Deepseek V4 Flash 0731 Spark — Hardware Requirements & GPU Compatibility
Chat
Deepseek V4 Flash 0731 Spark is a 60.3B-parameter open language model from 0xSero in the DeepSeek V4 family. It supports a context window of up to 1,048,576 tokens. At Q4_K_M it needs about 36.51 GB of VRAM — see which GPUs and Macs can run it below.
Doesn't fit a 24 GB GPU
Smallest consumer fit → Mac Studio M1 Max (32 GB) · Q2_K · 26.1 GB · ~11 tok/s
Macs from 32 GB (Q2_K · ~11 tok/s)
MIT
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Deepseek V4 Flash 0731 Spark
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve 0xSero/deepseek-v4-flash-0731-spark --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 A40 (48 GB) (11 GB headroom) · Q4_K_M
- Generation speed
- ~12 tok/s
- generation speed
- Cost per 1M output tokens
- $7.84
- per 1M output tokens
How Much VRAM Does Deepseek V4 Flash 0731 Spark Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 26.0 GB | 26.3 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Does not fit: needs 26.1 GB | |
| 29.7 GB | 30.1 GB | 3-bit medium — 3-bit medium quantization | Does not fit: needs 29.9 GB | |
| 36.5 GB | 36.9 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Does not fit: needs 36.7 GB | |
| 43.3 GB | 43.6 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Does not fit: needs 43.5 GB | |
| 50.1 GB | 50.4 GB | 6-bit, very good — 6-bit quantization, very good quality | Does not fit: needs 50.2 GB | |
| 60.6 GB | 61 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 60.8 GB | |
| 121.0 GB | 121.3 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 121.1 GB |
Selected Q4_K_M
4-bit medium quantization36.7GBat 16K context
- Weights
- 36.2 GB
- KV cache
- 0.2 GB
- Overhead
- 0.3 GB
Doesn't fit RTX 4090 (reference) · needs 12.7 GB more
Which Hardware Can Run Deepseek V4 Flash 0731 Spark?
Q4_K_M · 36.5 GBDeepseek V4 Flash 0731 Spark (Q4_K_M) requires 36.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 48+ GB is recommended. Using the full 1049K context window can add up to 11.5 GB, bringing total usage to 48.0 GB. No consumer GPU has enough memory. 67 devices with unified memory can run Deepseek V4 Flash 0731 Spark, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M5 Max (48 GB).
32–40 GB
Q4_K_M · best fit Q3_K_M (29.9 GB)
1 GPU · 2 Mac classes · 1 device family
RTX 5090~39Simulate output on RTX 5090M5 Max36 GB~11Simulate output on Mac Studio M5 Max (36 GB)M3 Pro*36 GB~4Simulate output on MacBook Pro 14" M3 Pro (36 GB)Orin~5Simulate output on NVIDIA Jetson AGX Orin 32GB48 GB
48 GB tier →fits · 11.3 GB free · up to 512K ctx
2 Mac classes · 1 device family
M5 Max48 GB~12Simulate output on Mac Studio M5 Max (48 GB)M5 Pro48 GB~6Simulate output on Mac Mini M5 Pro (48 GB)Snapdragon X2 Elite Extreme (X2E-96-100)~4Simulate output on Snapdragon X2 Elite Extreme Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Deepseek V4 Flash 0731 Spark
DeepSeek V4 family · 12 models →Similar models
2| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen2 57B A14B InstructAlibaba | — | 35.7 GB | —needs > 24 GB | same size |
| Mixtral 34Bx2 MoE 60Bcloudyu | — | 40.8 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- 0xSero
- Family
- DeepSeek V4
- Parameters
- 60.3B
- Architecture
- DeepseekV4ForCausalLM
- Context Length
- 1,048,576 tokens
- Vocabulary Size
- 129,280
- Release Date
- 2026-08-01
- License
- MIT
Frequently Asked Questions
How much VRAM does Deepseek V4 Flash 0731 Spark need?
Deepseek V4 Flash 0731 Spark requires 36.5 GB of VRAM at Q4_K_M, or 121.0 GB at BF16. Full 1049K context adds up to 11.5 GB (48.0 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 60.3B × 4.8 bits ÷ 8 = 36.2 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
Can NVIDIA GeForce RTX 5090 run Deepseek V4 Flash 0731 Spark?
Yes, at Q3_K_M (29.7 GB) or lower. Higher quantizations like Q4_K_M (36.5 GB) exceed the NVIDIA GeForce RTX 5090's 32 GB.
What's the best quantization for Deepseek V4 Flash 0731 Spark?
For Deepseek V4 Flash 0731 Spark, Q4_K_M (36.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (43.3 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 26.0 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Deepseek V4 Flash 0731 Spark on a Mac?
Yes — Mac Studio M1 Max (32 GB) and 84 other Macs can run Deepseek V4 Flash 0731 Spark, 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 36.5 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio M5 Max (48 GB).
Can I run Deepseek V4 Flash 0731 Spark locally?
Yes — Deepseek V4 Flash 0731 Spark can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Deepseek V4 Flash 0731 Spark?
At Q4_K_M, Deepseek V4 Flash 0731 Spark can reach ~132 tok/s on AMD Instinct MI350X. 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 ÷ 36.5 × 0.65 = ~142 tok/s
Estimated speed at Q4_K_M (36.5 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Deepseek V4 Flash 0731 Spark?
At Q4_K_M, the download is about 36.19 GB. The full-precision BF16 version is 120.63 GB. The smallest option (Q2_K) is 25.63 GB.
Which GPUs can run Deepseek V4 Flash 0731 Spark?
No single consumer GPU has enough VRAM to run Deepseek V4 Flash 0731 Spark at Q4_K_M (36.5 GB). Multi-GPU or professional hardware is required.
Which devices can run Deepseek V4 Flash 0731 Spark?
67 devices with unified memory can run Deepseek V4 Flash 0731 Spark at Q4_K_M (36.5 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (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.