Hy Memory 30B A3B — Hardware Requirements & GPU Compatibility
ChatFunctions
Hy-Memory-30B-A3B is a specialized model from Tencent Hunyuan that acts as the memory writer in external long-term memory systems. Rather than chatting directly, it replaces the general-purpose LLM inside frameworks such as Mem0, MemOS, and Hy-Memory, deciding what to keep, how precisely to store it, and how new information should update the existing memory bank. It is a mixture-of-experts model with about 30 billion total and 3 billion active parameters, trained with on-policy distillation followed by reinforcement learning. A Q4 build fits a 24 GB GPU, and the small active set makes it quick on modest hardware. The context window is 262,144 tokens, and the license is Apache 2.0. Its card reports a 55.49 average across four memory harnesses, edging out Claude Opus 5 at 54.62 and DeepSeek V4.1 Flash at 53.64. It supports English and Chinese. It is best treated as a component for agent memory pipelines, not a general assistant.
Runs great on a 24 GB GPU
Runs from 16 GB (Q3_K_M · 16K)Best on 24 GB (Q4_K_M · ~184 tok/s · 64K)Macs from 32 GB (~32 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Hy Memory 30B A3B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
ollama run hf.co/rapatsky/Hy-Memory-30B-A3B-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 4.9 GB free · 16K context
Download: Hy-Memory-30B-A3B-Q4_K_M.gguf18.24 GBrapatsky
Other ways
each opens step-by-stepRun in cloud
Fits on RTX A5000 (24 GB) (4 GB headroom) · Q4_K_M
- Generation speed
- ~158 tok/s
- generation speed
- Cost per 1M output tokens
- $0.28
- per 1M output tokens
How Much VRAM Does Hy Memory 30B A3B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 13.2 GB | 14.7 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 128K context · ~215 tok/s | |
| 15.1 GB | 16.6 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~203 tok/s | |
| 15.4 GB | 16.9 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 128K context · ~200 tok/s | |
| 17.3 GB | 18.8 GB | 4-bit · legacy+ — 4-bit legacy quantization with offset | Fits: 128K context · ~190 tok/s | |
| 18.4 GB | 20 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 64K context · ~184 tok/s | |
| 21.8 GB | 23.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 32K context · ~168 tok/s | |
| 25.2 GB | 26.7 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 25.9 GB | |
| 30.5 GB | 32 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 31.2 GB |
Selected Q4_K_M
4-bit medium quantization19.1GBat 16K context
- Weights
- 18.0 GB
- KV cache
- 0.8 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 4.9 GB free · ~184 tok/s
≥50 tok/s needs ≥163 GB/s · RTX 5090 ~237, RTX 3090 Ti ~184
Which Hardware Can Run Hy Memory 30B A3B?
Q4_K_M · 18.4 GBHy Memory 30B A3B (Q4_K_M) requires 18.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 24+ GB is recommended. Using the full 262K context window can add up to 12.8 GB, bringing total usage to 31.2 GB. 8 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 125 devices with unified memory can run Hy Memory 30B A3B, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Mini M5 Pro (24 GB).
20 GB
fits · 0.9 GB free · up to 32K ctx
1 GPU
RX 7900 XT~74Simulate output on RX 7900 XT32–40 GB
fits · 12.9 GB free · up to 256K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~237Simulate output on RTX 5090M5 Max36 GB~65Simulate output on Mac Studio M5 Max (36 GB)M532 GB~37Simulate output on MacBook Air 13" M5 (32 GB)Orin~61Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~39Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Hy Memory 30B A3B
Hunyuan family · 6 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Gemma 4 31B ITGoogle | 106.2 | 21.7 GB | ~32 | same size |
| Llm Jp 4.1 32B A3b Thinkingllm-jp | — | 20.1 GB | ~188 | same size |
| Muse Glimmer 30BMeta | — | 19.6 GB | ~36 | same size |
| NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA | — | 19.3 GB | ~197 | same size |
| GLM 4.7 FlashZ.ai | — | 19.9 GB | ~142 | same size |
| Olmo 3.1 32B InstructAllen AI | — | 21.5 GB | ~33 | same size |
Specifications
- Publisher
- Tencent
- Family
- Hunyuan
- Parameters
- 30.1B
- Active parameters
- 3B per token
- Architecture
- HYV3ForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 120,832
- Release Date
- 2026-10-15
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Hy Memory 30B A3B need?
Hy Memory 30B A3B requires 18.4 GB of VRAM at Q4_K_M, or 60.5 GB at BF16. Full 262K context adds up to 12.8 GB (31.2 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 30.1B × 4.8 bits ÷ 8 = 18 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 13.2 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Hy Memory 30B A3B?
Yes, at Q5_K_M (21.8 GB) or lower. Higher quantizations like Q6_K (25.2 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Hy Memory 30B A3B?
For Hy Memory 30B A3B, Q4_K_M (18.4 GB) offers the best balance of quality and VRAM usage. Q5_0 (19.2 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 13.2 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Hy Memory 30B A3B on a Mac?
Yes, but only at lower quantizations. The smallest Mac that can run Hy Memory 30B A3B is MacBook Pro 14" M3 Pro (18 GB) at Q2_K; 109 of the 156 Macs we list can run it at some quantization. For Q4_K_M (18.4 GB) you need a Mac with more unified memory.
Can I run Hy Memory 30B A3B locally?
Yes — Hy Memory 30B A3B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Hy Memory 30B A3B?
At Q4_K_M, Hy Memory 30B A3B can reach ~100 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~184 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 1.8 GB active ÷ (8000 × 0.65) = 0.35 ms, plus 48 layers × 0.055 ms = 2.64 ms, so 1000 ÷ 2.99 ms = ~334 tok/s
Estimated speed at Q4_K_M (18.4 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Hy Memory 30B A3B?
At Q4_K_M, the download is about 18.04 GB. The full-precision BF16 version is 60.13 GB. The smallest option (Q2_K) is 12.78 GB.
Which GPUs can run Hy Memory 30B A3B?
8 consumer GPUs can run Hy Memory 30B A3B at Q4_K_M. Top options include NVIDIA GeForce RTX 5090, AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX. 1 GPU have plenty of headroom for comfortable inference.
Which devices can run Hy Memory 30B A3B?
125 devices with unified memory can run Hy Memory 30B A3B at Q4_K_M (18.4 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Asus ROG Flow Z13 (2025, Ryzen AI Max+ 395, 128 GB), Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.