Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Hy3 GPU requirements: VRAM and cheapest GPU

Hy3 has 298.8B parameters, published in BF16. It is a mixture-of-experts model: 8 of its 192 experts work on each token, about 24.4B parameters, so it generates at roughly the speed of a model that size. Every expert still has to sit in GPU memory, which is why the memory figures below are for the full model. It is released under apache-2.0, which allows commercial use.

Parameters
298.8B
Smallest footprint (INT4)
about 177 GB
Cheapest GPU per hour (FP16)
$2.39
Type
Text
  • Parameters 298.8B
  • Active per token 24.4B
  • Type Text
  • Task text-generation
  • Max context 256K
  • Licence apache-2.0
  • Downloads 885.6K
  • Likes 994

Calculating for Hy3

Precision
Workload
More settings
Serving engine
Requests at once

Hy3 needs about 661 GB of GPU memory

Inference (vLLM), FP16 / BF16, 8K tokens context, one request at a time.

Where the memory goes

  • Weights 598 GB
  • KV cache 2.7 GB
  • Runtime overhead 61 GB
  • Total 661 GB

GPU counts keep 10% of each card free for spikes. Plan extra if you expect long prompts or many users at once.

GPUs that fit, cheapest firstEvery GPU type we track that can hold this setup on one node, ranked by total hourly cost
RankConfigurationPer GPU / hrGPUs neededTotal / hrTotal / month (24/7)Provider
14 × MI300X 192GBCheapestCDNA 3 · HBM3$2.394$9.56$6,979RunPod (opens in a new tab)4x MI300X
28 × RTX PRO 6000 96GBBlackwell · GDDR7$1.808$14.40$10,512Nebius (opens in a new tab)8x RTX PRO 6000 96GB
38 × H100 NVL 94GBHopper · HBM3$3.198$25.52$18,630RunPod (opens in a new tab)8x H100 NVL
46 × H200 141GBHopper · HBM3e$3.796sold as 8 GPUs; you need 6$30.32$22,134RunPod (opens in a new tab)8x H200 NVL
53 × B300 288GBBlackwell Ultra · HBM3e$8.993sold as 4 GPUs; you need 3$35.96$26,251RunPod (opens in a new tab)4x B300
64 × GB200 186GBBlackwell · HBM3e$10.504$42.00$30,660CoreWeave (opens in a new tab)GB200 NVL72
75 × B200 180GBBlackwell · HBM3e$7.695sold as 8 GPUs; you need 5$61.52$44,910Lambda (opens in a new tab)8x HGX B200
More than 8: needs several nodes: H100 80GB, A100 80GB, L40S 48GB, RTX 6000 Ada 48GB, RTX A6000 48GB, A40 48GB, A100 40GB, RTX 5090 32GB, RTX 4090 24GB, RTX 3090 24GB, L4 24GB, A10 24GB, T4 16GB, V100 16GB
Cost per cloudCheapest offer that fits, per provider. Approximate, prices collected Nov 16, 2026.
ProviderInstancePer GPU / hrGPUs neededTotal / hrTotal / month (24/7)
RunPod (opens in a new tab)Cheapest4x MI300Xus$2.394 × MI300X 192GB about $9.56about $6,979
Nebius (opens in a new tab)8x RTX PRO 6000 96GBeu/us$1.808 × RTX PRO 6000 96GB about $14.40about $10,512
Vast.ai (opens in a new tab)8x RTX PRO 6000 S (median of 4 offers)worldwide marketplace$2.238 × RTX PRO 6000 96GB about $17.87about $13,045
CoreWeave (opens in a new tab)RTX PRO 6000 Blackwell Server Edition (High Memory)us$2.508 × RTX PRO 6000 96GB about $20.00about $14,600
Vultr (opens in a new tab)vbm-256c-2048gb-8-mi300x-gpu (bare metal)us$3.994 × MI300X 192GB sold as 8 GPUs; you need 4about $31.92about $23,302
AWS (opens in a new tab)g7e.48xlargeus-east-1$4.148 × RTX PRO 6000 96GB about $33.14about $24,195
Google Cloud (opens in a new tab)g4-standard-384us-central1$4.508 × RTX PRO 6000 96GB about $36.00about $26,280
Lambda (opens in a new tab)8x HGX B200us$7.695 × B200 180GB sold as 8 GPUs; you need 5about $61.52about $44,910
Azure (opens in a new tab)ND128isr_NDR_GB200_v6eastus$27.044 × GB200 186GB about $108.16about $78,957

Prices collected Nov 16, 2026. Every cost is approximate: on-demand, US dollars, mostly US regions (marketplaces such as Vast.ai and Lium are worldwide).

Memory by precision and workload

With a 8K-token context and one request at a time. LoRA and full fine-tuning start from 16-bit weights and QLoRA from 4-bit, so the other cells are blank.
PrecisionInferenceLoRA fine-tuneQLoRA fine-tuneFull fine-tune
FP16 / BF16661 GB771 GB–5,267 GB
INT8349 GB–––
INT4177 GB–286 GB–

Cheapest GPU by precision

Inference with an 8K context and one request at a time.

FP16 / BF16

about 661 GB · 8K tokens

The weights as published. Nothing lost, but the largest footprint.

4 × MI300X 192GB

RunPod (opens in a new tab) · 4x MI300X

Per GPU / hr
$2.39
GPUs needed
4
Total / hr
about $9.56
per month, 24/7, approx.
$6,979

INT8

about 349 GB · 8K tokens

Weights at half the size. Most chat and RAG work shows no visible difference.

3 × MI300X 192GB

RunPod (opens in a new tab) · 4x MI300X

sold as 4 GPUs; you need 3
Per GPU / hr
$2.39
GPUs needed
3
Total / hr
about $9.56
per month, 24/7, approx.
$6,979

INT4

about 177 GB · 8K tokens

Weights at a quarter of the size. Small drops show up first in maths and code.

5 × A40 48GB

RunPod (opens in a new tab) · 8x A40

sold as 8 GPUs; you need 5
Per GPU / hr
$0.59
GPUs needed
5
Total / hr
about $4.72
per month, 24/7, approx.
$3,446

Inference vs fine-tuning

  • Inference (FP16)

    1×

    of the memory inference needs

    661 GB

    4 × MI300X 192GB · about $9.56

  • LoRA fine-tune

    1.2×

    of the memory inference needs

    771 GB

    5 × MI300X 192GB · about $19.12

  • QLoRA fine-tune

    0.4×

    of the memory inference needs

    286 GB

    7 × A40 48GB · about $4.72

  • Full fine-tune

    8×

    of the memory inference needs

    5,267 GB

    more than one 8-GPU node

Serving it at FP16 fits on 4 × MI300X 192GB (RunPod). LoRA needs 5 × MI300X 192GB (RunPod), QLoRA 7 × A40 48GB (RunPod) and a full fine-tune more than one 8-GPU node. Training keeps activations and optimiser state in memory as well as the weights, which is why it so often needs more GPUs than serving the same model.

Will it run on a Mac or a single RTX 4090?

RTX 4090 (24 GB)

Not on its own. Even at INT4 it needs about 177 GB, more than the card’s 24 GB. Two or more cards, or a bigger data-centre GPU, will do it.

Mac with Apple silicon

Yes, with enough memory. At INT4 it needs about 177 GB, so a Mac with 256 GB or more of unified memory (macOS lets the GPU use about three quarters of it). Expect slower generation than on a data-centre GPU.

Similar models

Open models of a similar size.

Calculate another model

Hy3: common questions

How much VRAM does Hy3 need?

About 661 GB at FP16, 349 GB at INT8 and 177 GB at INT4, with an 8K context and one request at a time. Longer contexts and more parallel requests add KV cache on top.

What is the cheapest GPU to run Hy3?

4 × MI300X 192GB (about $2.39 per GPU / hour) on RunPod, at about $9.56 an hour for FP16 / BF16 inference with an 8K context. Prices collected Nov 16, 2026; they change often.

Can I fine-tune Hy3 on a single GPU?

Not on one GPU. Even QLoRA needs about 286 GB, more than an 80 GB GPU holds. LoRA needs about 771 GB and a full fine-tune about 5,267 GB.

How are these numbers calculated?

Weights are parameters times bytes per parameter (2 at FP16, 1 at INT8, 0.5 at INT4). The KV cache comes from the model’s layers, KV heads, head size and context length. We add about 10% plus 1 GB for the runtime and fill each GPU to 90%. Prices are on-demand list prices in US regions, collected from each provider’s public pricing.

Can Hy3 run on a consumer GPU like the RTX 4090?

Not on its own. Even at INT4 it needs about 177 GB, more than the card’s 24 GB. Two or more cards, or a bigger data-centre GPU, will do it.

Does Hy3 fit on a single H100 80GB?

No. Even at INT4 it needs about 177 GB, so you need several GPUs.

How many GPUs do I need for Hy3?

At FP16 with an 8K context: 10 × 80 GB, 16 × 48 GB or 31 × 24 GB GPUs. At INT4 you need about a quarter of the memory for the weights.

How much VRAM does Hy3 need at INT8?

About 349 GB: 314 GB for the weights, 2.7 GB for an 8K KV cache and 33 GB of runtime overhead.

How much does it cost to run Hy3 24/7 for a month?

About $6,979 on the cheapest setup we found (4 × MI300X 192GB (about $2.39 per GPU / hour), $9.56 an hour for 730 hours). The cheapest on AWS, Google Cloud or Azure is about $24,195 a month.

FP16 vs INT8 vs INT4: which should I use?

FP16 keeps full quality and needs the most memory (about 661 GB here). INT8 halves the weights with little loss for most chat work. INT4 (AWQ, GPTQ or GGUF Q4) cuts them to a quarter, about 177 GB in total, with a small drop in quality that shows most in maths and code.

Can Hy3 run on a Mac?

Yes, with enough memory. At INT4 it needs about 177 GB, so a Mac with 256 GB or more of unified memory (macOS lets the GPU use about three quarters of it). Expect slower generation than on a data-centre GPU.

vLLM or llama.cpp?

vLLM is built for serving many requests at once on NVIDIA and AMD GPUs and reserves most of the GPU memory for KV cache. llama.cpp runs GGUF files on almost anything, including Macs and CPUs, and suits one user or a small team. Memory needs are similar for the same precision.

Prices collected Nov 16, 2026. Every cost is approximate: on-demand, US dollars, mostly US regions (marketplaces such as Vast.ai and Lium are worldwide).

How we calculate

  • Weights. Parameters times bytes per parameter: 2 at FP16, 1 at INT8, 0.5 at INT4, plus about 5% for the quantisation scales at INT8 and INT4.
  • KV cache. 2 × layers × KV heads × head size × context × requests × 2 bytes, from the model’s config.json. Where a setting is missing we estimate it and say so.
  • Overhead. About 10% plus 1 GB for the runtime, in line with what vLLM reserves.
  • Fine-tuning. LoRA keeps 16-bit weights and adds a small adapter, its optimiser and activations. QLoRA does the same over 4-bit weights. A full fine-tune needs about 16 bytes per parameter with mixed-precision Adam (weights, gradients and optimiser states), plus activations.
  • GPUs. We fill each GPU to 90% of its memory and stop at 8 per node. Costs are on-demand list prices in US regions; marketplace clouds show the median offer.