What GPU do I need to run tencent/Hy3-FP8?
298.8B parameters, published in F8_E4M3. View on Hugging Face
Set up Hy3-FP8Hy3-FP8 is published by tencent on Hugging Face, with 40,013 downloads and 70 likes to date. It's a HYV3ForCausalLM model built for text-generation, published natively in F8_E4M3.
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision, plus a fixed overhead for KV-cache, activations, and fragmentation. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| FP8 (native) | 278.3 GB | 333.9 GB | RTX 4090 | 7 | $3.39/hr |
| INT4 (quantized) | 139.1 GB | 167.0 GB | RTX A5000 | 7 | $1.23/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 Hy3-FP8 at its published (F8_E4M3) precision: 7× RTX 4090, at $0.485/hr per GPU ($3.39/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
Hy3-FP8: common questions
Can Hy3-FP8 run on a single GPU?
No. At FP8 (native) it needs 333.9 GB of VRAM, more than a single 32 GB desktop card holds. The cheapest capable card in the live feed is a 48.0 GB RTX 4090, and it takes 7 of them.
Is Hy3-FP8 already quantized?
Yes. It is published in FP8, one byte per parameter, so the 333.9 GB figure is already a quantized footprint rather than a full-precision one. Only INT4 goes below it, at 167.0 GB. A GPU without FP8 tensor cores cannot run it as published, which is why cards here are matched on precision support and not on VRAM alone.
How many GPUs do I need to run Hy3-FP8?
7 at FP8 (native). It needs 333.9 GB of VRAM and the cheapest capable live offer is a 48.0 GB RTX 4090, so 7 of them come to $3.39/hr in total.
Does quantizing Hy3-FP8 lower the GPU bill?
Yes. At FP8 (native) the cheapest live fit is 7 RTX 4090 cards at $3.39/hr. At INT4 (quantized) it drops to 7 RTX A5000 cards at $1.23/hr, provided a quantized checkpoint exists for it.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Hy3 models
- Hy3-preview (298.8B, BF16)
- Hy3 (298.8B, BF16)
Related reading: RTX 4090 pricing and specs, and The best GPUs for AI, ranked.
