Configs › DeepSeek-R1-Distill-Qwen-32B on RTX 5090: 66.5 tok/s measured
DeepSeek-R1-Distill-Qwen-32B on RTX 5090: 66.5 tok/s measured
Measured DeepSeek-R1-Distill-Qwen-32B speeds: 66.5 tok/s on RTX 5090 (Q4_K_M, llama.cpp/cuda). 4 configs, each linked to its source, snapshot 2026-10-15.
Snapshot 2026-10-15 · 394 configs on AI boxes and 423 on GPU cards · 123 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.
Quick answers
- How fast is DeepSeek-R1-Distill-Qwen-32B on RTX 5090?
- 66.5 tok/s single-stream decode (Q4_K_M, llama.cpp/cuda), the fastest of 4 measured configs on NVIDIA RTX 5090 (32GB), measured by paulplee, source last checked 2026-10-15. source
DeepSeek-R1-Distill-Qwen-32B on GPU cards, ranked on their own
GPU cards run the model from their own VRAM: very fast on what fits, but most cannot load what a 128 GB AI box can. Ranked on their own, never against the boxes.
| Model | Hardware | Quant | Backend | Decode | Context | Trust | Source |
|---|---|---|---|---|---|---|---|
| DeepSeek-R1-Distill-Qwen-32B | RTX 5090 | Q4_K_M | llama.cpp/cuda | 66.5 tok/s | 8192 | Structured table | source |
| DeepSeek-R1-Distill-Qwen-32B | RTX 5090 | Q3_K_M | llama.cpp/cuda | 66.4 tok/s | 8192 | Structured table | source |
| DeepSeek-R1-Distill-Qwen-32B | RTX 5090 | Q5_K_M | llama.cpp/cuda | 58.8 tok/s | 8192 | Structured table | source |
| DeepSeek-R1-Distill-Qwen-32B | RTX 5090 | Q6_K | llama.cpp/cuda | 52.1 tok/s | 8192 | Structured table | source |
More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.