Internlm Chat 20B — Hardware Requirements & GPU Compatibility
Chat
InternLM-Chat-20B is a 20-billion-parameter chat model from a collaboration between the Shanghai AI Laboratory, SenseTime, CUHK, and Fudan University, built by applying supervised fine-tuning and RLHF on top of the InternLM-20B base model. InternLM-20B used a deliberately deep 60-layer architecture, versus the 32-40 layers typical of 7B/13B models of its era, trained on over 2.3 trillion tokens of English, Chinese, and code data, giving it stronger reasoning, math, and coding scores than contemporaries like Llama 2 13B and Baichuan2-13B. As a 20-billion-parameter dense model, it fits on a single consumer GPU once quantized. Context length is 4,096 tokens natively, extendable to about 16,384 tokens through inference-time extrapolation, per the model card. It is released under the Apache 2.0 license for the code, with model weights free for academic research and available for free commercial use after applying for a license from the developers. It was published in September 2023, predating the InternLM2 generation.
Runs well on a 24 GB GPU
Runs from 16 GB (Q4_K_M · 2K)Best on 24 GB (Q4_K_M · ~44 tok/s · 4K)Macs from 18 GB (Q2_K · ~9 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Internlm Chat 20B
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 internlm/internlm-chat-20b --max-model-len 4096vLLM 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) (6 GB headroom) · Q4_K_M
- Generation speed
- ~41 tok/s
- generation speed
- Cost per 1M output tokens
- $1.34
- per 1M output tokens
How Much VRAM Does Internlm Chat 20B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 11.3 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 4K context · ~58 tok/s | |
| 12.6 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 4K context · ~52 tok/s | |
| 14.8 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 4K context · ~44 tok/s | |
| 17.1 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 4K context · ~38 tok/s | |
| 19.3 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 4K context · ~34 tok/s | |
| 22.8 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 2K context · ~29 tok/s | |
| 42.8 GB | — | 16-bit · full — Full half-precision — baseline for inference | Does not fit: needs 42.8 GB |
Selected Q4_K_M
4-bit medium quantization17.3GBat 4K context
- Weights
- 12.0 GB
- KV cache
- 5.0 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 6.7 GB free · ~44 tok/s
≥50 tok/s needs ≥1.14 TB/s · RTX 5090 ~79
Which Hardware Can Run Internlm Chat 20B?
Q4_K_M · 14.8 GBInternlm Chat 20B (Q4_K_M) requires 14.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 20+ GB is recommended. Using the full 4K context window can add up to 2.5 GB, bringing total usage to 17.3 GB. 26 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 122 devices with unified memory can run Internlm Chat 20B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M5 Pro (24 GB).
20 GB
fits · 2.7 GB free · up to 4K ctx
1 GPU
RX 7900 XT~32Simulate output on RX 7900 XT32–40 GB
fits · 14.7 GB free · up to 4K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~79Simulate output on RTX 5090M5 Max36 GB~22Simulate output on Mac Studio M5 Max (36 GB)M532 GB~7Simulate output on MacBook Air 13" M5 (32 GB)Orin~9Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~6Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Internlm Chat 20B?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
Here are all 3 boards that have results for this model, sorted by their weight in the llmrun Score; badges show who ran the test.
- MMLU
Epoch AI · Sep 2023Lab-reported
Lab-reported57.4%
#52/85 open
- BIG-Bench Hard
Epoch AI · Sep 2023Lab-reported
Lab-reported42.4%
#27/38 open
- GSM8K
Epoch AI · Sep 2023Lab-reported
Lab-reported15.7%
#57/62 open
Alternatives to Internlm Chat 20B
InternLM family · 7 models →Similar models
4| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| InternVL3 5 GPT OSS 20B A4B Preview HFOpenGVLab | — | 13.6 GB | ~180 | same size |
| Tri 21B Thinktrillionlabs | — | 45.1 GB | —needs > 24 GB | same size |
| ERNIE 4.5 21B A3B PTBaidu | — | 13.8 GB | ~254 | same size |
| GPT Neox 20BEleutherAI | — | 13.7 GB | ~48 | same size |
Specifications
- Publisher
- InternLM
- Family
- InternLM
- Parameters
- 20B
- Architecture
- InternLMForCausalLM
- Context Length
- 4,096 tokens
- Vocabulary Size
- 103,168
- Release Date
- 2023-09-18
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Internlm Chat 20B need?
Internlm Chat 20B requires 14.8 GB of VRAM at Q4_K_M, or 42.8 GB at FP16. Full 4K context adds up to 2.5 GB (17.3 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 20B × 4.8 bits ÷ 8 = 12 GB
KV Cache + Overhead ≈ 2.8 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 ≈ 5.3 GB (at full 4K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Internlm Chat 20B?
Yes, at Q8_0 (22.8 GB) or lower. Higher quantizations like FP16 (42.8 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Internlm Chat 20B?
For Internlm Chat 20B, Q4_K_M (14.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (17.1 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 11.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Internlm Chat 20B on a Mac?
Yes — MacBook Pro 14" M3 Pro (18 GB) and 108 other Macs can run Internlm Chat 20B, the smallest at Q2_K. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 14.8 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M5 Pro (24 GB).
Can I run Internlm Chat 20B locally?
Yes — Internlm Chat 20B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Internlm Chat 20B?
At Q4_K_M, Internlm Chat 20B can reach ~324 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~44 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 ÷ 14.8 × 0.65 = ~351 tok/s
Estimated speed at Q4_K_M (14.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Internlm Chat 20B?
At Q4_K_M, the download is about 12.00 GB. The full-precision FP16 version is 40.00 GB. The smallest option (Q2_K) is 8.50 GB.
Which GPUs can run Internlm Chat 20B?
26 consumer GPUs can run Internlm Chat 20B at Q4_K_M. Top options include AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, NVIDIA GeForce RTX 3090 Ti, AMD Radeon RX 6800. 7 GPUs have plenty of headroom for comfortable inference.
Which devices can run Internlm Chat 20B?
122 devices with unified memory can run Internlm Chat 20B at Q4_K_M (14.8 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.