Can DeepSeek R1 Distill 70B run on NVIDIA L40 48GB?

Q: What should I upgrade first if DeepSeek R1 Distill 70B feels slow on NVIDIA L40 48GB?

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.

BARELY — Tight on Memory

B63Good

Estimated from fit model

DeepSeek R1 Distill 70B needs ~53.6 GB VRAM. NVIDIA L40 48GB has 48.0 GB. With Q4_K_M quantization, expect ~10 tok/s.

Runtime: OllamaCapacity: OffloadBandwidth: HighStack: BasicBottleneck: 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.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 53.6 GB, 10.2 tok/s, Very compromised (needs ~4.4 GB host RAM)

53.6 GB required48.0 GB available

112% VRAM needed

5.6 GB over capacity — needs offload or smaller quantization

Fit status

Very compromised (needs ~4.4 GB host RAM)

Decode

10.2 tok/s

TTFT

18959 ms

Safe context

Memory

53.6 GB / 48.0 GB

Offload

10%

Memory breakdown

Weights42.7 GB

KV Cache4.9 GB

Runtime1.2 GB

Headroom4.8 GB

See how fast it feels

See how fast it feelsDeepSeek R1 Distill 70B on NVIDIA L40 48GB
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: 10.2 tok/s decode · 19.0s TTFT (warm) · 26 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 4.4 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	B	Runs with offload (needs ~2.6 GB host RAM)	11.3 tok/s	9374 ms	4K
Coding	B	Very compromised (needs ~4.4 GB host RAM)	10.2 tok/s	18959 ms	4K
Agentic Coding	F	Too heavy	8.5 tok/s	33134 ms	4K
Reasoning	B	Very compromised (needs ~4.4 GB host RAM)	10.2 tok/s	22406 ms	4K
RAG	F	Too heavy	7.8 tok/s	45042 ms

Quantization options

How DeepSeek R1 Distill 70B (70B params) fits at each quantization level on NVIDIA L40 48GB (48.0 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	27.3 GB	Low	A74
Q3_K_SBest for your GPU	3	34.3 GB	Low	A74

Get started

Copy-paste commands to run DeepSeek R1 Distill 70B on your machine.

Run

ollama run deepseek-r1:70b

Upgrade options

Hardware that runs DeepSeek R1 Distill 70B well

NVIDIA A16 64GBBudget pick

64 GB VRAM (+16)

Removes host-memory offload, which is usually the single biggest latency and throughput win.11.9 tok/s decode

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

Adds memory headroom for longer context windows and future model growth.

~$6,500 MSRP

RTX PRO 6000 Blackwell Workstation Edition 96GBBest value

96 GB VRAM (+48)1792 GB/s (+928)

Removes host-memory offload, which is usually the single biggest latency and throughput win.38.3 tok/s decode

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

Raises estimated decode speed by about 275%.

~$9,999 MSRP

RTX PRO 6000 Blackwell Server Edition 96GBNVIDIA upgrade

96 GB VRAM (+48)1597 GB/s (+733)

Removes host-memory offload, which is usually the single biggest latency and throughput win.34.2 tok/s decode

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

Raises estimated decode speed by about 235%.

~$9,999 MSRP

Frequently asked questions

Can NVIDIA L40 48GB run DeepSeek R1 Distill 70B?

Yes, NVIDIA L40 48GB can run DeepSeek R1 Distill 70B with a B grade (Very compromised (needs ~4.4 GB host RAM)). Expected decode speed: 10.2 tok/s.

How much VRAM does DeepSeek R1 Distill 70B need?

DeepSeek R1 Distill 70B (70B parameters) requires approximately 53.6 GB of memory with Q4_K_M quantization.

What is the best quantization for DeepSeek R1 Distill 70B?

The recommended quantization for DeepSeek R1 Distill 70B is Q4_K_M, which balances quality and memory efficiency.

What speed will DeepSeek R1 Distill 70B run at on NVIDIA L40 48GB?

On NVIDIA L40 48GB, DeepSeek R1 Distill 70B achieves approximately 10.2 tokens per second decode speed with a time-to-first-token of 18959ms using Q4_K_M quantization.

Can NVIDIA L40 48GB run DeepSeek R1 Distill 70B for coding?

For coding workloads, DeepSeek R1 Distill 70B on NVIDIA L40 48GB receives a B grade with 10.2 tok/s and 4K context.

What context window can DeepSeek R1 Distill 70B use on NVIDIA L40 48GB?

On NVIDIA L40 48GB, DeepSeek R1 Distill 70B can safely use up to 4K tokens of context. The model's official context limit is 131K, but available memory constrains the safe maximum.

See all results for NVIDIA L40 48GB See all hardware for DeepSeek R1 Distill 70B

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/deepseek-r1-70b-on-l40-48gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: