DeepSeek V4 Flash 0731 VRAM requirements
DeepSeek V4 Flash 0731 needs about 190 GB of VRAM at Q4_K_M with 8K context, so it needs a 640 GB setup such as 8× H100 SXM; a 24 GB card is too small. DeepSeek V4 Flash 0731 has 304.2B parameters (DeepSeek-V4-Flash-Vision-Exp has the same text config plus a vision encoder; its checkpoint is 466M parameters (0.9 GB) larger, so these figures cover it too), of which about 13B are used per token; all 256 experts still have to be in memory. With the same 8K context and one request it needs 313 GB at FP8 and 624 GB at FP16/BF16. The published weights take 155 GB (4.4 bits/weight). It does not fit on a 24 GB card at Q4_K_M; the smallest setup here that holds it with an 8K context is the M3 Ultra Mac Studio (512 GB, 384 GB usable).
190 GB at Q4_K_M, 8K context, one request
- FP8
- 313 GB
- FP16 / BF16
- 624 GB
- Published weights
- 155 GB
- Smallest setup, Q4_K_M
- M3 Ultra Mac Studio (512 GB, 384 GB usable)
Worked example: DeepSeek V4 Flash 0731 with 32K tokens
Inputs: DeepSeek V4 Flash 0731 · Q4_K_M weights · 32,768 tokens of context · one request · FP16 KV cache
- Weights (304.2B at Q4_K_M)
- 171 GB
- KV cache (86 KB per token)
- 2.69 GB
- Buffers and runtime (0.5 GB + 10%)
- 17.9 GB
- Total
- 192 GB
The smallest setup here that holds it is the M3 Ultra Mac Studio (512 GB, 384 GB usable), with about 192 GB to spare. Change the inputs in the calculator
What makes DeepSeek V4 Flash 0731's memory use different
As a mixture-of-experts model it reads about 7.32 GB of its 171 GB Q4_K_M weights per generated token (4%), so it writes like a much smaller model while needing memory for all of them.
How much VRAM does DeepSeek V4 Flash 0731 need?
DeepSeek V4 Flash 0731 needs 190 GB at Q4_K_M, 332 GB at Q8_0 and 624 GB at FP16 with 8,192 tokens of context and one request. Each total below is the weights plus the FP16 KV cache and the runtime overhead (0.5 GB plus 10%), and each row opens the calculator with that setting. What the GGUF names mean.
| Precision | Weights | Total | Smallest setup |
|---|---|---|---|
| As published (4.4 bits/weight) | 155 GB | 172 GB | B200 (180 GB) |
| FP16 / BF16 | 567 GB | 624 GB | 8× H100 SXM (640 GB) |
| FP8 / INT8 | 283 GB | 313 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| INT4 (AWQ / GPTQ) | 150 GB | 167 GB | B200 (180 GB) |
| GGUF Q8_0 | 301 GB | 332 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF Q6_K | 232 GB | 257 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF Q5_K_M | 201 GB | 222 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF Q4_K_M | 171 GB | 190 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| GGUF IQ4_XS | 154 GB | 171 GB | B200 (180 GB) |
| GGUF Q3_K_M | 138 GB | 154 GB | B200 (180 GB) |
| GGUF IQ3_XXS | 117 GB | 130 GB | H200 (141 GB) |
| GGUF Q2_K | 119 GB | 132 GB | H200 (141 GB) |
Fine-tuning DeepSeek V4 Flash 0731? DeepSeek V4 Flash 0731 VRAM for LoRA, QLoRA and full training.
DeepSeek V4 Flash 0731 GGUF files on Hugging Face
The 3 GGUF files llama.cpp picks for DeepSeek V4 Flash 0731 with -hf <repo>:<QUANT>, with the bytes the Hugging Face file list reported on 2026-09-29. Each total is that file plus the FP16 KV cache and 0.5 GB + 10%, one request. "vs estimate" compares the file with the bits-per-weight size the table above uses.
| Quant | File | Size | vs estimate | 8K | 32K | 128K |
|---|---|---|---|---|---|---|
| Q2_K | ggml-org/ | 109 GB117,349,438,592 bytes | −8% | 121 GB | 124 GB | 133 GB |
| Q3_K_M | unsloth/ | 119 GB128,078,484,032 bytes | −14% | 132 GB | 135 GB | 144 GB |
| IQ4_XS | unsloth/ | 127 GB136,662,446,656 bytes | −17% | 141 GB | 143 GB | 152 GB |
GB here is GiB (1024³ bytes), as everywhere on this site.
Which DeepSeek V4 Flash 0731 GGUF fits a 16, 24, 32, 48 or 80 GB GPU or a Mac
From the file sizes above: the largest file that fits each machine with an 8K context and 0.5 GB left free, the longest context it then has room for, and the longest context for Q4_K_M. One request, FP16 KV cache.
| Hardware | Largest GGUF, 8K | Its longest context | Q4_K_M, longest context |
|---|---|---|---|
| RTX 5060 Ti 16GB | None fits | — | — |
| RTX 4090 (24 GB) | None fits | — | — |
| RTX 5090 (32 GB) | None fits | — | — |
| L40S (48 GB) | None fits | — | — |
| H100 SXM (80 GB) | None fits | — | — |
| M4 Pro Mac (64 GB, 48 GB usable) | None fits | — | — |
| M4 Max Mac (128 GB, 96 GB usable) | None fits | — | — |
DeepSeek V4 Flash 0731 at 8K, 32K, 128K and 1M (full) tokens of context
The tables use an independent-layer upper-bound proxy for its 43-layer shared, compressed cache. Each extra token of context adds 86 KB of FP16 cache per request. The last column is the smallest setup here, one card or a group, that holds it at Q4_K_M. How the KV cache works.
| Context | KV cache, FP16 | Total at Q4_K_M | Total at Q8_0 | Smallest setup, Q4_K_M |
|---|---|---|---|---|
| 8K tokens | 688 MB | 190 GB | 332 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| 32K tokens | 2.69 GB | 192 GB | 335 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| 128K tokens | 10.8 GB | 201 GB | 343 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| 1M tokens | 86.0 GB | 284 GB | 426 GB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
How fast DeepSeek V4 Flash 0731 writes
Tokens per second for one request with 8,192 tokens of context, estimated from memory bandwidth and the parameters read per token. A dash means it does not fit on one card. Try other settings in the speed calculator.
| Hardware | Bandwidth | Q4_K_M | FP8 |
|---|---|---|---|
| RTX 3060 12GB | 360 GB/s | — | — |
| RTX 4090 | 1,008 GB/s | — | — |
| RTX 5090 | 1,792 GB/s | — | — |
| M4 Max Mac (128 GB) | 546 GB/s | — | — |
| M3 Ultra Mac Studio (512 GB) | 819 GB/s | 27–47 | 17–29 |
| H100 SXM | 3,350 GB/s | — | — |
| H200 | 4,800 GB/s | — | — |
Longest context on a multi-GPU group
No single card in this table holds it at Q4_K_M, Q8_0 or FP8, so these rows use GPU groups: how many tokens of context DeepSeek V4 Flash 0731 fits on each, in one tensor-parallel group with one request, an FP16 KV cache and 0.5 GB left free. "Full" means the model's whole context window fits.
| Group | Memory | Q4_K_M | Q8_0 | FP8 |
|---|---|---|---|---|
| 2× RTX 3060 12GB | 24 GB | No | No | No |
| 2× RTX 3090 | 48 GB | No | No | No |
| 2× RTX 4090 | 48 GB | No | No | No |
| 2× RTX 5090 | 64 GB | No | No | No |
| 4× RTX 3090 | 96 GB | No | No | No |
| 8× H100 SXM | 640 GB | 1M (full) | 1M (full) | 1M (full) |
| 8× H200 | 1,128 GB | 1M (full) | 1M (full) | 1M (full) |
Model details
- Parameters
- 304.2B (304,180,418,494); DeepSeek-V4-Flash-Vision-Exp has the same text config plus a vision encoder; its checkpoint is 466M parameters (0.9 GB) larger, so these figures cover it too
- Experts
- 256 routed experts, all loaded
- Active per token
- 13B
- Layers
- The tables use an independent-layer upper-bound proxy for its 43-layer shared, compressed cache
- Attention cache
- 1 KV heads × 512
- Context length
- 1,048,576 tokens
- Published weights
- 155 GB (4.4 bits/weight)
- On Hugging Face
- deepseek-ai/
DeepSeek-V4-Flash-0731 - Reported benchmarks
- None published in the model card for GPQA, MMLU-Pro or LiveCodeBench.
Why this estimate looks this way
At Q4_K_M and an 8K-token context, DeepSeek V4 Flash 0731 uses 171 GB for weights, 688 MB for its FP16 KV cache and 17.7 GB for estimated runtime overhead, totaling 190 GB. The overhead is a 0.5 GB base plus 10% of weights and cache.
This is a mixture-of-experts model: all 256 experts contribute to the 304.2B parameters held in memory, even though only about 13B parameters run per token. Using only active parameters would understate VRAM.
This model shares and compresses its KV cache across layers, which the calculator does not model, so the KV cache figure is an upper bound.
The architecture and published checkpoint size come from the model's config.json and weight files. These are estimates rather than measured peak memory; inference engines can reserve extra buffers or preallocate the full configured cache.
DeepSeek V4 Flash 0731 next to similar-sized models
The three models here closest to it in parameter count, at Q4_K_M with 8,192 tokens of context.
| Model | Parameters | Total | KV per token | Smallest setup |
|---|---|---|---|---|
| DeepSeek V4 Flash 0731 | 304.2B, 13B active | 190 GB | 86 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| MiMo V2.6 Flash | 310.8B, 15B active | 193 GB | 23 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| DeepSeek V4 Flash | 290.9B, 13B active | 182 GB | 86 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
| IQuest-Q1 320B-A15B | 320.3B, 15B active | 201 GB | 100 KB | M3 Ultra Mac Studio (512 GB, 384 GB usable) |
DeepSeek V4 Flash 0731 VRAM questions
How much VRAM do I need to run DeepSeek V4 Flash 0731 locally?
About 190 GB at Q4_K_M with an 8K context and one request, by the bits-per-weight estimate, and 624 GB at BF16. The smallest setup listed here that holds Q4_K_M is the M3 Ultra Mac Studio (512 GB, 384 GB usable).
Can I run DeepSeek V4 Flash 0731 on a 24 GB GPU?
No. The RTX 4090 (24 GB) alone cannot hold it: the smallest listed GGUF, Q2_K at 109 GB, needs 121 GB with an 8K context.
Which quantization of DeepSeek V4 Flash 0731 fits in 16 GB of VRAM?
The RTX 5060 Ti 16GB alone cannot hold it: the smallest listed GGUF, Q2_K at 109 GB, needs 121 GB with an 8K context.
Can I run DeepSeek V4 Flash 0731 on a Mac?
The M4 Pro Mac (64 GB, 48 GB usable) alone cannot hold it: the smallest listed GGUF, Q2_K at 109 GB, needs 121 GB with an 8K context. The M4 Max Mac (128 GB, 96 GB usable) alone cannot hold it: the smallest listed GGUF, Q2_K at 109 GB, needs 121 GB with an 8K context. By default macOS lets the GPU use about 75% of unified memory.
How much more VRAM does DeepSeek V4 Flash 0731 need for 32K or 128K tokens of context?
Its FP16 KV cache is 688 MB at 8K, 2.69 GB at 32K, 10.8 GB at 128K for one request, so 128K adds 10.1 GB over 8K, plus 10% overhead. A q8_0 cache
Does DeepSeek V4 Flash 0731 need less VRAM because only 13B parameters are active?
No. All 256 experts stay loaded, so the Q4_K_M weights take 171 GB. The active parameters decide speed: each generated token reads about 7.32 GB of them.
Badge for your model card
Paste it into a Hugging Face or GitHub README; it shows the Q4_K_M total above and links to this page. Q8_0 and FP16 badges.
Numbers read from the model files on Hugging Face on . Compare it with other models in the reproducible model and GPU report.