Can I run Mistral 7B v0.3 on CPU only, dual-channel DDR4 (16GB)?
Yes: Mistral 7B v0.3 runs on CPU only, dual-channel DDR4 (16GB) with Q8_0 at roughly 3 tokens/s. Estimated memory 9.0 GiB at 8,192 tokens of context.
With Q8_0 you can push the context to about 25,856 tokens before memory runs out (F16 KV cache).
Memory by quantisation (8,192-token context)
| Quantisation | Memory | Verdict | Est. speed |
|---|---|---|---|
| F16Unquantized reference | 15.3 GiB | Needs CPU offload | — |
| Q8_0Near-lossless | 9.0 GiB | Fits | ~3 tok/s |
| Q6_KVery close to F16 | 7.3 GiB | Fits | ~4 tok/s |
| Q5_K_MGreat quality/size | 6.6 GiB | Fits | ~5 tok/s |
| Q4_K_MBest default for most | 5.9 GiB | Fits | ~5 tok/s |
| IQ4_XSSmaller 4-bit (imatrix) | 5.4 GiB | Fits | ~6 tok/s |
| Q3_K_MNoticeable loss | 5.1 GiB | Fits | ~6 tok/s |
| Q2_KLast resort; big loss | 4.6 GiB | Fits | ~7 tok/s |
CPU only, dual-channel DDR4 (16GB) offers about 16 GiB of memory (80% usable for a model) at 45 GB/s. Speed is a bandwidth-bound estimate of single-stream generation at 65% efficiency; real results vary by runtime, drivers and settings, so measure on your machine.
How this is calculated
Weights ≈ parameters × bits per weight ÷ 8. KV cache ≈ 2 × 32 layers × 8 KV heads × 128 head dim × 2 bytes per token. Decode speed ≈ efficiency × memory bandwidth ÷ bytes read per token. Learn the maths in VRAM math and memory bandwidth, or try the VRAM calculator.
Mistral 7B v0.3 on other hardware
Similar models on CPU only, dual-channel DDR4 (16GB)
Running slowly or out of memory? See CUDA out of memory and the quantisation lesson.