RTX 5090 vs Radeon AI PRO R9700 for local AI
Today the Radeon AI PRO R9700 is cheaper: $1,820 against $4,920.99 for the RTX 5090. Per gigabyte of memory the Radeon AI PRO R9700 wins, $56 against $153. For Qwen3.8 27B · Q4_K_M the RTX 5090 is faster, about 71 against 19 tokens per second.
| NVIDIA GeForce RTX 5090 | AMD Radeon AI PRO R9700 | |
|---|---|---|
| Lowest price today, delivered | $4,920.99 on eBay | $1,820 on eBay |
| Price per GB | $153/GB | $56/GB |
| Memory | 32 GB GDDR7 | 32 GB GDDR6 ECC |
| Memory bandwidth | 1,792 GB/s | 640 GB/s |
| FP16 compute, dense | 210 TFLOPS | 96 TFLOPS |
| Reference models that fit at Q4_K_M | 11 of 28 | 11 of 28 |
| Power | 575 W | 300 W |
| Form factor | Dual slot (Founders Edition), partner cards up to 4 slots | Dual slot, full height |
| Interface | PCIe 5 | PCIe 5 |
| Software platform | cuda | rocm |
| Launched | 2025 | 2025 |
| Launch price | $1,999 | $1,299 |
Qwen3.8 27B · Q4_K_M by quantization
| Quant | RTX 5090 | Radeon AI PRO R9700 |
|---|---|---|
| FP16 | does not fit | does not fit |
| Q8_0 | ~41 tok/s | ~11 tok/s |
| Q6_K | ~53 tok/s | ~14 tok/s |
| Q5_K_M | ~61 tok/s | ~17 tok/s |
| Q4_K_M | ~71 tok/s | ~19 tok/s |
| Q3_K_M | ~88 tok/s | ~24 tok/s |
| IQ2_M | ~130 tok/s | ~35 tok/s |
Fit means the weights plus an 8K context cache fit in memory; speed is the estimated decode rate. Prices are the cheapest live eBay listing that passed our filters, item plus shipping. How we estimate.