.28 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 0) | 24.94 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 16) | 23.78 | Not recorded | 69c0c9020636e5bf4afc3b59 | | FP16 | 21.38 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 0) | 19.34 | Not recorded | 69c0c9020636e5bf4afc3b59 | | AWQ (awq\_gemm) | 18.28 | Not recorded | 69c0c9020636e5bf4afc3b59 | | TTQ (r = 16) | 17.43 | Not recorded | 69c0c9020636e5bf4afc3b59 | | FP16 | 14.1 | Not recorded | 69c0c9020636e5bf4afc3b59 | | FP16 | 10.43 | Not recorded | 69c0c9020636e5bf4afc3b59 | | FP16 | 6.86 | Not recorded | 69c0c9020636e5bf4afc3b59 | ### TTQ (r = 0) Throughput (k tokens/sec): 80.63. Result ID: 69c11e0747f34ce924a96c10. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"0.6B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 0) Throughput (k tokens/sec): 78.68. Result ID: 69c11e0747f34ce924a96c11. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1.7B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (awq\_gemm) Throughput (k tokens/sec): 75.76. Result ID: 69c11e0747f34ce924a96c04. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"0.6B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (marlin\_gemm) Throughput (k tokens/sec): 73.98. Result ID: 69c11e0747f34ce924a96c0c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"4B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (marlin\_gemm) Throughput (k tokens/sec): 73.7. Result ID: 69c11e0747f34ce924a96c0b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1.7B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (marlin\_gemm) Throughput (k tokens/sec): 73.32. Result ID: 69c11e0747f34ce924a96c0a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"0.6B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (awq\_gemm) Throughput (k tokens/sec): 71.6. Result ID: 69c11e0747f34ce924a96c05. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1.7B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 0) Throughput (k tokens/sec): 69.11. Result ID: 69c11e0747f34ce924a96c12. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"4B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 16) Throughput (k tokens/sec): 66.67. Result ID: 69c11e0747f34ce924a96c16. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"0.6B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 16) Throughput (k tokens/sec): 59.08. Result ID: 69c11e0747f34ce924a96c17. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1.7B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### FP16 Throughput (k tokens/sec): 57.58. Result ID: 69c11e0747f34ce924a96bfe. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"0.6B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (awq\_gemm) Throughput (k tokens/sec): 54.61. Result ID: 69c11e0747f34ce924a96c06. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"4B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 16) Throughput (k tokens/sec): 49.67. Result ID: 69c11e0747f34ce924a96c18. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"4B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### FP16 Throughput (k tokens/sec): 41.29. Result ID: 69c11e0747f34ce924a96bff. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"1.7B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (marlin\_gemm) Throughput (k tokens/sec): 38.91. Result ID: 69c11e0747f34ce924a96c0d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"8B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (marlin\_gemm) Throughput (k tokens/sec): 37.52. Result ID: 69c11e0747f34ce924a96c0e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"14B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 0) Throughput (k tokens/sec): 36.52. Result ID: 69c11e0747f34ce924a96c13. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"8B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (awq\_gemm) Throughput (k tokens/sec): 32.68. Result ID: 69c11e0747f34ce924a96c07. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"8B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (marlin\_gemm) Throughput (k tokens/sec): 30.32. Result ID: 69c11e0747f34ce924a96c0f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"32B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 16) Throughput (k tokens/sec): 30.32. Result ID: 69c11e0747f34ce924a96c19. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"8B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (awq\_gemm) Throughput (k tokens/sec): 25.28. Result ID: 69c11e0747f34ce924a96c08. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"14B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 0) Throughput (k tokens/sec): 24.94. Result ID: 69c11e0747f34ce924a96c14. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"14B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 16) Throughput (k tokens/sec): 23.78. Result ID: 69c11e0747f34ce924a96c1a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"14B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### FP16 Throughput (k tokens/sec): 21.38. Result ID: 69c11e0747f34ce924a96c00. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"4B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 0) Throughput (k tokens/sec): 19.34. Result ID: 69c11e0747f34ce924a96c15. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"32B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### AWQ (awq\_gemm) Throughput (k tokens/sec): 18.28. Result ID: 69c11e0747f34ce924a96c09. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"32B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### TTQ (r = 16) Throughput (k tokens/sec): 17.43. Result ID: 69c11e0747f34ce924a96c1b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"32B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### FP16 Throughput (k tokens/sec): 14.1. Result ID: 69c11e0747f34ce924a96c01. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"8B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### FP16 Throughput (k tokens/sec): 10.43. Result ID: 69c11e0747f34ce924a96c02. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"14B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. ### FP16 Throughput (k tokens/sec): 6.86. Result ID: 69c11e0747f34ce924a96c03. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"32B","Hardware":"NVIDIA A40 GPU"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.