Granite Speech models
7 Granite Speech models from IBM on Hugging Face, from 473M to 8.6B parameters, published by IBM in BF16, with contexts from 4K to 128K tokens. The smallest official model, granite-speech-5.0-470m-turboctc, needs about 1.1 GB of VRAM at its published precision; the cheapest live fit is RTX 4070 Super at $0.121/hr.
Part of the Granite series
Pick a size
One row per official Granite Speech size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| granite-speech-5.0-470m-turboctc | 473M | 1.1 GB | 0.5 GB | 0.3 GB | RTX 4070 Super | $0.121/hr | not published for this architecture |
| granite-speech-4.1-2b-plus | 2.1B | 4.7 GB | 2.4 GB | 1.2 GB | RTX 4070 Super | $0.121/hr | 2.5 GB |
| granite-speech-4.1-2b | 2.3B | 5.2 GB | 2.6 GB | 1.3 GB | RTX 4070 Super | $0.121/hr | 2.5 GB |
| granite-speech-3.3-2b | 3.0B | 6.7 GB | 3.4 GB | 1.7 GB | RTX 4070 Super | $0.121/hr | 2.5 GB |
| granite-speech-3.2-8b | 8.5B | 19.0 GB | 9.5 GB | 4.7 GB | RTX A5000 | $0.176/hr | 5.0 GB |
| granite-speech-3.3-8b | 8.6B | 19.3 GB | 9.7 GB | 4.8 GB | RTX A5000 | $0.176/hr | 5.0 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (7)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| granite-speech-5.0-470m-turboctc | 473M | BF16 | 1.1 GB | RTX 4070 Super | $0.121/hr |
| granite-speech-5.0-470m-turboctc-nc | 473M | BF16 | 1.1 GB | RTX 4070 Super | $0.121/hr |
| granite-speech-4.1-2b-plus | 2.1B | BF16 | 4.7 GB | RTX 4070 Super | $0.121/hr |
| granite-speech-4.1-2b | 2.3B | BF16 | 5.2 GB | RTX 4070 Super | $0.121/hr |
| granite-speech-3.3-2b | 3.0B | BF16 | 6.7 GB | RTX 4070 Super | $0.121/hr |
| granite-speech-3.2-8b | 8.5B | BF16 | 19.0 GB | RTX A5000 | $0.176/hr |
| granite-speech-3.3-8b | 8.6B | BF16 | 19.3 GB | RTX A5000 | $0.176/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
