RTX A6000 vs RTX 6000 Ada for local AI
Today the RTX A6000 is cheaper: $4,500 against $7,698.43 for the RTX 6000 Ada. Per gigabyte of memory the RTX A6000 wins, $93 against $160. For Qwen3.8 27B · Q4_K_M the RTX 6000 Ada is faster, about 38 against 30 tokens per second.
| NVIDIA RTX A6000 | NVIDIA RTX 6000 Ada | |
|---|---|---|
| Lowest price today, delivered | $4,500 on eBay | $7,698.43 on eBay |
| Price per GB | $93/GB | $160/GB |
| Memory | 48 GB GDDR6 ECC | 48 GB GDDR6 ECC |
| Memory bandwidth | 768 GB/s | 960 GB/s |
| FP16 compute, dense | 155 TFLOPS | 182 TFLOPS |
| Reference models that fit at Q4_K_M | 12 of 28 | 12 of 28 |
| Power | 300 W | 300 W |
| Form factor | Dual slot, full height | Dual slot, full height |
| Interface | PCIe 4 | PCIe 4 |
| Software platform | cuda | cuda |
| Launched | 2020 | 2022 |
| Launch price | $4,650 | $6,800 |
Qwen3.8 27B · Q4_K_M by quantization
| Quant | RTX A6000 | RTX 6000 Ada |
|---|---|---|
| FP16 | does not fit | does not fit |
| Q8_0 | ~17 tok/s | ~22 tok/s |
| Q6_K | ~23 tok/s | ~28 tok/s |
| Q5_K_M | ~26 tok/s | ~33 tok/s |
| Q4_K_M | ~30 tok/s | ~38 tok/s |
| Q3_K_M | ~38 tok/s | ~47 tok/s |
| IQ2_M | ~54 tok/s | ~68 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.