Can the Instinct MI50 32GB run InternLumina-U2?
InternLumina-U2·16B·Instinct MI50 32GB·32 GB VRAM
Yes. The Instinct MI50 32GB runs InternLumina-U2 fully in VRAM at Q8_0 (about 19 GB of 32 GB), at roughly 33 tokens/s.
- best quant
- Q8_0
- needs
- 19 GB
- usable
- 32 GB
- est. speed
- ~33 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the InternLumina-U2 page →
Every quantisation of InternLumina-U2 on the Instinct MI50 32GB
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 7.1 GB | fits | ~107 tok/s |
| Q3_K_M | 9.6 GB | fits | ~72 tok/s |
| Q4_K_M | 11 GB | fits | ~58 tok/s |
| Q5_K_M | 13 GB | fits | ~49 tok/s |
| Q6_K | 15 GB | fits | ~43 tok/s |
| Q8_0 pick | 19 GB | fits | ~33 tok/s |
| FP16 / BF16 | 33 GB | offload | ~11 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 |
|---|---|---|---|---|
| 12B | Q8_0 | ≈16 GBFits in memory | ~44 tok/s | |
| 14.7B | Q8_0 | ≈17 GBFits in memory | ~36 tok/s | |
| 17.5B | Q8_0 | ≈20 GBFits in memory | ~30 tok/s | |
| 10.7B | Q8_0 | ≈8.9 GBFits in memory | ~49 tok/s | |
| 12B | Q8_0 | ≈16 GBFits in memory | ~44 tok/s | |
| 9.4B | Q8_0 | ≈10 GBFits in memory | ~57 tok/s |
Other GPUs that run InternLumina-U2
Frequently asked questions
- Can the Instinct MI50 32GB run InternLumina-U2?
- Yes. The Instinct MI50 32GB runs InternLumina-U2 fully in VRAM at Q8_0 (about 19 GB of 32 GB), at roughly 33 tokens/s.
- How fast does InternLumina-U2 run on the Instinct MI50 32GB?
- An estimate from memory bandwidth gives about 33 tokens/s at Q8_0.
- Which quantisation of InternLumina-U2 should I use on the Instinct MI50 32GB?
- Q8_0: the largest one that fits. Expected quality: 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.