Builds / AMD Radeon AI PRO R9700 32 GB
Radeon AI PRO R9700 32 GB, vLLM FP16 (Puget Systems): Llama 3.1 8B at 29.1 tok/s, DeepSeek R1 Distill 8B 34.5, Qwen3 8B 27.7
Last updated:
Source: pugetsystems.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
Legacy models only. Every model measured on this rig now has a newer successor: Llama 3.1 8B → Muse Glimmer 30B, DeepSeek R1 Distill 8B → DeepSeek V4.1 Flash. The hardware numbers are still a useful reference.
- GPU(s)
- AMD Radeon AI PRO R9700 32 GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 32
- CPU
- AMD Ryzen Threadripper PRO 5995WX (64 cores)
Amazon · eBay (used) · Newegg - Motherboard
- ASUS Pro WS WRX80E-SAGE SE
Amazon · eBay (used) · Newegg - RAM
- 128 GB
Amazon · eBay (used) · Newegg - PSU
- not stated
- Other
- Puget Systems test workstation, bare-metal Rocky Linux 10.1, PCIe 5.0 x16 per card, containerized runtimes (Podman/Docker) with PCIe peer-to-peer; one card used
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Llama 3.1 8B legacy | FP16 (unquantized) | vLLM v0.20.2 ROCm, HIP graphs | 29.1 | not stated | context: max-model-len 16,384📏 NVIDIA GenAI-Perf streaming, 500 input / 500 output tokens, concurrency 1; TTFT 113 ms, ITL 34 ms; 102 tok/s total at 4 users, 153 at 8 |
| DeepSeek R1 Distill 8B legacy | FP16 (unquantized) | vLLM v0.20.2 ROCm, HIP graphs | 34.5 | not stated | context: max-model-len 16,384📏 NVIDIA GenAI-Perf streaming, 500 input / 500 output tokens, concurrency 1; TTFT 111 ms (DeepSeek-R1-Distill-Llama-8B) |
| Qwen3 8B legacy | FP16 (unquantized) | vLLM v0.20.2 ROCm, HIP graphs | 27.7 | not stated | context: max-model-len 16,384📏 NVIDIA GenAI-Perf streaming, 500 input / 500 output tokens, concurrency 1, 120-second window for thinking tokens; TTFT 92 ms, ITL 36 ms |
Vendor-lab test by Puget Systems (posted 18 Aug 2026, updated 10 Sep 2026). Puget corrected an earlier, inflated Llama 8B figure (61.7 tok/s, above what 640 GB/s allows); the numbers here are the corrected ones. Same article: Qwen2.5 3B Instruct FP16 53.3 tok/s for one user (183 at 4, 251 at 8). The dual-card Qwen3.6 27B results are a separate RigBase build.
Current GPU street price (estimate, not the builder's cost)
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.
- 1× Radeon AI Pro R9700: ≈€1,848 each (≈$2,071), new, approx. price as of 2026-10-24, lowest listing on billiger.de (PowerColor Radeon AI PRO R9700 32 GB GDDR6)
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.
Source: https://www.pugetsystems.com/labs/articles/amd-radeon-ai-pro-r9700-dual-gpu-ai-inference-performance/
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)