Can the Instinct MI50 32GB run Apriel-Voice-27B-preview?
Apriel-Voice-27B-preview·27.2B·Instinct MI50 32GB·32 GB VRAM
Yes. The Instinct MI50 32GB runs Apriel-Voice-27B-preview fully in VRAM at Q6_K (about 24 GB of 32 GB), at roughly 25 tokens/s.
- best quant
- Q6_K
- needs
- 24 GB
- usable
- 32 GB
- est. speed
- ~25 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the Apriel-Voice-27B-preview page →
Every quantisation of Apriel-Voice-27B-preview on the Instinct MI50 32GB
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 11 GB | fits | ~63 tok/s |
| Q3_K_M | 15 GB | fits | ~42 tok/s |
| Q4_K_M | 19 GB | fits | ~34 tok/s |
| Q5_K_M | 21 GB | fits | ~29 tok/s |
| Q6_K pick | 24 GB | fits | ~25 tok/s |
| Q8_0 | 31 GB | tight | ~19 tok/s |
| FP16 / BF16 | 56 GB | offload | ~1.3 tok/s |
Speed is a bandwidth-based estimate for generating text (1024 GB/s); prompt processing and long contexts are slower.
Measured speed on the Instinct MI50 32GB
No community reports for this pair yet.
Report your own speed →Omni models that fit the Instinct MI50 32GB comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 35.3BA3B | Q6_K | ≈30 GBFits in memory | ~228 tok/s | |
| 24.3B | Q8_0 | ≈26 GBFits in memory | ~22 tok/s | |
| 31.7BA3B | Q6_K | ≈27 GBFits in memory | ~228 tok/s | |
| 33BA3B | Q6_K | ≈28 GBFits in memory | ~228 tok/s | |
| 31.7BA3B | Q6_K | ≈27 GBFits in memory | ~228 tok/s | |
| 17.5B | Q8_0 | ≈20 GBFits in memory | ~30 tok/s |
Other GPUs that run Apriel-Voice-27B-preview
Frequently asked questions
- Can the Instinct MI50 32GB run Apriel-Voice-27B-preview?
- Yes. The Instinct MI50 32GB runs Apriel-Voice-27B-preview fully in VRAM at Q6_K (about 24 GB of 32 GB), at roughly 25 tokens/s.
- How fast does Apriel-Voice-27B-preview run on the Instinct MI50 32GB?
- An estimate from memory bandwidth gives about 25 tokens/s at Q6_K.
- Which quantisation of Apriel-Voice-27B-preview should I use on the Instinct MI50 32GB?
- Q6_K: the largest one that fits. Expected quality: near-lossless compared with the full-precision weights.
Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.