How much VRAM does Jamba 1.5 Large 398B (94B active) Instruct need?
Jamba 1.5 Large 398B (94B active) Instruct needs about 236.7 GB of VRAM at Q4_K_M, or 414.3 GB at Q8_0 and 779.1 GB at FP16. No single GPU in our set holds it even at Q4_K_M; it needs multi-GPU or offload. Weights plus runtime overhead. Budget separately for KV cache, which at 256K is a material share.
Weights-plus-runtime footprint across the six most common quantizations. KV cache is context-dependent and comes on top. At 256K it is gigabytes on its own, before the weights; size it precisely in the GPU Memory Calculator.
This is a mixture-of-experts model: all 398B parameters must sit in memory, but only about 94B are active per token — so it loads like a 398B model and runs closer to a 94B one.
Can your GPU run Jamba 1.5 Large 398B (94B active) Instruct?
Pick your card — the answer below is computed by the same engine that produces every figure on this page.
H200 141GB cannot hold Jamba 1.5 Large 398B (94B active) Instruct at Q4_K_M: it needs 236.7 GB against 133.9 GB usable; a heavier quantization, a larger card, or splitting across GPUs are the ways forward; as a mixture-of-experts model all 398B parameters must be resident even though only 94B are active per token.
Weights plus runtime overhead. KV cache is not included — it depends on context length and this model’s published architecture.
| Weights (Q4_K_M) | 224.7 GB |
| Runtime overhead | 12.0 GB |
| Estimated total | 236.7 GB |
| Usable VRAM (95% of 141 GB) | 133.9 GB |
Across every GPU we track
- RTX 3060 12GBNot enough-225.3 GB
- RTX 4060 Ti 16GBNot enough-221.5 GB
- RTX 4070 Ti SUPERNot enough-221.5 GB
- RTX 3090Not enough-213.9 GB
- RTX 4090Not enough-213.9 GB
- RTX 5090Not enough-206.3 GB
- A100 40GBNot enough-198.7 GB
- RTX A6000Not enough-191.1 GB
- L40SNot enough-191.1 GB
- A100 80GBNot enough-160.7 GB
- H100 80GBNot enough-160.7 GB
- H200 141GBNot enough-102.8 GB
Estimates, not guarantees: real usage moves with runtime, driver, batch size and context. Size a specific context window in the GPU Memory Calculator.
How was this number calculated? Every figure here comes from Bitpute’s documented calculation methodology — parameters, precision, quantization, runtime overhead and usable VRAM, each formula written out in full.
Memory by quantization
| Quant | Weights | + runtime overhead |
|---|---|---|
| FP16 | 741.3 GB | 779.1 GB |
| Q8_0 | 393.8 GB | 414.3 GB |
| Q6_K | 303.9 GB | 319.9 GB |
| Q5_K_M | 263.6 GB | 277.6 GB |
| Q4_K_M | 224.7 GB | 236.7 GB |
| Q4_0 | 208.5 GB | 219.7 GB |
Overhead = 0.75 GB + 5% of weights (CUDA context, buffers). Add KV cache on top. Weights dominate at this size, but a 256K window still costs gigabytes.
Which quantization should you actually run?
Jamba 1.5 Large 398B (94B active) Instruct is a multi-GPU or datacenter model: even Q4_K_M needs about 236.7 GB, beyond any single card. Plan for several 80 GB GPUs or a hosted endpoint; FP16 (779.1 GB) is server-only. As a mixture-of-experts model it must hold all 398B in VRAM but activates only about 94B per token, so it runs much faster than its footprint suggests — memory is the limit here, not speed.
Single-GPU fit at Q4_K_M (236.7 GB + KV)
RTX 3060 12GBRTX 4060 Ti 16GBRTX 4070 Ti SUPERRTX 3090RTX 4090RTX 5090A100 40GBRTX A6000L40SA100 80GBH100 80GBH200 141GB
Common questions
How much VRAM does Jamba 1.5 Large 398B (94B active) Instruct need?
Jamba 1.5 Large 398B (94B active) Instruct needs about 236.7 GB of VRAM at Q4_K_M, 414.3 GB at Q8_0, or 779.1 GB at FP16. That is weights plus about 0.75 GB of runtime overhead and 5% of weight size; KV cache is additional and depends on context length.
What GPU can run Jamba 1.5 Large 398B (94B active) Instruct?
No single GPU in our set fits it at Q4_K_M; it needs multiple GPUs or CPU offload. At FP16 no single GPU in our set is sufficient.
Can Jamba 1.5 Large 398B (94B active) Instruct run on 24 GB?
Not at Q4_K_M: it needs about 236.7 GB, more than the ~22.8 GB usable on a 24 GB card. Use a larger card, multiple GPUs, or offload.