| | | | |
| Product | AMD Instinct™ MI350P - 144GB HBM3e - PCIe 5.0 x16 - Passive Cooling | NVIDIA® L40S ADA GPU Computing Accelerator - 48GB GDDR6 - PCIe 4.0 x16 - Passive Cooling | NVIDIA® H200 NVL GPU - 141GB HBM3e - PCIe 5.0 x16 - Passive Cooling - 600W | NVIDIA® RTX PRO 6000 Blackwell Server Edition - 96GB GDDR7 - PCIe 5.0 x16 - Passive Cooling (4xDP) - 600W |
| Action | Select | Select | Select | Select |
| Main Specifications |
| Product Series |
AMD Instinct MI350 Series | Nvidia L40S | Nvidia H200 | |
| Core Type |
| NVIDIA TENSOR | NVIDIA TENSOR | |
| Core Clock Speed |
2200 MHz | | | |
| Host Interface |
PCI Express 5.0 x16 | PCI Express 4.0 x16 | PCI Express 5.0 x16 | PCI Express 5.0 x16 |
| GPU Architecture |
CDNA4 | Ada Lovelace | Hopper | |
| Product Type |
| | | Server |
| Product Line |
| | | NVIDIA Professional Graphics |
| Memory Technology |
| | | GDDR7 |
| Memory Capacity |
| | | 96 GB with ECC |
| Max Displays |
| | | 4 Displays |
| Detailed Specifications |
| Streaming Processor Cores |
8,192 | 18,176 | | 24,064 CUDA Parallel Processing Cores |
| Matrix Cores |
512 | | | |
| Compute Units |
128 | | | |
| NVIDIA Tensor Cores |
| 568 | Gen 4 | | 752 (5th Gen) |
| NVIDIA RT Cores |
| 142 | Gen 3 | | 188 (4th Gen) |
| Memory Interface |
| | | 512-bit |
| Max Memory Size |
144 GB HBM3E | 48GB GDDR6 with ECC | 141 GB | |
| Max Memory Bandwidth |
4 TB/s | 864 GB/s | 4.8TB/s | |
| Peak FP64 |
| | 30 TFLOPS | |
| Peak FP64 Tensor Core |
| | 60 TFLOPS | |
| INT8 Tensor Core |
| 733 teraFLOPS | 3,341 TFLOPS | |
| TF32 Tensor Core |
| 183 teraFLOPS | 835 TFLOPS | |
| Peak BFLOAT16 Tensor Core |
| 362.05 teraFLOPS | 1,671 TFLOPS | |
| Peak FP16 Tensor Core |
| 362.05 teraFLOPS | 1,671 TFLOPS | |
| Peak FP8 Tensor Core |
| 733 teraFLOPS | 3,341 TFLOPS | |
| Peak INT4 Tensor Core |
| 733 teraFLOPS | | |
| Total NVLink Bandwidth |
| Not supported | | |
| Multi-Instance GPUs |
| No | | |
| NVENC | NVDEC |
| 3x l 3x (includes AV1 encode and decode) | | |
| Secure Boot with Root of Trust |
| Yes | | |
| NEBS Ready |
| Level 3 | | |
| Peak Microscaling Four-bit Precision Matrix (MXFP4) Performance |
4.6 PFLOPs | | | |
| Peak Microscaling Six-bit Precision Matrix (MXFP6) Performance |
4.6 PFLOPs | | | |
| Peak Half Precision (FP16) Performance |
72 TFLOPs | | | |
| Peak Half Precision Matrix (FP16) Performance |
1.15 PFLOPs | | | |
| Peak Double Precision Matrix (FP64) Performance |
36 TFLOPs | | | |
| Peak Double Precision (FP64) Performance |
36 TFLOPs | | | |
| Peak INT8 Performance |
2.3 POPs | | | |
| Peak bfloat16 |
1.15 PFLOPs | | | |
| Transistor Count |
73 Billion | | | |
| DisplayPort Connectors |
| 4x DisplayPort 1.4a | | |
| Cooling |
Passive | Passive | Passive | |
| Dual Slot |
Double Slot | | | |
| Dimensions |
10.5" (267 mm) Board Length | 4.4" (H) x 10.5" (L) | | 4.4” H x 10.5” L, FHFL Dual Slot |
| Form Factor |
PCIe® Add-in Card | | Dual-slot air-cooled | |
| Lithography |
TSMC 3nm | 6nm FinFET | | | 4N NVIDIA Custom Process |
| Supplementary Power Connectors |
| 1x 16-pin | | 1x PCIe CEM5 16-pin |
| Max Graphics Card Power (W) |
600W TBP (Max); 450W TBP configurable | 350W | Up to 600W (configurable) | 400 W – 600 W (configurable) |
| Processor |
| | | NVIDIA Blackwell Architecture |
| Memory Bandwidth |
| | | Up to 1.6 TB/s |
| Minimum Recommended Power, Single Card (W) |
| | | 1000 |
| Minimum Recommended Power, 2-Way (W) |
| | | 1600 |
| Minimum Recommended Power, 3-Way (W) |
| | | 2000 |
| Minimum Recommended Power, 4-Way (W) |
| | | 2400 |
| Thermal Solution |
| | | Passive |
| Slot Height |
| | | 2-Slot |
| Action | Select | Select | Select | Select |