VRAM Reference

Mistral Large VRAM Requirements: FP16 / INT4 / KV-Cache (2026)

Mistral Large runs on 123B parameters with a 128K-token context window. Weights are deterministic param math; KV-cache uses the registry layer geometry where published, otherwise a modeled GQA estimate — every figure is provenance-labeled below.

FP16 weights74 GB

INT4 weights37 GB

Context128K tokens

Single-GPU verdictFits (INT4, 73 GB)

74 GB FP16 weights

Calculated EstimateHIGH

SourceOpenGPU Radar models-registry.json (param-count weight model)

VerifiedSep 29, 2026

Value

74 GB

Methodology

weights = parametersB × bytes-per-parameter (FP16 = 2 B, INT4 = 0.5 B)

Refreshed daily from provider APIs and market scrapingSep 29, 2026

37 GB INT4 weights

Calculated EstimateHIGH

SourceOpenGPU Radar models-registry.json (param-count weight model)

VerifiedSep 29, 2026

Value

37 GB

Methodology

weights = parametersB × bytes-per-parameter (FP16 = 2 B, INT4 = 0.5 B)

Refreshed daily from provider APIs and market scrapingSep 29, 2026

128K context window

Manufacturer SpecHIGH

SourceModel vendor context-window specification

VerifiedSep 29, 2026

Value

128,000 tokens

Refreshed daily from provider APIs and market scrapingSep 29, 2026

Methodology →

How much VRAM does Mistral Large need?

Weights: 74 GB at FP16, 37 GB at INT4 (parameters × bytes-per-parameter: FP16 = 2 B, INT4 = 0.5 B).

Full stack at 128K: 280.3 GB FP16 / 73.4 GB INT4 — including KV-cache, CUDA overhead, activations, and fragmentation headroom from the canonical VRAM engine.

Single-GPU verdict: INT4 fits 80 GB datacenter cards (A100/H100); FP16 needs 320 GB → multi-GPU.

VRAM by Precision & Context

From the canonical VRAM engine: weights + KV-cache + CUDA overhead + activations + fragmentation headroom. KV-cache uses published layer geometry where available; otherwise a modeled GQA estimate (labeled in the badge).

Precision	Context	Weights	KV-Cache	Total (est.)
FP16	4K	246.0 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 246 GB Methodology Formula: parameterCountB × bytesPerParam. Precision: FP16 = 2 bytes/param. (parameterCountB in billions, result in GB). Assumptions & Parameters paramCountB: 123 quantization: fp16 bytesPerParam: 2 Refreshed daily from provider APIs and market scrapingSep 26, 2026	0.23 GB Calculated EstimateMEDIUM SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 0.23 GB Methodology Estimated via GQA model — architectural parameters not available. Assumptions & Parameters contextLength: 4096 quantization: fp16 estimatedVia: Modeled GQA estimate Refreshed daily from provider APIs and market scrapingSep 26, 2026	272.6 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 272.62 GB Methodology Total = weights + kvCache + cudaOverhead + activation + fragmentationHeadroom (10%). Assumptions & Parameters paramCountB: 123 quantization: fp16 contextLength: 4096 batchSize: 1 numLayers: 0 numKvHeads: 0 headDim: 0 gqaRatio: 1 architecture: mistral-large bytesPerParam: 2 bytesPerKvElement: 2 cudaOverheadGb: 1.2 activationGb: 0.4 fragmentationHeadroomPct: 10 isMla: false Refreshed daily from provider APIs and market scrapingSep 26, 2026
FP16	33K	246.0 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 246 GB Methodology Formula: parameterCountB × bytesPerParam. Precision: FP16 = 2 bytes/param. (parameterCountB in billions, result in GB). Assumptions & Parameters paramCountB: 123 quantization: fp16 bytesPerParam: 2 Refreshed daily from provider APIs and market scrapingSep 26, 2026	1.85 GB Calculated EstimateMEDIUM SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 1.85 GB Methodology Estimated via GQA model — architectural parameters not available. Assumptions & Parameters contextLength: 32768 quantization: fp16 estimatedVia: Modeled GQA estimate Refreshed daily from provider APIs and market scrapingSep 26, 2026	274.4 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 274.4 GB Methodology Total = weights + kvCache + cudaOverhead + activation + fragmentationHeadroom (10%). Assumptions & Parameters paramCountB: 123 quantization: fp16 contextLength: 32768 batchSize: 1 numLayers: 0 numKvHeads: 0 headDim: 0 gqaRatio: 1 architecture: mistral-large bytesPerParam: 2 bytesPerKvElement: 2 cudaOverheadGb: 1.2 activationGb: 0.4 fragmentationHeadroomPct: 10 isMla: false Refreshed daily from provider APIs and market scrapingSep 26, 2026
FP16	128K	246.0 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 246 GB Methodology Formula: parameterCountB × bytesPerParam. Precision: FP16 = 2 bytes/param. (parameterCountB in billions, result in GB). Assumptions & Parameters paramCountB: 123 quantization: fp16 bytesPerParam: 2 Refreshed daily from provider APIs and market scrapingSep 26, 2026	7.24 GB Calculated EstimateMEDIUM SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 7.24 GB Methodology Estimated via GQA model — architectural parameters not available. Assumptions & Parameters contextLength: 128000 quantization: fp16 estimatedVia: Modeled GQA estimate Refreshed daily from provider APIs and market scrapingSep 26, 2026	280.3 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 280.33 GB Methodology Total = weights + kvCache + cudaOverhead + activation + fragmentationHeadroom (10%). Assumptions & Parameters paramCountB: 123 quantization: fp16 contextLength: 128000 batchSize: 1 numLayers: 0 numKvHeads: 0 headDim: 0 gqaRatio: 1 architecture: mistral-large bytesPerParam: 2 bytesPerKvElement: 2 cudaOverheadGb: 1.2 activationGb: 0.4 fragmentationHeadroomPct: 10 isMla: false Refreshed daily from provider APIs and market scrapingSep 26, 2026
INT4	4K	61.5 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 61.5 GB Methodology Formula: parameterCountB × bytesPerParam. Precision: INT4 = 0.5 bytes/param. (parameterCountB in billions, result in GB). Assumptions & Parameters paramCountB: 123 quantization: int4 bytesPerParam: 0.5 Refreshed daily from provider APIs and market scrapingSep 26, 2026	0.12 GB Calculated EstimateMEDIUM SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 0.12 GB Methodology Estimated via GQA model — architectural parameters not available. Assumptions & Parameters contextLength: 4096 quantization: int4 estimatedVia: Modeled GQA estimate Refreshed daily from provider APIs and market scrapingSep 26, 2026	69.5 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 69.54 GB Methodology Total = weights + kvCache + cudaOverhead + activation + fragmentationHeadroom (10%). Assumptions & Parameters paramCountB: 123 quantization: int4 contextLength: 4096 batchSize: 1 numLayers: 0 numKvHeads: 0 headDim: 0 gqaRatio: 1 architecture: mistral-large bytesPerParam: 0.5 bytesPerKvElement: 1 cudaOverheadGb: 1.2 activationGb: 0.4 fragmentationHeadroomPct: 10 isMla: false Refreshed daily from provider APIs and market scrapingSep 26, 2026
INT4	33K	61.5 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 61.5 GB Methodology Formula: parameterCountB × bytesPerParam. Precision: INT4 = 0.5 bytes/param. (parameterCountB in billions, result in GB). Assumptions & Parameters paramCountB: 123 quantization: int4 bytesPerParam: 0.5 Refreshed daily from provider APIs and market scrapingSep 26, 2026	0.93 GB Calculated EstimateMEDIUM SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 0.93 GB Methodology Estimated via GQA model — architectural parameters not available. Assumptions & Parameters contextLength: 32768 quantization: int4 estimatedVia: Modeled GQA estimate Refreshed daily from provider APIs and market scrapingSep 26, 2026	70.4 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 70.43 GB Methodology Total = weights + kvCache + cudaOverhead + activation + fragmentationHeadroom (10%). Assumptions & Parameters paramCountB: 123 quantization: int4 contextLength: 32768 batchSize: 1 numLayers: 0 numKvHeads: 0 headDim: 0 gqaRatio: 1 architecture: mistral-large bytesPerParam: 0.5 bytesPerKvElement: 1 cudaOverheadGb: 1.2 activationGb: 0.4 fragmentationHeadroomPct: 10 isMla: false Refreshed daily from provider APIs and market scrapingSep 26, 2026
INT4	128K	61.5 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 61.5 GB Methodology Formula: parameterCountB × bytesPerParam. Precision: INT4 = 0.5 bytes/param. (parameterCountB in billions, result in GB). Assumptions & Parameters paramCountB: 123 quantization: int4 bytesPerParam: 0.5 Refreshed daily from provider APIs and market scrapingSep 26, 2026	3.62 GB Calculated EstimateMEDIUM SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 3.62 GB Methodology Estimated via GQA model — architectural parameters not available. Assumptions & Parameters contextLength: 128000 quantization: int4 estimatedVia: Modeled GQA estimate Refreshed daily from provider APIs and market scrapingSep 26, 2026	73.4 GB Calculated EstimateHIGH SourceDeterministic VRAM Canonical Engine VerifiedSep 26, 2026 Value 73.39 GB Methodology Total = weights + kvCache + cudaOverhead + activation + fragmentationHeadroom (10%). Assumptions & Parameters paramCountB: 123 quantization: int4 contextLength: 128000 batchSize: 1 numLayers: 0 numKvHeads: 0 headDim: 0 gqaRatio: 1 architecture: mistral-large bytesPerParam: 0.5 bytesPerKvElement: 1 cudaOverheadGb: 1.2 activationGb: 0.4 fragmentationHeadroomPct: 10 isMla: false Refreshed daily from provider APIs and market scrapingSep 26, 2026

Cloud GPUs That Fit Mistral Large

Fit = full-stack total (weights + KV at 128K) ≤ GPU VRAM. Rates are the lowest observed on-demand rows in data/providers.json, refreshed daily.

GPU	VRAM	FP16 fit	INT4 fit	Lowest on-demand
H200 SXM5	141 GB	OOM	✓ fits	$2.79/hr (Vast.ai)
B200 Blackwell	192 GB	OOM	✓ fits	$3.99/hr (Vast.ai)
H100 SXM5	80 GB	OOM	✓ fits	$1.89/hr (Vast.ai)
A100 80GB SXM4	80 GB	OOM	✓ fits	$1.59/hr (Lambda Labs)
L40S	48 GB	OOM	OOM	$0.69/hr (Vast.ai)
RTX 4090	24 GB	OOM	OOM	$0.34/hr (Vast.ai)

Recommended GPUs for Mistral Large

Minimum = smallest single GPU that fits the full INT4 stack (73.4 GB). Optimal = registry recommendation, falling back to the lowest observed on-demand rate among fitting cards. Every card links to its canonical GPU page.

Minimum80 GB · $1.89/hr

H100 SXM5

Smallest fitting card — 80 GB vs 73.4 GB INT4 full stack.

View GPU page →

Optimal141 GB · $2.79/hr

H200 SXM5

Registry recommendation (1x H200).

View GPU page →

Next steps

⚡

Open in VRAM calculatorPre-filled with Mistral Large

✓

Full model profileBenchmarks, providers & run commands

GPU

Frequently Asked Questions

How much VRAM does Mistral Large need?▾

Mistral Large needs 74 GB at FP16 (weights) and 37 GB at INT4 from the param-count weight model. Full-stack totals at 128K context are 280.3 GB (FP16) and 73.4 GB (INT4) including KV-cache, CUDA overhead, and fragmentation headroom.

Can Mistral Large run on a single GPU?▾

Yes — the registry recommendation is 1x H200, and the NVIDIA H200 SXM5 (141GB HBM3e) fits the INT4 workload single-GPU. FP16 full-stack needs 280.3 GB.

What context length fits Mistral Large?▾

Registry context window is 128,000 tokens. KV-cache grows linearly with context — the quantization table shows totals at 4K, 33K, 128K context for FP16 and INT4.

Weight model: parameters × bytes-per-parameter. KV-cache: GQA formula or modeled estimate where layer geometry is unpublished.Methodology →