
ASUS XA NB3I-E12
9U NVIDIA HGX B300 8-GPU server with dual Intel Xeon 6 for heavy AI training and inference.
- GPUs: NVIDIA HGX B300 8-GPU
- CPUs: Dual Intel Xeon 6
- Networking: 8× embedded ConnectX-8 InfiniBand, dual 10 GbE
Model ESC A8A-E12U
7U eight-GPU AMD Instinct MI325X server for large AI models and HPC.
The ASUS ESC A8A-E12U is a 7U dual-socket server with eight AMD Instinct MI325X accelerators and AMD EPYC 9005 processors, built for high-end AI training and large-model inference.
Each MI325X carries 256 GB of HBM3E, and a one-GPU-to-one-NIC topology with AMD Infinity Fabric provides the bandwidth needed for distributed training. It runs the open AMD ROCm software stack.
256 GB HBM3E per accelerator for trillion-parameter models.
Infinity Fabric mesh with a dedicated NIC per GPU.
Room for high-speed networking and storage.
Separate CPU and GPU air tunnels with dual-rotor fans.
Six 3000 W 80 PLUS Titanium supplies.
Open software supporting PyTorch, TensorFlow and JAX.
| CPUs | 2× AMD EPYC 9005 / 9004, up to 400 W |
|---|---|
| Memory | 24× DDR5 DIMMs up to 6400 MHz (1DPC), up to 3 TB |
| GPUs | 8× AMD Instinct MI325X (256 GB HBM3E each) |
|---|---|
| Expansion | Up to 11 PCIe slots: 8× Gen5 x16 (GPU group), 1× Gen5 x16 + 1× Gen5 x8 (CPU1), 1× Gen5 x16 (CPU2) |
| LAN | 2× 10 GbE (Intel X710-AT2), 1× management port |
|---|---|
| Front I/O | 4× USB 3.2 Gen1, 1× VGA |
| Form factor | 7U |
|---|---|
| Dimensions | 885 × 447 × 306.65 mm |
| Weight | 108 kg net / 142.4 kg gross |
| Power | 5+1 redundant 3000 W 80 PLUS Titanium |
| Management | ASUS Control Center, ASMB11-iKVM |
| Operating temperature | 10 °C to 35 °C |
Tell us your workload and timeline for the ASUS ESC A8A-E12U. We reply within one business day with a configuration, pricing and lead time.