Is unified memory on NVIDIA DGX Spark 128GB as fast as VRAM for Codestral 22B?

Not always. NVIDIA DGX Spark 128GB can often fit larger models thanks to unified memory, but a discrete GPU with dedicated high-bandwidth VRAM may still decode faster once the model fits. For this combination, the important distinction is capacity versus sustained throughput.

Can Codestral 22B run on NVIDIA DGX Spark 128GB?

Q: What should I upgrade first if Codestral 22B feels slow on NVIDIA DGX Spark 128GB?

Prioritize bandwidth, not only capacity. If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.

YES — With F16

B55Good

Estimated from fit model

Codestral 22B needs ~61.8 GB VRAM. NVIDIA DGX Spark 128GB has 0 MB. With F16 quantization, expect ~6 tok/s.

Runtime: OllamaCapacity: RoomyBandwidth: LowStack: BasicBottleneck: Memory bandwidth

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.

Codestral 22B at Q4_K_M needs 17.1 GB — too much for NVIDIA DGX Spark 128GB (0.0 GB). Runs at F16 (61.8 GB) with maximum quality. 8 quantization levels fit.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 30.1 GB, 13.1 tok/s, Runs well

30.1 GB required108.8 GB available

28% VRAM used

Fit status

Runs well

Decode

13.1 tok/s

TTFT

14755 ms

Safe context

33K

Memory

30.1 GB / 108.8 GB

Memory breakdown

Weights13.4 GB

KV Cache2.4 GB

Runtime1.2 GB

Headroom13.1 GB

See how fast it feels

See how fast it feelsCodestral 22B on NVIDIA DGX Spark 128GB
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: 13.1 tok/s decode · 14.8s TTFT (warm) · 33 tok/s prefill

What limits this setup

The model fits in shared memory, but shared-memory bandwidth is now the real limiter.

Fit does not mean dedicated-VRAM speed

Unified or shared memory can make a model technically fit, but sustained tokens per second may still trail a discrete high-bandwidth GPU with less total memory.

Shared-memory contention still exists

The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.

Best improvement path

Prioritize bandwidth, not only capacity

If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.

Performance by workload

Workload	Grade	Fit	Decode	TTFT	Context
Chat	F	Too heavy	2.2 tok/s	48065 ms	4K
Coding	F	Too heavy	2.2 tok/s	88119 ms	4K
Agentic Coding	F	Too heavy	2.2 tok/s	128174 ms	4K
Reasoning	F	Too heavy	2.2 tok/s	104141 ms	4K
RAG	F	Too heavy	2.2 tok/s	160217 ms	4K

Quantization options

How Codestral 22B (22B params) fits at each quantization level on NVIDIA DGX Spark 128GB (92.2 GB usable).

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	8.6 GB	Low	C50
Q3_K_S	3	10.8 GB	Low	C50
NVFP4	4

Get started

Copy-paste commands to run Codestral 22B on your machine.

Run

ollama run codestral

Upgrade options

Hardware that runs Codestral 22B well

NVIDIA H200 141GBBudget pick

141 GB VRAM (+13)4800 GB/s (+4527)

Raises estimated decode speed by about 2251%.308 tok/s decode

Raises estimated decode speed by about 2251%.

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

~$30,000 MSRP

NVIDIA H200 PCIe 141GBBest value

141 GB VRAM (+13)4800 GB/s (+4527)

Raises estimated decode speed by about 2251%.308 tok/s decode

Raises estimated decode speed by about 2251%.

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

~$30,000 MSRP

NVIDIA B200 180GBNVIDIA upgrade

180 GB VRAM (+52)8000 GB/s (+7727)

Raises estimated decode speed by about 2251%.308 tok/s decode

Raises estimated decode speed by about 2251%.

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

~$30,000 MSRP

Frequently asked questions

Can NVIDIA DGX Spark 128GB run Codestral 22B?

Yes, NVIDIA DGX Spark 128GB can run Codestral 22B at F16 quantization (Runs well). The recommended Q4_K_M requires 17.1 GB which exceeds available memory, but at F16 it needs only 61.8 GB. Expected decode speed: 5.5 tok/s.

How much VRAM does Codestral 22B need?

Codestral 22B (22B parameters) requires approximately 17.1 GB at Q4_K_M quantization. On NVIDIA DGX Spark 128GB, it fits at F16 using 61.8 GB.

What is the best quantization for Codestral 22B?

The recommended quantization is Q4_K_M, but on NVIDIA DGX Spark 128GB the best fitting quantization is F16, which uses 61.8 GB.

What speed will Codestral 22B run at on NVIDIA DGX Spark 128GB?

On NVIDIA DGX Spark 128GB, Codestral 22B achieves approximately 5.5 tokens per second decode speed with a time-to-first-token of 35419ms using F16 quantization.

Can NVIDIA DGX Spark 128GB run Codestral 22B for coding?

For coding workloads, Codestral 22B on NVIDIA DGX Spark 128GB receives a F grade with 2.2 tok/s and 4K context.

What context window can Codestral 22B use on NVIDIA DGX Spark 128GB?

On NVIDIA DGX Spark 128GB, Codestral 22B can safely use up to 33K tokens of context at F16 quantization. The model's official context limit is 33K, but available memory constrains the safe maximum.

See all results for NVIDIA DGX Spark 128GB See all hardware for Codestral 22B

Embed this result▼

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

<iframe src="https://willitrunai.com/embed/codestral-22b-on-dgx-spark-128gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: