MiniCPM SALA — Hardware Requirements & GPU Compatibility
ChatFunctions
MiniCPM SALA is a 9.5B-parameter open language model from OpenBMB in the MiniCPM family. It supports a context window of up to 524,288 tokens. At BF16 it needs about 19.32 GB of VRAM — see which GPUs and Macs can run it below.
Runs well on a 24 GB GPU
Runs on 24 GB (BF16 · ~34 tok/s · 128K)Macs from 24 GB (BF16 · ~11 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run MiniCPM SALA
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
vllm serve openbmb/MiniCPM-SALA --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3090 (24 GB) (4 GB headroom) · BF16
- Generation speed
- ~32 tok/s
- generation speed
- Cost per 1M output tokens
- $1.74
- per 1M output tokens
How Much VRAM Does MiniCPM SALA Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 19.3 GB | 20.3 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 128K context · ~34 tok/s |
Selected BF16
Brain floating point 1619.8GBat 16K context
- Weights
- 19.0 GB
- KV cache
- 0.5 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 4.2 GB free · ~34 tok/s
≥50 tok/s needs ≥1.49 TB/s · RTX 5090 ~60
Which Hardware Can Run MiniCPM SALA?
BF16 · 19.3 GBMiniCPM SALA (BF16) requires 19.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 26+ GB is recommended. Using the full 524K context window can add up to 17.1 GB, bringing total usage to 36.4 GB. 8 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 122 devices with unified memory can run MiniCPM SALA, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M1 Max (32 GB).
20 GB
fits · 0.2 GB free · up to 16K ctx
1 GPU
RX 7900 XT~25Simulate output on RX 7900 XT32–40 GB
fits · 12.2 GB free · up to 256K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~60Simulate output on RTX 5090M5 Max36 GB~17Simulate output on Mac Studio M5 Max (36 GB)M532 GB~6Simulate output on MacBook Air 13" M5 (32 GB)Orin~7Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~5Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at BF16
Alternatives to MiniCPM SALA
MiniCPM family · 18 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| SenseNova Loop 9B Previewsensenova | — | 8.2 GB | ~103 | same size |
| Granite Switch 4.1 8BIBM | — | 8.4 GB | ~111 | same size |
| ZDTaichu5.0 9BTaichuAI | — | 9.2 GB | ~101 | same size |
| Ornith 1.5 9Bornith-ai | — | 6.6 GB | ~103 | same size |
| Apertus V1.5 8Bswiss-ai | — | 8.5 GB | ~111 | same size |
| ZAYA1 VL 8BZyphra | — | 7.5 GB | ~346 | same size |
Specifications
- Publisher
- OpenBMB
- Family
- MiniCPM
- Parameters
- 9.5B
- Architecture
- MiniCPMSALAForCausalLM
- Context Length
- 524,288 tokens
- Vocabulary Size
- 73,448
- Release Date
- 2026-02-11
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does MiniCPM SALA need?
MiniCPM SALA requires 19.3 GB of VRAM at BF16. Full 524K context adds up to 17.1 GB (36.4 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 9.5B × 16 bits ÷ 8 = 19 GB
KV Cache + Overhead ≈ 0.3 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 ≈ 17.4 GB (at full 524K context)
VRAM usage by quantization
Can I run MiniCPM SALA on a Mac?
Yes — Mac Mini M5 Pro (24 GB) and 106 other Macs can run MiniCPM SALA. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 19.3 GB of usable unified memory (RAM minus macOS overhead).
Can I run MiniCPM SALA locally?
Yes — MiniCPM SALA can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is MiniCPM SALA?
At BF16, MiniCPM SALA can reach ~248 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~34 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 ÷ 19.3 × 0.65 = ~269 tok/s
Estimated speed at BF16 (19.3 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of MiniCPM SALA?
At BF16, the download is about 18.95 GB.
Which GPUs can run MiniCPM SALA?
8 consumer GPUs can run MiniCPM SALA at BF16. Top options include NVIDIA GeForce RTX 5090, AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX. 1 GPU has plenty of headroom for comfortable inference.
Which devices can run MiniCPM SALA?
122 devices with unified memory can run MiniCPM SALA at BF16 (19.3 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.