Lambda Hyperplane A100
A new standard for
deep learning hardware
See up to 40% training performance improvements with the new 4x and 8x NVIDIA® A100 Tensor Core GPU servers from Lambda.
4 NVIDIA A100 Tensor Core GPUs with NVLink™ & Mellanox InfiniBand
System Specifications
-
GPU Details
4x NVIDIA Tesla A100 SXM4 (40GB or 80GB) + NVLink
-
Processor
2x AMD EPYC™ Processors (Up to 64 cores)
-
SYSTEM RAM
512 GB
-
Storage
Up to 60TB NVMe
-
NETWORK INTERFACE
Up to 4x Mellanox InfiniBand HDR 200Gbps Cards
8 NVIDIA A100 Tensor Core GPUs with NVLink, NVSwitch™ & Mellanox InfiniBand
System Specifications
-
GPU Details
8x NVIDIA Tesla A100 SXM4 (40GB or 80GB) + NVSwitch
-
Processor
2x AMD EPYC™ or Intel Processors
-
SYSTEM RAM
1 TB
-
STORAGE
Up to 96TB NVMe
-
NETWORK INTERFACE
Up to 9x Mellanox InfiniBand HDR 200Gbps Cards
Major deep learning frameworks pre-installed
Cluster-ready deep learning infrastructure
-
Multi-node distributed training
The new Lambda Hyperplane 8-A100 Supports up to 9x Mellanox ConnectX-6 VPI HDR InfiniBand cards for up to 1.8 Terabits of internode connectivity.
-
NVIDIA multi-instance GPU (MIG) support
The A100 GPUs inside the Hyperplane can now be seamlessly divided into 7 virtual GPUs each for up to 56 virtual GPUs in a Hyperplane 8.
-
Engineered for you
Leverage Lambda support to plan your next server or cluster build and ensure it meets the needs of your specific deep learning workloads.
GPU Specifications
Comparing the A100 & V100
| NVIDIA Tesla A100 SXM4-40GB | NVIDIA Tesla V100 SXM3-32GB | |
|---|---|---|
| FP32 CUDA Cores | 6912 | 5120 |
| Clock Speed | 1410 MHz | 1530MHz |
| Theoretical FP32 TFLOPS | 19.5 TFLOPS | 15.7 TFLOPS |
| VRAM | 40 GB HBM2e | 32 GB HBM2 |
| Memory Bandwidth | 1,555 GBps | 900 GBps |
| GPU Interconnect | 12 NVLink Connections (600 GBps) | 6 NVLink Connections (300 GBps) |
| Process Node | TSMC 7nm | TSMC 12nm FFN |
| TDP (W) | 400 W | 300 W |
| Power Efficiency | 45.2 GFLOPS / W | 48.6 GFLOPS / W |