DeepSeekDeepSeek V4DeepseekV4ForCausalLM

DeepSeek V4 Flash DSpark — Hardware Requirements & GPU Compatibility

Chat

DeepSeek V4 Flash DSpark packs roughly 165 billion parameters into a Flash-designated variant of DeepSeek's newer V4 line, aimed at faster inference than the standard release. It is a chat-oriented model intended for general dialogue and instruction-following tasks rather than a specialized one, continuing DeepSeek's open-weight release pattern. It supports a 1M token context window, well suited to long-document and long-context workloads that need to retain a great deal of information in a single pass. It is released under the MIT license for broad local and commercial use. At this parameter count, DeepSeek V4 Flash DSpark needs multi-GPU or server-class hardware to run locally, even once quantized.

165.3B params · MoE13B active1049K contextMITJun 2026155.5K downloads300 likes369 quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

DeepSeek V4 Flash DSpark runs on a 128 GB+ Mac or a datacenter GPU

Best local pick → Mac Pro M2 Ultra (128 GB) · Q4_K_M · 99.6 GB · ~44 tok/s

Macs from 64 GB (IQ2_XXS · ~67 tok/s)

MIT

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run DeepSeek V4 Flash DSpark

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on H200 (141 GB) (41 GB headroom) · Q4_K_M

Generation speed: ~205 tok/s; generation speed
Cost per 1M output tokens: $4.86; per 1M output tokens

Deployon RunPodBest optionfrom $3.59/hr

Compare GPUs →

Run DeepSeek V4 Flash DSpark

99.6 GB VRAM at 16K context · doesn't fit the RTX 4090 (75.6 GB short)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/yailabs/DeepSeek-V4-Flash-GGUF
yailabs/DeepSeek-V4-Flash-GGUF does not list Q4_K_M; Ollama picks a quant itself (Q4_K_M if present). Check the repo's Files tab.
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/yailabs/DeepSeek-V4-Flash-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/yailabs/DeepSeek-V4-Flash-GGUF
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf yailabs/DeepSeek-V4-Flash-GGUF -c 16384 -ngl 99 --port 8080
yailabs/DeepSeek-V4-Flash-GGUF does not list Q4_K_M; llama.cpp falls back to Q4_K_M or the first file.

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/yailabs/DeepSeek-V4-Flash-GGUF
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf yailabs/DeepSeek-V4-Flash-GGUF -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve deepseek-ai/DeepSeek-V4-Flash-DSpark --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~364 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model TensorFold/DeepSeek-V4-Flash-DSpark-MLX
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model TensorFold/DeepSeek-V4-Flash-DSpark-MLX

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$yailabs/DeepSeek-V4-Flash-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
H200 SXM (141 GB)CheapestCheapest41 GB free	~205 tok/s	$4.86	Deploy on RunPodfrom $3.59/hr (opens in new tab)	141 GB · 41 GB free · ~205 tok/s · $4.86 / 1M tokens
H200 NVL (141 GB)41 GB free	~205 tok/s	$5.74	Deploy on Vast.aifrom $4.24/hr (opens in new tab)	141 GB · 41 GB free · ~205 tok/s · $5.74 / 1M tokens
B200 (192 GB)92 GB free	~258 tok/s	$6.43	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 92 GB free · ~258 tok/s · $6.43 / 1M tokens
B300 (288 GB)188 GB free	~258 tok/s	$7.46	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 188 GB free · ~258 tok/s · $7.46 / 1M tokens

Compare GPUs →

How Much VRAM Does DeepSeek V4 Flash DSpark Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
45.8 GB	46.1 GB	47.2 GB	~45.5 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Does not fit: needs 45.9 GB
70.6 GB	70.9 GB	72 GB	~70.2 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Does not fit: needs 70.7 GB
80.9 GB	81.2 GB	82.3 GB	~80.6 GB	3-bit medium — 3-bit medium quantization	Does not fit: needs 81.1 GB
99.5 GB	99.8 GB	100.9 GB	~99.2 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Does not fit: needs 99.6 GB
118.1 GB	118.4 GB	119.5 GB	~117.8 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Does not fit: needs 118.2 GB
136.7 GB	137 GB	138.1 GB	~136.3 GB	6-bit, very good — 6-bit quantization, very good quality	Does not fit: needs 136.8 GB
165.6 GB	165.9 GB	167 GB	~165.3 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Does not fit: needs 165.8 GB
330.9 GB	331.2 GB	332.3 GB	~330.5 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 331 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

99.6GBat 16K context

Weights: 99.2 GB
KV cache: 0.2 GB
Overhead: 0.3 GB

Doesn't fit RTX 4090 (reference) · needs 75.6 GB more

≥30 tok/s needs ≥388 GB/s · M3 Ultra ~45, M2 Ultra ~44

Or rent H200 from $3.59/hr.

Which Hardware Can Run DeepSeek V4 Flash DSpark?

Q4_K_M · 99.5 GB

DeepSeek V4 Flash DSpark (Q4_K_M) requires 99.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 130+ GB is recommended. Using the full 1049K context window can add up to 11.5 GB, bringing total usage to 111 GB. No consumer GPU has enough memory. 18 devices with unified memory can run DeepSeek V4 Flash DSpark, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (128 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8–16 GB · RTX 4060, RTX 3060 12GB, RTX 5060 Ti 16GBDoesn't fit — smallest quant needs 45.9 GB8 GB tier →12 GB tier →16 GB tier →

24 GB · RTX 4090referenceDoesn't fit — smallest quant needs 45.9 GB24 GB tier →

Mac 32 GB · Mac Mini M4Doesn't fit — smallest quant needs 45.9 GBMacs →

Mac 64 GB · Mac Studio M4 MaxIQ2_XXS · 45.9 GB · ~56 tok/s · 1M45.9~561MMacs →

Alternatives to DeepSeek V4 Flash DSpark

DeepSeek V4 family · 12 models →

Similar models

1tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
BloomBigScience	—	122.4 GB	—needs > 24 GB	same size

Fine-tunes & variants(2)community builds of DeepSeek V4 Flash DSpark

Specifications

Publisher: DeepSeek
Family: DeepSeek V4
Parameters: 165.3B
Active parameters: 13B per token
Architecture: DeepseekV4ForCausalLM
Context Length: 1,048,576 tokens
Vocabulary Size: 129,280
Release Date: 2026-06-27
License: MIT

Resources

Frequently Asked Questions

How much VRAM does DeepSeek V4 Flash DSpark need?

DeepSeek V4 Flash DSpark requires 99.5 GB of VRAM at Q4_K_M, or 330.9 GB at BF16. Full 1049K context adds up to 11.5 GB (111 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 165.3B × 4.8 bits ÷ 8 = 99.2 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 ≈ 11.8 GB (at full 1049K context)

VRAM usage by quantization

Q4_K_M

99.5 GB

Q4_K_M + full context

111.0 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run DeepSeek V4 Flash DSpark?

No — DeepSeek V4 Flash DSpark requires at least 45.8 GB at IQ2_XXS, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for DeepSeek V4 Flash DSpark?

For DeepSeek V4 Flash DSpark, Q4_K_M (99.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (118.1 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 45.8 GB.

VRAM requirement by quantization

IQ2_XXS

45.8 GB~53%

Q3_K_M

80.9 GB~83%

Q4_K_M ★

99.5 GB~89%

Q5_K_M

118.1 GB~92%

Q6_K

136.7 GB~95%

BF16

330.9 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run DeepSeek V4 Flash DSpark on a Mac?

Yes — Mac Pro M2 Ultra (64 GB) and 41 other Macs can run DeepSeek V4 Flash DSpark, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 99.5 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (128 GB).

Can I run DeepSeek V4 Flash DSpark locally?

Yes — DeepSeek V4 Flash DSpark can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is DeepSeek V4 Flash DSpark?

At Q4_K_M, DeepSeek V4 Flash DSpark can reach ~98 tok/s on AMD Instinct MI350X. 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 → 7.8 GB active ÷ (8000 × 0.65) = 1.50 ms, plus 43 layers × 0.055 ms = 2.37 ms, so 1000 ÷ 3.87 ms = ~258 tok/s

Estimated speed at Q4_K_M (99.5 GB)

NVIDIA B200

~258 tok/s

NVIDIA B300

~258 tok/s

NVIDIA GH200 Grace Hopper Superchip

~207 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of DeepSeek V4 Flash DSpark?

At Q4_K_M, the download is about 99.16 GB. The full-precision BF16 version is 330.53 GB. The smallest option (IQ2_XXS) is 45.45 GB.

Which GPUs can run DeepSeek V4 Flash DSpark?

No single consumer GPU has enough VRAM to run DeepSeek V4 Flash DSpark at Q4_K_M (99.5 GB). Multi-GPU or professional hardware is required.

Which devices can run DeepSeek V4 Flash DSpark?

18 devices with unified memory can run DeepSeek V4 Flash DSpark at Q4_K_M (99.5 GB), including ASUS Ascent GX10, Mac Pro M2 Ultra (128 GB), Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.