Can DeepSeek V2.5 236B run on NVIDIA H200 PCIe 141GB?

Q: What should I upgrade first if DeepSeek V2.5 236B feels slow on NVIDIA H200 PCIe 141GB?

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.

YES — With Q2_K

A76Great

Estimated from fit model

DeepSeek V2.5 236B needs ~165.6 GB VRAM. NVIDIA H200 PCIe 141GB has 141.0 GB. With Q2_K quantization, expect ~55 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 217.6 GB — too much for NVIDIA H200 PCIe 141GB (141.0 GB). Runs at Q2_K (165.6 GB) with low quality.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 217.6 GB, exceeds 141.0 GB available

217.6 GB required141.0 GB available

154% VRAM needed

76.6 GB over capacity — needs offload or smaller quantization

Fit status

Too heavy

Decode

23.8 tok/s

TTFT

8127 ms

Safe context

Memory

217.6 GB / 141.0 GB

Offload

40%

Memory breakdown

Weights144.0 GB

KV Cache58.6 GB

Runtime0.9 GB

Headroom14.1 GB

See how fast it feels

With memory offload — actual speed may be lower

See how fast it feelsDeepSeek V2.5 236B on NVIDIA H200 PCIe 141GB
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: 23.8 tok/s decode · 8.1s TTFT (warm) · 60 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 13.7 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	F	Too heavy	32.0 tok/s	3300 ms	4K
Coding	F	Too heavy	23.8 tok/s	8127 ms	4K
Agentic Coding	F	Too heavy	14.6 tok/s	19230 ms	4K
Reasoning	F	Too heavy	23.8 tok/s	9605 ms	4K
RAG	F	Too heavy	14.6 tok/s	24037 ms	4K

Quantization options

How DeepSeek V2.5 236B (236B params) fits at each quantization level on NVIDIA H200 PCIe 141GB (141.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_KBest for your GPU	2	92.0 GB	Low	A82
Q3_K_S	3	115.6 GB	Low	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

Upgrade options

Hardware that runs DeepSeek V2.5 236B well

AMD Instinct MI350X 288GBBudget pick

288 GB VRAM (+147)8000 GB/s (+3200)

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

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 192GBBest value

192 GB VRAM (+51)8000 GB/s (+3200)

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

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

Raises estimated decode speed by about 253%.

~$35,000 MSRP

NVIDIA GB200 192GBNVIDIA upgrade

192 GB VRAM (+51)8000 GB/s (+3200)

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

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

Raises estimated decode speed by about 253%.

~$60,000 MSRP

Frequently asked questions

Can NVIDIA H200 PCIe 141GB run DeepSeek V2.5 236B?

Yes, NVIDIA H200 PCIe 141GB can run DeepSeek V2.5 236B at Q2_K quantization (Very compromised (needs ~13.7 GB host RAM)). The recommended Q4_K_M requires 217.6 GB which exceeds available memory, but at Q2_K it needs only 165.6 GB. Expected decode speed: 55.2 tok/s.

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

DeepSeek V2.5 236B (236B parameters) requires approximately 217.6 GB at Q4_K_M quantization. On NVIDIA H200 PCIe 141GB, it fits at Q2_K using 165.6 GB.

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

The recommended quantization is Q4_K_M, but on NVIDIA H200 PCIe 141GB the best fitting quantization is Q2_K, which uses 165.6 GB.

What speed will DeepSeek V2.5 236B run at on NVIDIA H200 PCIe 141GB?

On NVIDIA H200 PCIe 141GB, DeepSeek V2.5 236B achieves approximately 55.2 tokens per second decode speed with a time-to-first-token of 3505ms using Q2_K quantization.

Can NVIDIA H200 PCIe 141GB run DeepSeek V2.5 236B for coding?

For coding workloads, DeepSeek V2.5 236B on NVIDIA H200 PCIe 141GB receives a F grade with 23.8 tok/s and 4K context.

What context window can DeepSeek V2.5 236B use on NVIDIA H200 PCIe 141GB?

On NVIDIA H200 PCIe 141GB, DeepSeek V2.5 236B can safely use up to 9K tokens of context at Q2_K quantization. The model's official context limit is 131K, but available memory constrains the safe maximum.

See all results for NVIDIA H200 PCIe 141GB 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-h200-pcie-141gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: