Servers optimized for GPU Compute provide huge boost in performance compared to CPU only servers. With NVIDIAs latest generation of GPUs – Ampere, high performance computing is more cost effective than ever before. Several new GPUs have been introduced in 2021, enabling effective GPU acceleration for a wide variety of workloads:
NVIDIA A100. High performance and double precision computing. Artificial intelligence and machine learning. Big data processing. GPU virtualization with MIG support.
NVIDIA A40. Virtual Desktop Infrastructure for engineers and high-end visual computing. Memory demanding render and single precision workloads.
NVIDIA A30. Double precision computing. All features with minimal power draw. GPU virtualization with MIG support.
NVIDIA A10. Virtual Desktop Infrastructure for mainstream graphics users. Cost effective GPU compute and AI workloads.
Specifications in brief:
| Product | A100 | A40 | A30 | A10 |
| Performance: double precision tensor core – TF64 | 19,5 TFLOPS | - | 10,3 TFLOPS | - |
| Performance: single precision tensor core – TF32 | 156 TFLOPS | 74,8 TFLOPS | 82 TFLOPS | 62,5 TFLOPS |
| Memory size | 40 or 80 GB HBM2 | 48 GB GDDR6 | 24 GB HBM2 | 24 GB GDDR6 |
| Memory bandwidth | 1555+ GB/s | 696 GB/s | 933 GB/s | 600 GB/s |
| Multi-instance GPU (MIG) | Up to 7 | - | Up to 4 | - |
| Power draw | Up to 300 W | 300 W | 165 W | 150 W |
| Monitor support | 0 | 0 | 0 | 0 |
| PCI-e form factor | Double width Full length Passive cooler | Double width Full length Passive cooler | Double width Full length Passive cooler | Single width Full length Passive cooler |
