What GPU do I need to run microsoft/Phi-3.5-MoE-instruct?
41.9B parameters, published in BF16. View on Hugging Face
Set up Phi-3.5-MoE-instructPhi-3.5-MoE-instruct is published by microsoft on Hugging Face, with 134,117 downloads and 575 likes to date. It's a PhiMoEForCausalLM model built for text-generation, 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 | 78.0 GB | 93.6 GB | RTX PRO 6000 WS | 1 | $1.62/hr |
| cheaper alt. | RTX A5000 | 4 | $0.704/hr | ||
| FP8 (quantized) | 39.0 GB | 46.8 GB | L40 | 1 | $0.759/hr |
| cheaper alt. | RTX 4000 SFF Ada | 3 | $0.594/hr | ||
| INT4 (quantized) | 19.5 GB | 23.4 GB | RTX A5000 | 1 | $0.176/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 Phi-3.5-MoE-instruct at its published (BF16) precision: 1× RTX PRO 6000 WS, at $1.62/hr per GPU ($1.62/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
Phi-3.5-MoE-instruct: 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.
KV-cache size for Phi-3.5-MoE-instruct is not published for this architecture, so no figure is shown.
Phi-3.5-MoE-instruct: common questions
Can Phi-3.5-MoE-instruct run on a single GPU?
Yes, but not on a desktop card. At BF16 it needs 93.6 GB of VRAM, more than a single 32 GB desktop card holds. The cheapest live single-card fit is a 96.0 GB RTX PRO 6000 WS at $1.62/hr.
What is the least VRAM Phi-3.5-MoE-instruct can run in?
23.4 GB, at INT4 (quantized), which fits a 24 GB card, against 93.6 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 Phi-3.5-MoE-instruct lower the GPU bill?
Yes. At BF16 the cheapest live fit is one RTX PRO 6000 WS at $1.62/hr. At INT4 (quantized) it drops to one RTX A5000 at $0.176/hr, provided a quantized checkpoint exists for it.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Phi-3.5 models
- Phi-3.5-vision-instruct (4.1B, BF16)
- VLM2Vec-Full (4.1B, BF16)
- Phi-3.5-mini-instruct (3.8B, BF16)
Alternatives at this size
Other models for text-generation within about a third of Phi-3.5-MoE-instruct's 41.9B parameters, from other model lines.
- Qwen3-32B (32.8B, BF16)
- GLM-4.7-Flash (31.2B, BF16)
- NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 (31.6B, BF16)
- granite-4.1-30b (28.9B, BF16)
- Ornith-1.5-35B-A3B (36.0B, BF16)
More on Phi-3.5-MoE-instruct
Fits on a 96 GB GPU at BF16: every model that fits in 96 GB.
Fits on a 48 GB GPU at FP8: every model that fits in 48 GB.
Fits on a 24 GB GPU at INT4: every model that fits in 24 GB.
Best chat and assistants models: how Phi-3.5-MoE-instruct ranks against the rest.
Compare Phi-3.5-MoE-instruct with Qwen-72B, Llama-3.3-70B-Instruct, Llama-3.1-70B-Instruct, Qwen3-Coder-Next, Qwen2.5-72B-Instruct.
Related reading: RTX PRO 6000 WS 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.
