Home / Shop / IBM / Networking (IBM / Lenovo)
- Description
- Specifications
4X67A84824 is the ThinkSystem NVIDIA L4 24GB PCIe Gen4 Passive GPU, Lenovo feature code BS2C and NVIDIA part number 900-2G193-2700-001. It is an NVIDIA Ada Lovelace Tensor Core GPU with 24 GB of GDDR6 memory at 300 GB/s, ECC, and a PCIe Gen 4 x16 interface in a single-slot, low-profile card of 169 mm by 69 mm that relies on server airflow for cooling. Maximum power is 72 W and no auxiliary power cable is needed.
Peak throughput is 30.3 TFLOPS in FP32, 121 TFLOPS in FP16 and BF16 Tensor operations (242 TFLOPS with structural sparsity), 242.5 TFLOPS in FP8 and 242.5 TOPS in INT8. There are no display connectors, no NVLink and no Multi-Instance GPU support, and vGPU use is through NVIDIA vPC/vApps and RTX Virtual Workstation software. Compute APIs are CUDA 12.0, DirectCompute and OpenCL 3.0.
The option ships with the card carrying a full-height bracket, plus a low-profile bracket and documentation. Lenovo supports it in ThinkEdge SE360 V2, SE450 and SE455 V3 and in ThinkSystem SR630 V2, SR630 V3, SR650 V2, SR650 V3, SR655 V3, SR665 V3, SR670 V2, SR675 V3, SR850 V3, SR860 V3, ST650 V2, ST650 V3 and several other V3 and V4 models.
| Part number | 4X67A84824 |
| Feature code | BS2C |
| NVIDIA part number | 900-2G193-2700-001 |
| GPU architecture | NVIDIA Ada Lovelace |
| Memory | 24 GB GDDR6, 300 GB/s, ECC |
| Interface | PCIe Gen 4 x16 |
| Form factor | low profile, single slot, 169 mm x 69 mm |
| Maximum power | 72 W |
| Cooling | passive |
| FP32 | 30.3 TFLOPS |
| FP16 Tensor | 121 TFLOPS (242 with sparsity) |
| INT8 | 242.5 TOPS (485 with sparsity) |
| Display connectors | none |
| Controlled GPU status | Controlled |

