Compare hardware for your workload
The same model and the same workload on different machines. There is no universal winner — each column lists what that machine does better and worse for this job.
Model
Quantization
Use case
Tool
Context
Add a machine
Qwen3-Coder 30B-A3B Instruct · Agentic coding / medium repository
| Machine | 64 GB unified memory · 273 GB/s | 128 GB unified memory · 614 GB/s | 256 GB unified memory · 819 GB/s | 32 GB VRAM (1792 GB/s) + 64 GB RAM | 128 GB unified memory · 256 GB/s |
|---|---|---|---|---|---|
| Comfort rating | Comfortable | Comfortable | Comfortable | Comfortable | Comfortable |
| Quantization · runtime | 5-bit · LM Studio | 6-bit · LM Studio | 6-bit · LM Studio | Q6_K · Ollama | Q6_K · Lemonade |
| Memory needed | 24 GB | 27 GB | 27 GB | 28 GB | 28 GB |
| Headroom | 28 GB | 88 GB | 216 GB | 3.5 GB | 86 GB |
| Generation | 30–40 tok/s | 45–80 tok/s | 50–70 tok/s | 150–200 tok/s | 30–40 tok/s |
| Prompt processing | 740–1000 tok/s | 4530–7560 tok/s | 2440–3300 tok/s | 12220–16530 tok/s | 770–1040 tok/s |
| First prompt | ≈30 s | ≈4 s | ≈9 s | ≈2 s | ≈25 s |
| Per turn / step | ≈15 s | ≈7 s | ≈8 s | ≈3 s | ≈15 s |
| Typical task | ≈8 min | ≈3 min | ≈4 min | ≈70 s | ≈7 min |
| Max practical context | 64K | 128K | 128K | 64K | 64K |
| Confidence | Medium confidence | Low confidence | Medium confidence | Medium confidence | Medium confidence |
| Approx. price | $2,199 | $5,599 | $6,299 | $3,028 | $2,000 |
| Tradeoffs |
|
|
|
|
|