RTX 3090 vs RX 7900 XTX for local AI
Today the RX 7900 XTX is cheaper: $900 against $1,497 for the RTX 3090. Per gigabyte of memory the RX 7900 XTX wins, $37 against $62. For Qwen3.8 27B · Q4_K_M the RTX 3090 is faster, about 37 against 29 tokens per second.
| NVIDIA GeForce RTX 3090 | AMD Radeon RX 7900 XTX | |
|---|---|---|
| Lowest price today, delivered | $1,497 on eBay | $900 on eBay |
| Price per GB | $62/GB | $37/GB |
| Memory | 24 GB GDDR6X | 24 GB GDDR6 |
| Memory bandwidth | 936 GB/s | 960 GB/s |
| FP16 compute, dense | 71 TFLOPS | 123 TFLOPS |
| Reference models that fit at Q4_K_M | 11 of 28 | 11 of 28 |
| Power | 350 W | 355 W |
| Form factor | Triple slot, full height | Triple slot, full height |
| Interface | PCIe 4 | PCIe 4 |
| Software platform | cuda | rocm |
| Launched | 2020 | 2022 |
| Launch price | $1,499 | $999 |
Qwen3.8 27B · Q4_K_M by quantization
| Quant | RTX 3090 | RX 7900 XTX |
|---|---|---|
| FP16 | does not fit | does not fit |
| Q8_0 | does not fit | does not fit |
| Q6_K | ~27 tok/s, 5K ctx | ~22 tok/s, 5K ctx |
| Q5_K_M | ~32 tok/s | ~25 tok/s |
| Q4_K_M | ~37 tok/s | ~29 tok/s |
| Q3_K_M | ~46 tok/s | ~36 tok/s |
| IQ2_M | ~66 tok/s | ~52 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.