Can DeepSeek V2.5 236B run on NVIDIA B200 180GB?

YES — With NVFP4

A78Great

Estimated from fit model

DeepSeek V2.5 236B needs ~209.7 GB VRAM. NVIDIA B200 180GB has 180.0 GB. With NVFP4 quantization, expect ~95 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: HighStack: StandardBottleneck: Host offload

Operating mode

Choose the run profile you care about

Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

DeepSeek V2.5 236B at Q4_K_M needs 221.5 GB — too much for NVIDIA B200 180GB (180.0 GB). Runs at NVFP4 (209.7 GB) with medium quality. 3 quantization levels fit.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 221.5 GB, exceeds 180.0 GB available

221.5 GB required180.0 GB available

123% VRAM needed

41.5 GB over capacity — needs offload or smaller quantization

Fit status

Too heavy

Decode

76.2 tok/s

TTFT

2541 ms

Safe context

Memory

221.5 GB / 180.0 GB

Offload

20%

Memory breakdown

Weights144.0 GB

KV Cache58.6 GB

Runtime0.9 GB

Headroom18.0 GB

See how fast it feels

With memory offload — actual speed may be lower

See how fast it feelsDeepSeek V2.5 236B on NVIDIA B200 180GB
1st promptCold start — includes initialization
>Write a Python function to sort a list
Here's a Python function that sorts a list using quicksort:

def quicksort(arr):
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

# Example usage
numbers = [3, 6, 8, 10, 1, 2, 1]
print(quicksort(numbers))  # [1, 1, 2, 3, 6, 8, 10]

The algorithm picks a pivot, partitions the list into smaller, equal, and larger elements, then recursively sorts each partition and concatenates the results.
2nd promptWarm — model ready, no init overhead
>Now add type hints and a docstring
def quicksort(arr: list[int]) -> list[int]:
    """Sort a list using the quicksort algorithm.

    Args:
        arr: List of integers to sort.

    Returns:
        A new sorted list.
    """
    if len(arr) <= 1:
        return arr
    pivot = arr[len(arr) // 2]
    left   = [x for x in arr if x < pivot]
    middle = [x for x in arr if x == pivot]
    right  = [x for x in arr if x > pivot]
    return quicksort(left) + middle + quicksort(right)

Added type hints for the input and return type, plus a Google-style docstring.
Estimated: 76.2 tok/s decode · 2.5s TTFT (warm) · 191 tok/s prefill

What limits this setup

It fits through host-memory offload, and offload is the main reason performance drops.

CPU or host-memory offload is active

About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.

Very little memory headroom

You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.

Best improvement path

Remove offload with more accelerator memory

Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Increase host RAM if you keep offloading

This setup may need roughly 18.7 GB of extra host RAM just for the offloaded portion, before OS and other tools.

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	A	Runs with offload (needs ~9.1 GB host RAM)	96.3 tok/s	1097 ms	5K
Coding	F	Too heavy	76.2 tok/s	2541 ms	5K
Agentic Coding	F	Too heavy	51.7 tok/s	5444 ms	5K
Reasoning	F	Too heavy	76.2 tok/s	3003 ms	5K
RAG	F	Too heavy	51.7 tok/s	6805 ms	5K

Quantization options

How DeepSeek V2.5 236B (236B params) fits at each quantization level on NVIDIA B200 180GB (180.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	92.0 GB	Low	A82
Q3_K_S	3	115.6 GB	Low	A82
NVFP4	4	132.2 GB	Medium	A82
Q4_K_MBest for your GPU	4	144.0 GB	Medium	A82
Q5_K_M	5	169.9 GB	High	F0
Q6_K	6	193.5 GB	High	F0
Q8_0	8	252.5 GB	Very High	F0
F16	16	483.8 GB	Maximum	F0

Get started

Copy-paste commands to run DeepSeek V2.5 236B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "deepseek-ai/DeepSeek-V2.5" \
  --hf-file "DeepSeek-V2.5-Q4_K_M.gguf" \
  -c 4096 -ngl 99

アップグレードオプション

DeepSeek V2.5 236Bを快適に動かすハードウェア

AMD Instinct MI350X 288GBコスパ重視

288 GB VRAM (+108)

Makes the model fit on the accelerator instead of staying completely out of reach.109.3 tok/sデコード

Makes the model fit on the accelerator instead of staying completely out of reach.

Removes host-memory offload, which is usually the single biggest latency and throughput win.

〜$8,000 MSRP

B100 192GBコスパ最良

192 GB VRAM (+12)

Makes the model fit on the accelerator instead of staying completely out of reach.84 tok/sデコード

Makes the model fit on the accelerator instead of staying completely out of reach.

〜$35,000 MSRP

NVIDIA GB200 192GBNVIDIAアップグレード

192 GB VRAM (+12)

Makes the model fit on the accelerator instead of staying completely out of reach.84 tok/sデコード

Makes the model fit on the accelerator instead of staying completely out of reach.

〜$60,000 MSRP

Frequently asked questions

Can NVIDIA B200 180GB run DeepSeek V2.5 236B?

Yes, NVIDIA B200 180GB can run DeepSeek V2.5 236B at NVFP4 quantization (Very compromised (needs ~18.7 GB host RAM)). The recommended Q4_K_M requires 221.5 GB which exceeds available memory, but at NVFP4 it needs only 209.7 GB. Expected decode speed: 95.4 tok/s.

How much VRAM does DeepSeek V2.5 236B need?

DeepSeek V2.5 236B (236B parameters) requires approximately 221.5 GB at Q4_K_M quantization. On NVIDIA B200 180GB, it fits at NVFP4 using 209.7 GB.

What is the best quantization for DeepSeek V2.5 236B?

The recommended quantization is Q4_K_M, but on NVIDIA B200 180GB the best fitting quantization is NVFP4, which uses 209.7 GB.

What speed will DeepSeek V2.5 236B run at on NVIDIA B200 180GB?

On NVIDIA B200 180GB, DeepSeek V2.5 236B achieves approximately 95.4 tokens per second decode speed with a time-to-first-token of 2030ms using NVFP4 quantization.

Can NVIDIA B200 180GB run DeepSeek V2.5 236B for coding?

For coding workloads, DeepSeek V2.5 236B on NVIDIA B200 180GB receives a F grade with 76.2 tok/s and 5K context.

What context window can DeepSeek V2.5 236B use on NVIDIA B200 180GB?

On NVIDIA B200 180GB, DeepSeek V2.5 236B can safely use up to 8K tokens of context at NVFP4 quantization. The model's official context limit is 131K, but available memory constrains the safe maximum.

What should I upgrade first if DeepSeek V2.5 236B feels slow on NVIDIA B200 180GB?

Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

See all results for NVIDIA B200 180GB See all hardware for DeepSeek V2.5 236B

Embed this result▼

Paste this snippet into any page to show a live fit card.

<iframe src="https://willitrunai.com/embed/deepseek-v2.5-236b-on-b200-180gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: