What GPU do I need to run nvidia/Cosmos3-Super-Text2Image-4Step?
64.0B parameters, published in BF16. View on Hugging Face
Set up Cosmos3-Sup…Text2Image-4StepCosmos3-Super-Text2Image-4Step is published by nvidia on Hugging Face, with 97,850 downloads and 14 likes to date. It's a Cosmos3ForConditionalGeneration model built for text-to-image, published natively in BF16.
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision, plus a fixed overhead for activation memory and allocator fragmentation. Diffusion and video models carry no KV-cache. The real driver of extra memory is output resolution and frame count, which this flat overhead does not model. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| BF16 | 119.2 GB | 143.0 GB | AMD MI300X | 1 | $2.63/hr |
| cheaper alt. | RTX A5000 | 6 | $1.06/hr | ||
| FP8 (quantized) | 59.6 GB | 71.5 GB | RTX PRO 6000 | 1 | $1.49/hr |
| cheaper alt. | RTX 4000 SFF Ada | 4 | $0.792/hr | ||
| INT4 (quantized) | 29.8 GB | 35.8 GB | RTX A6000 | 1 | $0.363/hr |
| cheaper alt. | RTX 3060 | 3 | $0.330/hr |
A GPU is only matched to a row if its hardware supports that precision, and the primary recommendation is always a single-GPU fit when one exists.
INT4 requires a quantized checkpoint actually published for this model, check its Hugging Face page before relying on this row.
Cheapest way to run Cosmos3-Super-Text2Image-4Step at its published (BF16) precision: 1× AMD MI300X, at $2.63/hr per GPU ($2.63/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
Cosmos3-Super-Text2Image-4Step: KV cache by context length
The KV cache is the memory the attention layers hold for every token of context, on top of the weights. It is not part of the flat 1.2x overhead in the table above, grows with context length and with every concurrent request, and is why a long-context deployment needs more VRAM than the table shows.
Grouped-query attention: every layer caches keys and values for a few shared KV heads. Cached per token: 262,144 bytes at 16-bit.
| Context | KV cache, one sequence |
|---|---|
| 4K | 1.0 GB |
| 32K | 8.0 GB |
| 128K | 32.0 GB |
| 256K (model maximum) | 64.0 GB |
Computed from the layer, head and window counts in the model's own configuration (nvidia/Cosmos3-Super-Text2Image-4Step), at 16 bits per cached value. Each concurrent sequence needs its own cache, on top of the weights.
Cosmos3-Super-Text2Image-4Step: common questions
Can Cosmos3-Super-Text2Image-4Step run on a single GPU?
Yes, but not on a desktop card. At BF16 it needs 143.0 GB of VRAM, more than a single 32 GB desktop card holds. The cheapest live single-card fit is a 192.0 GB AMD MI300X at $2.63/hr.
Does quantizing Cosmos3-Super-Text2Image-4Step lower the GPU bill?
Yes. At BF16 the cheapest live fit is one AMD MI300X at $2.63/hr. At INT4 (quantized) it drops to one RTX A6000 at $0.363/hr, provided a quantized checkpoint exists for it.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Cosmos models
- Cosmos3-Super-Image2Video (64.6B, BF16)
- Cosmos-H-Surgical (15.2B, BF16)
- Cosmos-Reason2-8B (8.8B, BF16)
- Cosmos-Reason1-7B (8.3B, BF16)
- Cosmos-Reason2-2B (2.4B, BF16)
More on Cosmos3-Super-Text2Image-4Step
Fits on a 192 GB GPU at BF16: every model that fits in 192 GB.
Fits on an 80 GB GPU at FP8: every model that fits in 80 GB.
Fits on a 48 GB GPU at INT4: every model that fits in 48 GB.
Best image generation models: how Cosmos3-Super-Text2Image-4Step ranks against the rest.
Related reading: AMD MI300X pricing and specs, Fast diffusion inference on GPU VMs, and What AI inference is.
