Mamba 2.8B HF — Hardware Requirements & GPU Compatibility
Chat
Mamba 2.8B HF is a 2.8B-parameter open language model from State Spaces. At Q4_K_M it needs about 1.83 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 32K)Best on 24 GB (Q4_K_M · ~358 tok/s · 32K)Macs from 8 GB (Q4_K_M · ~39 tok/s)
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Mamba 2.8B HF
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve state-spaces/mamba-2.8b-hf --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX 3060 12GB (8 GB headroom) · Q4_K_M
- Generation speed
- ~128 tok/s
- generation speed
- Cost per 1M output tokens
- $0.15
- per 1M output tokens
How Much VRAM Does Mamba 2.8B HF Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 1.3 GB | 5.4 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 32K context · ~508 tok/s | |
| 1.5 GB | 5.6 GB | 3-bit medium — 3-bit medium quantization | Fits: 32K context · ~443 tok/s | |
| 1.8 GB | 5.9 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 32K context · ~358 tok/s | |
| 2.2 GB | 6.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 32K context · ~302 tok/s | |
| 2.5 GB | 6.6 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 32K context · ~261 tok/s | |
| 3.0 GB | 7.1 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 32K context · ~215 tok/s | |
| 6.1 GB | 10.2 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 32K context · ~108 tok/s |
Selected Q4_K_M
4-bit medium quantization3.7GBat 16K context
- Weights
- 1.3 GB
- KV cache
- 2.2 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 20.3 GB free · ~358 tok/s
≥50 tok/s needs ≥141 GB/s · RTX 5090 ~637, RTX 3090 Ti ~358
Which Hardware Can Run Mamba 2.8B HF?
Q4_K_M · 1.8 GBMamba 2.8B HF (Q4_K_M) requires 1.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 3+ GB is recommended. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Mamba 2.8B HF, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 4.3 GB free · up to 32K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~270Simulate output on RTX 3080RTX 2080 Ti~219Simulate output on RTX 2080 TiRTX 3070 Ti~216Simulate output on RTX 3070 TiM3*8 GB~39Simulate output on iMac 24" M3 (8 GB)Orin Nano~36Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~26Simulate output on iPhone 1720 GB
fits · 16.3 GB free · up to 32K ctx
1 GPU
RX 7900 XT~262Simulate output on RX 7900 XT32–40 GB
fits · 28.3 GB free · up to 32K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~637Simulate output on RTX 5090M5 Max36 GB~176Simulate output on Mac Studio M5 Max (36 GB)M532 GB~59Simulate output on MacBook Air 13" M5 (32 GB)Orin~73Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~48Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| LFM2.5 2.6BLiquid AI | — | 2.3 GB | ~321 | same size |
| Granite 3.0 2B InstructIBM | — | 2.2 GB | ~320 | same size |
| Gemma 2 2B ITGoogle | — | 2.5 GB | ~379 | same size |
| Phi 2Microsoft | — | 2.6 GB | ~248 | same size |
| Nemotron Flash 3BNVIDIA | — | 7.6 GB | ~109 | same size |
| LFM2 2.6B LongevityLiquid AI | — | 2.3 GB | ~333 | same size |
Specifications
- Publisher
- State Spaces
- Parameters
- 2.8B
- Architecture
- MambaForCausalLM
- Vocabulary Size
- 50,280
- Release Date
- 2024-03-05
Frequently Asked Questions
How much VRAM does Mamba 2.8B HF need?
Mamba 2.8B HF requires 1.8 GB of VRAM at Q4_K_M, or 6.1 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 2.8B × 4.8 bits ÷ 8 = 1.7 GB
KV Cache + Overhead ≈ 0.1 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.
VRAM usage by quantization
What's the best quantization for Mamba 2.8B HF?
For Mamba 2.8B HF, Q4_K_M (1.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (2.2 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 1.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Mamba 2.8B HF on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Mamba 2.8B HF. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 1.8 GB of usable unified memory (RAM minus macOS overhead).
Can I run Mamba 2.8B HF locally?
Yes — Mamba 2.8B HF can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Mamba 2.8B HF?
At Q4_K_M, Mamba 2.8B HF can reach ~2623 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~358 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 1.83 × 0.65 = ~2842 tok/s
Estimated speed at Q4_K_M (1.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Mamba 2.8B HF?
At Q4_K_M, the download is about 1.66 GB. The full-precision BF16 version is 5.54 GB. The smallest option (Q2_K) is 1.18 GB.
Which GPUs can run Mamba 2.8B HF?
52 consumer GPUs can run Mamba 2.8B HF at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run Mamba 2.8B HF?
163 devices with unified memory can run Mamba 2.8B HF at Q4_K_M (1.8 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (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.