What GPU do I need to run google/diffusiongemma-26B-A4B-it?
25.8B parameters, published in BF16. View on Hugging Face
Set up diffusiongemma-26B-A4B-itdiffusiongemma-26B-A4B-it is published by google on Hugging Face, with 1,290,354 downloads and 1,199 likes to date. It's a DiffusionGemmaForBlockDiffusion model built for image-text-to-text, published natively in BF16.
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision times a flat 1.2 overhead for activations and fragmentation. The KV-cache grows with context and is not in that factor; it is listed per model below where the architecture is published. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| BF16 | 48.1 GB | 57.7 GB | A100 | 1 | $1.10/hr |
| cheaper alt. | RTX A5000 | 3 | $0.528/hr | ||
| FP8 (quantized) | 24.1 GB | 28.9 GB | RTX 4090 | 1 | $0.470/hr |
| cheaper alt. | RTX 4080 | 2 | $0.347/hr | ||
| INT4 (quantized) | 12.0 GB | 14.4 GB | RTX A4000 | 1 | $0.167/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 diffusiongemma-26B-A4B-it at its published (BF16) precision: 1× A100, at $1.10/hr per GPU ($1.10/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
diffusiongemma-26B-A4B-it: 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.
Sliding-window attention: windowed layers stop growing at the window, the full-attention layers keep growing. Cached per token: 245,760 bytes at 16-bit while the context is inside the window.
| Context | KV cache, one sequence |
|---|---|
| 4K | 0.35 GB |
| 32K | 1.4 GB |
| 128K | 5.2 GB |
| 256K (model maximum) | 10.2 GB |
Computed from the layer, head and window counts in the model's own configuration (google/diffusiongemma-26B-A4B-it), at 16 bits per cached value. Each concurrent sequence needs its own cache, on top of the weights.
diffusiongemma-26B-A4B-it: common questions
Can diffusiongemma-26B-A4B-it run on a single GPU?
Yes, but not on a desktop card. At BF16 it needs 57.7 GB of VRAM, more than a single 32 GB desktop card holds. The cheapest live single-card fit is a 80.0 GB A100 at $1.10/hr.
What is the least VRAM diffusiongemma-26B-A4B-it can run in?
14.4 GB, at INT4 (quantized), which fits a 16 GB card, against 57.7 GB at BF16. That row assumes an INT4 checkpoint has actually been published for this model, so check its Hugging Face page before planning around it.
Does quantizing diffusiongemma-26B-A4B-it lower the GPU bill?
Yes. At BF16 the cheapest live fit is one A100 at $1.10/hr. At INT4 (quantized) it drops to one RTX A4000 at $0.167/hr, provided a quantized checkpoint exists for it.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More google models
- medasr (105M, F32)
Alternatives at this size
Other models for image-text-to-text within about a third of diffusiongemma-26B-A4B-it's 25.8B parameters, from other model lines.
- gemma-4-31B-it (31.3B, BF16)
- Qwen3.6-27B (27.8B, BF16)
- Muse-Glimmer-30B (29.8B, BF16)
- Infinity-Parser2-Pro (35.1B, BF16)
- InternVL2-26B (25.5B, BF16)
More on diffusiongemma-26B-A4B-it
Fits on an 80 GB GPU at BF16: every model that fits in 80 GB.
Fits on a 32 GB GPU at FP8: every model that fits in 32 GB.
Fits on a 16 GB GPU at INT4: every model that fits in 16 GB.
Best vision-language models: how diffusiongemma-26B-A4B-it ranks against the rest.
Related reading: A100 pricing and specs, How much VRAM you need for LLMs, Serving LLMs with vLLM, vLLM vs TensorRT-LLM vs SGLang, and Best GPU for LLM inference.
