Skip to content
Updated minutes ago
NVIDIA

Nemotron 340B

NVIDIA · dense · 340B parameters · 131,072 context

Quality
85.0

Parameters

340B

Context Window

128K tokens

Architecture

Dense

Best GPU

B200 NVL (pair)

Cheapest API

$4.20/M

Quality Score

85/100

Intelligence Brief

Nemotron 340B is a 340B parameter DENSE model from NVIDIA, featuring Grouped Query Attention (GQA) with 96 layers and 18,432 hidden dimensions. With a 131,072 token context window, it supports tools, structured output, code, math, multilingual. On standardized benchmarks, it achieves MMLU 82, HumanEval 57, GSM8K 92. The most cost-effective API deployment is via nvidia at $4.20/M output tokens. For self-hosted inference, B200 NVL (pair) delivers optimal throughput at $19929/month.

Architecture Details

TypeDENSE
Total Parameters340B
Active Parameters340B
Layers96
Hidden Dimension18,432
Attention Heads96
KV Heads8
Head Dimension192
Vocab Size256,000

Memory Requirements

BF16 Weights

680.0 GB

FP8 Weights

340.0 GB

INT4 Weights

170.0 GB

KV-Cache per Token2359296 bytes
Activation Estimate8.00 GB

Fits on (multi-GPU with Tensor Parallelism)

Multi-GPU configurations use Tensor Parallelism (TP) to split model layers across GPUs. Requires NVLink or NVSwitch interconnect for optimal performance.

GPU Compatibility Matrix

Nemotron 340B is compatible with 7% of GPU configurations across 41 GPUs at 3 precision levels.

BF16 (Full)
FP8 (Half)
INT4 (Quarter)
Blackwell(7 GPUs)
B200 NVL (pair)360GB
B300288GB
B100 SXM192GB
GB200 NVL72 (per GPU)192GB
Hopper(7 GPUs)
H100 NVL 94GB (per GPU pair)188GB
H200 SXM141GB
H2096GB
GH20096GB
Ada Lovelace(11 GPUs)
L40S48GB
L4048GB
RTX 6000 Ada48GB
L2048GB
Ampere(16 GPUs)
A100 80GB SXM80GB
A100 80GB PCIe80GB
A1664GB
RTX A600048GB
Legend:No fitVery tightTightModerateGoodExcellent

GPU Recommendations

B200 NVL (pair)optimal

FP8 · 2 GPUs · tensorrt-llm

88/100

score

Throughput

280.0 tok/s

Latency (ITL)

3.6ms

Est. TTFT

1ms

Cost/Month

$19929

Cost/M Tokens

$27.08

Use this config →
B200 SXMoptimal

FP8 · 4 GPUs · tensorrt-llm

83/100

score

Throughput

280.0 tok/s

Latency (ITL)

3.6ms

Est. TTFT

1ms

Cost/Month

$17044

Cost/M Tokens

$23.16

Use this config →
H200 SXMoptimal

FP8 · 4 GPUs · tensorrt-llm

80/100

score

Throughput

280.0 tok/s

Latency (ITL)

3.6ms

Est. TTFT

1ms

Cost/Month

$10211

Cost/M Tokens

$13.88

Use this config →

Deployment Options

API

API Deployment

nvidia

$4.20/M

output tokens

Self-Hosted

Single GPU

Requires multi-GPU setup (340 GB VRAM needed)

Scale

Multi-GPU

B200 NVL (pair) x2

280.0 tok/s

TP· $19929/mo

API Pricing Comparison

ProviderInput $/MOutput $/MBadges
nvidia$4.20$4.20
Cheapest

Cost Analysis

ProviderInput $/MOutput $/M~Monthly Cost
nvidiaBest Value$4.20$4.20$42

Cost per 1,000 Requests

Short (500 tok)

$2.94

via nvidia

Medium (2K tok)

$11.76

via nvidia

Long (8K tok)

$42.00

via nvidia

Performance Estimates

Throughput by GPU

B200 NVL (pair)
280.0 tok/s
B200 SXM
280.0 tok/s
H200 SXM
280.0 tok/s

VRAM Breakdown (B200 NVL (pair), FP8)

Weights
Act
Weights 170.0 GBKV-Cache 4.8 GBActivations 64.0 GBOverhead 8.5 GB

Precision Impact

bf16

340.0 GB

weights/GPU

fp8

170.0 GB

weights/GPU

~280.0 tok/s

int4

85.0 GB

weights/GPU

Quality Benchmarks

Top 10%
94th percentile across all models
MMLU
82.0
Average (61th pctile)
HumanEval
57.0
Average (66th pctile)
GSM8K
92.0
Above Average (76th pctile)
MT-Bench
85.0
Bottom 25% (0th pctile)

Capabilities

Features

Tool Use Vision Code Math Reasoning Multilingual Structured Output

Supported Frameworks

tensorrt-llmvllmsglang

Supported Precisions

BF16 (default)FP8INT4

Where to Deploy Nemotron 340B

Similar Models

Similar specsCompare →
Cheaper, Smaller modelCompare →

Grok-2

314B params · moe

Quality: 78

from $10.00/M

Lower quality, More expensiveCompare →

Grok-3

314B params · dense

Quality: 91

from $15.00/M

Higher quality, More expensiveCompare →
Smaller context, Lower qualityCompare →

Frequently Asked Questions

How much VRAM does Nemotron 340B need for inference?

Nemotron 340B requires approximately 680.0 GB of VRAM at BF16 precision, 340.0 GB at FP8, or 170.0 GB at INT4 quantization. Additional VRAM is needed for KV-cache (2359296 bytes per token) and activations (~8.00 GB).

What is the best GPU for Nemotron 340B?

The top recommended GPU for Nemotron 340B is the B200 NVL (pair) (x2) using FP8 precision. It achieves approximately 280.0 tokens/sec at an estimated cost of $19929/month ($27.08/M tokens). Score: 88/100.

How much does Nemotron 340B inference cost?

Nemotron 340B API inference starts from $4.20/M input tokens and $4.20/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.