Llama 2 70B
Meta · dense · 70B parameters · 4,096 context
Parameters
70B
Context Window
4K tokens
Architecture
Dense
Best GPU
H200 SXM
Cheapest API
$0.90/M
Quality Score
62/100
Intelligence Brief
Llama 2 70B is a 70B parameter DENSE model from Meta, featuring Multi-Head Attention (MHA) with 80 layers and 8,192 hidden dimensions. With a 4,096 token context window, it supports code. On standardized benchmarks, it achieves MMLU 69.8, HumanEval 30, GSM8K 56. The most cost-effective API deployment is via together at $0.90/M output tokens. For self-hosted inference, H200 SXM delivers optimal throughput at $2553/month.
Architecture Details
Memory Requirements
BF16 Weights
140.0 GB
FP8 Weights
70.0 GB
INT4 Weights
35.0 GB
GPU Compatibility Matrix
Llama 2 70B is compatible with 38% of GPU configurations across 41 GPUs at 3 precision levels.
GPU Recommendations
FP8 · 1 GPU · tensorrt-llm
100/100
score
Throughput
560.0 tok/s
Latency (ITL)
1.8ms
Est. TTFT
0ms
Cost/Month
$2553
Cost/M Tokens
$1.73
FP8 · 1 GPU · tensorrt-llm
100/100
score
Throughput
478.0 tok/s
Latency (ITL)
2.1ms
Est. TTFT
0ms
Cost/Month
$940
Cost/M Tokens
$0.75
FP8 · 1 GPU · tensorrt-llm
100/100
score
Throughput
478.0 tok/s
Latency (ITL)
2.1ms
Est. TTFT
0ms
Cost/Month
$2838
Cost/M Tokens
$2.26
Deployment Options
API Deployment
together
$0.90/M
output tokens
Single GPU
H200 SXM
$2553/mo
Min VRAM: 70 GB
Multi-GPU
H100 SXM x2
560.0 tok/s
TP· $3587/mo
API Pricing Comparison
| Provider | Input $/M | Output $/M | Badges |
|---|---|---|---|
| together | $0.90 | $0.90 | Cheapest |
Cost Analysis
| Provider | Input $/M | Output $/M | ~Monthly Cost |
|---|---|---|---|
| togetherBest Value | $0.90 | $0.90 | $9 |
Cost per 1,000 Requests
Short (500 tok)
$0.63
via together
Medium (2K tok)
$2.52
via together
Long (8K tok)
$9.00
via together
Performance Estimates
Throughput by GPU
VRAM Breakdown (H200 SXM, FP8)
Precision Impact
bf16
140.0 GB
weights/GPU
fp8
70.0 GB
weights/GPU
~560.0 tok/s
int4
35.0 GB
weights/GPU
Quality Benchmarks
Capabilities
Features
Supported Frameworks
Supported Precisions
Where to Deploy Llama 2 70B
Self-Hosted Infrastructure
Similar Models
Code Llama 70B
70B params · dense
Quality: 60
from $0.90/M
WizardMath 70B
70B params · dense
Quality: 50
Claude Sonnet 4
70B params · dense
Quality: 86
from $15.00/M
o1-mini
70B params · dense
Quality: 83
from $12.00/M
o3-mini
70B params · dense
Quality: 86
from $4.40/M
Frequently Asked Questions
How much VRAM does Llama 2 70B need for inference?
Llama 2 70B requires approximately 140.0 GB of VRAM at BF16 precision, 70.0 GB at FP8, or 35.0 GB at INT4 quantization. Additional VRAM is needed for KV-cache (2621440 bytes per token) and activations (~2.50 GB).
What is the best GPU for Llama 2 70B?
The top recommended GPU for Llama 2 70B is the H200 SXM using FP8 precision. It achieves approximately 560.0 tokens/sec at an estimated cost of $2553/month ($1.73/M tokens). Score: 100/100.
How much does Llama 2 70B inference cost?
Llama 2 70B API inference starts from $0.90/M input tokens and $0.90/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.