⚡
Cluster
Mind
AI Infra Calculator
Calculator Mode
🔮 Inference
🏋️ Training
Configuration
Model Size
(billion parameters)
1B
3B
7B
13B
30B
70B
180B
405B
1T (1000B)
2T (2000B)
Quantization
FP32
BF16
FP16
INT8
INT4
Target Throughput
(req/sec)
1 req/s
5 req/s
10 req/s
50 req/s
100 req/s
500 req/s
1,000 req/s
5,000 req/s
10,000 req/s
Avg Output Tokens / Request
50 tokens
100 tokens
250 tokens
500 tokens
1,000 tokens
Context Length
2K tokens
4K tokens
8K tokens
16K tokens
32K tokens
128K tokens
Deployment Target
☁️ Cloud
🏢 On-Prem
📋 Copy Results
Training Configuration
Model Size
(billion parameters)
1B
3B
7B
13B
30B
70B
180B
405B
1T (1000B)
2T (2000B)
5T (5000B)
10T (10000B)
Dataset Size
(training tokens, billions)
1B tokens
10B tokens
100B tokens
300B tokens
1T tokens
2T tokens
5T tokens
10T tokens
20T tokens
Optimizer
(affects VRAM usage)
Adam (default) — 2× FP32 states
AdamW — 2× FP32 states
Adafactor — ~0.25× states
8-bit Adam (bitsandbytes) — 0.5× states
SGD + Momentum — 1× states
Lion — 1× states
SOAP — 2× FP32 states
Global Batch Size
(tokens)
256K tokens
512K tokens
1M tokens
2M tokens
4M tokens
Precision
BF16
FP32
FP8 Mixed
Gradient Checkpointing
Reduces activation memory (slower)
ZeRO Stage
None
ZeRO-1
ZeRO-2
ZeRO-3
Parallelism Strategy
Data Parallel only
Tensor + Data Parallel
Full 3D (TP+PP+DP)
Training Deployment Target
☁️ Cloud
🏢 On-Prem
📋 Copy Results