DLM 2.0 14B FP8 — Hardware Requirements & GPU Compatibility
Chat
24 downloads
2 likesSpecifications
- Publisher
- dataslab
- Parameters
- 14.8B
- Architecture
- Qwen3ForCausalLM
- Context Length
- 40,960 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2026-03-13
- License
- Apache 2.0
Get Started
HuggingFace
How Much VRAM Does DLM 2.0 14B FP8 Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| Q4_0 | 4.00 | 8.0 GB | 14.4 GB | 7.38 GB | 4-bit legacy quantization |
| Q4_1 | 4.50 | 8.9 GB | 15.3 GB | 8.31 GB | 4-bit legacy quantization with offset |
| Q5_0 | 5.00 | 9.9 GB | 16.2 GB | 9.23 GB | 5-bit legacy quantization |
| Q5_1 | 5.50 | 10.8 GB | 17.2 GB | 10.15 GB | 5-bit legacy quantization with offset |
| Q8_0 | 8.00 | 15.4 GB | 21.8 GB | 14.77 GB | 8-bit quantization, near-lossless |
Which GPUs Can Run DLM 2.0 14B FP8?
Q4_0 · 8.0 GBDLM 2.0 14B FP8 (Q4_0) requires 8.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 11+ GB is recommended. Using the full 41K context window can add up to 6.4 GB, bringing total usage to 14.4 GB. 28 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti.
Runs great
— Plenty of headroom
Decent
— Enough VRAM, may be tightWhich Devices Can Run DLM 2.0 14B FP8?
Q4_0 · 8.0 GB27 devices with unified memory can run DLM 2.0 14B FP8, including NVIDIA DGX H100, NVIDIA DGX A100 640GB.
Runs great
— Plenty of headroomRelated Models
Frequently Asked Questions
- How much VRAM does DLM 2.0 14B FP8 need?
DLM 2.0 14B FP8 requires 8.0 GB of VRAM at Q4_0, or 15.4 GB at Q8_0. Full 41K context adds up to 6.4 GB (14.4 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 14.8B × 4 bits ÷ 8 = 7.4 GB
KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 7 GB (at full 41K context)
VRAM usage by quantization
Q4_08.0 GBQ4_0 + full context14.4 GB- What's the best quantization for DLM 2.0 14B FP8?
For DLM 2.0 14B FP8, Q5_0 (9.9 GB) offers the best balance of quality and VRAM usage. Q5_1 (10.8 GB) provides better quality if you have the VRAM. The smallest option is Q4_0 at 8.0 GB.
VRAM requirement by quantization
Q4_08.0 GB~85%Q4_18.9 GB~88%Q5_0 ★9.9 GB~90%Q5_110.8 GB~92%Q8_015.4 GB~99%★ Recommended — best balance of quality and VRAM usage.
- Can I run DLM 2.0 14B FP8 on a Mac?
DLM 2.0 14B FP8 requires at least 8.0 GB at Q4_0, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run DLM 2.0 14B FP8 locally?
Yes — DLM 2.0 14B FP8 can run locally on consumer hardware. At Q4_0 quantization it needs 8.0 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is DLM 2.0 14B FP8?
At Q4_0, DLM 2.0 14B FP8 can reach ~364 tok/s on AMD Instinct MI300X. On NVIDIA GeForce RTX 4090: ~82 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 8.0 × 0.55 = ~364 tok/s
Estimated speed at Q4_0 (8.0 GB)
AMD Instinct MI300X~364 tok/sNVIDIA GeForce RTX 4090~82 tok/sNVIDIA H100 SXM~272 tok/sAMD Instinct MI250X~225 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of DLM 2.0 14B FP8?
At Q4_0, the download is about 7.38 GB. The full-precision Q8_0 version is 14.77 GB.
- Which GPUs can run DLM 2.0 14B FP8?
28 consumer GPUs can run DLM 2.0 14B FP8 at Q4_0 (8.0 GB). Top options include AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 6900 XT, AMD Radeon RX 6700 XT. 17 GPUs have plenty of headroom for comfortable inference.
- Which devices can run DLM 2.0 14B FP8?
27 devices with unified memory can run DLM 2.0 14B FP8 at Q4_0 (8.0 GB), including Mac Mini M4 (16 GB), Mac Mini M4 (32 GB), Mac Mini M4 Pro (24 GB), Mac Mini M4 Pro (48 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.