RTX 5090 vs RTX PRO 6000 for local AI
Today the RTX 5090 is cheaper: $4,920.99 against $16,499 for the RTX PRO 6000. Per gigabyte of memory the RTX 5090 wins, $153 against $171. For Qwen3.8 27B · Q4_K_M the RTX 5090 is faster, about 71 against 71 tokens per second. The RTX PRO 6000 holds more, 96 GB against 32 GB, and fits 18 of our 28 reference models at Q4_K_M where the RTX 5090 fits 11.
| NVIDIA GeForce RTX 5090 | NVIDIA RTX PRO 6000 Blackwell | |
|---|---|---|
| Lowest price today, delivered | $4,920.99 on eBay | $16,499 on eBay |
| Price per GB | $153/GB | $171/GB |
| Memory | 32 GB GDDR7 | 96 GB GDDR7 ECC |
| Memory bandwidth | 1,792 GB/s | 1,792 GB/s |
| FP16 compute, dense | 210 TFLOPS | 504 TFLOPS |
| Reference models that fit at Q4_K_M | 11 of 28 | 18 of 28 |
| Power | 575 W | 600 W |
| Form factor | Dual slot (Founders Edition), partner cards up to 4 slots | Dual slot, full height (Workstation Edition) |
| Interface | PCIe 5 | PCIe 5 |
| Software platform | cuda | cuda |
| Launched | 2025 | 2025 |
| Launch price | $1,999 | $8,565 |
Qwen3.8 27B · Q4_K_M by quantization
| Quant | RTX 5090 | RTX PRO 6000 |
|---|---|---|
| FP16 | does not fit | ~22 tok/s |
| Q8_0 | ~41 tok/s | ~41 tok/s |
| Q6_K | ~53 tok/s | ~53 tok/s |
| Q5_K_M | ~61 tok/s | ~61 tok/s |
| Q4_K_M | ~71 tok/s | ~71 tok/s |
| Q3_K_M | ~88 tok/s | ~88 tok/s |
| IQ2_M | ~130 tok/s | ~130 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.