⏱️ Training Time Estimator
Estimate duration & cost · Compare single / multi-GPU scaling
⚙️ Configuration
🔥 Model presets
ResNet-50
BERT-base
GPT-2 (124M)
LLaMA-7B
Stable Diffusion
📦 Dataset size (samples)
🧠 Model parameters (millions)
🎮 GPU type
H100 SXM (312 TFLOPS FP16) — $32.5/hr
A100 80GB (197 TFLOPS) — $22.8/hr
A100 40GB (156 TFLOPS) — $18.2/hr
A6000 (83 TFLOPS) — $13.5/hr
A5000 (71 TFLOPS) — $10.2/hr
RTX 4090 (65 TFLOPS) — $8.9/hr
RTX 4080 (51 TFLOPS) — $6.5/hr
RTX 4070 Ti (40 TFLOPS) — $5.2/hr
RTX 3090 (35 TFLOPS) — $4.1/hr
RTX 3080 (28 TFLOPS) — $3.2/hr
RTX 3070 (21 TFLOPS) — $2.5/hr
RTX 3060 (16 TFLOPS) — $1.9/hr
T4 (12 TFLOPS) — $1.2/hr
V100 (8 TFLOPS) — $0.9/hr
P100 (6 TFLOPS) — $0.6/hr
K80 (4.5 TFLOPS) — $0.45/hr
📏 Batch size
🔁 Epochs
🎯 Precision
FP32 (1x)
FP16 (2x throughput)
BF16 (2x throughput)
INT8 (4x throughput)
🚀 Calculate
📊 Results & multi‑GPU comparison
⏱️ Time per step
—
📐 Steps per epoch
—
🕒 Total training time
—
💰 Estimated cost (single GPU)
—
🔄 Multi‑GPU scaling (estimated)
1 GPU
—
—
2 GPUs
—
—
4 GPUs
—
—
8 GPUs
—
—
⚡ Scaling efficiency:
based on workload
* TFLOPS based on advertised mixed‑precision. Cloud rates approximate (on‑demand). Scaling efficiency assumes 0.85/0.75/0.65 for 2/4/8 GPUs.