سرور سوپرمیکرو Supermicro SuperServer 2029GP-TR
سرور سوپرمیکرو Supermicro SuperServer 4029GP-TRT
4029GP-TRT Dual Socket (LGA 3647) support 2nd Gen Intel Xeon Scalable processors 24 DIMMs; up to 6TB 3DS ECC DDR4-2933MHz RDIMM/LRDIMM, Supports Intel Optane DCPMM 8 PCI-E 3.0 x16 slots (support up to 8 double width GPU), 2 PCI-E 3.0 x8, 1 PCI-E 3.0 x4 Up to 24 Hot-swap 2.5" drive bays; 2x 10GBase-T LAN ports via Intel C622 2000W (2+2) Redundant Power Supplies Titanium Level
سرور هوش مصنوعی Supermicro SYS-522GA-NRT
| تجهیز | مشخصات | تعداد |
| Server | Supermicro SYS-522GA-NRT | 1 |
| CPU | Intel Xeon 6960P – 2.7GHz – 72 Core – 432MB Cache | 2 |
| Memory | HPE 128GB DDR5-6400 RDIMM 2Rx4 – P69730-B21 | 8 |
| GPU | NVIDIA H200 NVL 141GB | 2 |
| GPU Interconnect | NVIDIA NVLink 2-Way | 1 |
| NVMe SSD | HPE 7.68TB NVMe Gen4 U.3 RI – P64848-B21 | 2 |
| SATA SSD | HPE 960GB SATA 6G MU SFF – P40503-B21 | 2 |
کارت گرافیک NVIDIA Quadro RTX 6000 24GB
کارت گرافیک A100 Tensor Core 80GB
A100 Tensor Core
مشخصات فنی:
FP64 Tensor Core: 19.5 ترافلاپس
Tensor Float 32 (TF32): 156 ترافلاپس (312 ترافلاپس با تنک‌سازی)
FP16 Tensor Core: 312 ترافلاپس (624 ترافلاپس با تنک‌سازی)
INT8 Tensor Core: 624 ترااپس (1,248 ترااپس با تنک‌سازی)
حافظه GPU: 40 گیگابایت HBM2 یا 80 گیگابایت HBM2e
پهنای باند: تا 2,039 گیگابایت بر ثانیه
توان طراحی حرارتی: 250 وات (PCIe) تا 400 وات (SXM)
فرم‌فاکتورها: PCIe و SXM4
NVLink: تا 600 گیگابایت بر ثانیه
PCIe Gen4: 64 گیگابایت بر ثانیه
پشتیبانی از سیستم‌های NVIDIA HGX A100 با تا 16 GPU.
کارت گرافیک NVIDIA H100 Tensor Core GPU 80GB
NVIDIA H100 Tensor Core GPU
Architecture Overview
GPU Architecture: Hopper (H100)
Process Node: TSMC 4N (Custom 4nm)
Transistor Count: 80 billion
SMs (Streaming Multiprocessors): Up to 144 (varies by configuration)
CUDA Cores: Not explicitly stated (Hopper focuses on Tensor Cores rather than CUDA cores)
Tensor Cores: 4th Generation (with new FP8, FP16, TF32, FP64 acceleration)
PCIe Version: PCIe 5.0 (x16)
NVLink: 4th Gen NVLink (900 GB/s bidirectional bandwidth)
Multi-Instance GPU (MIG): Up to 7 instances per GPU
کارت گرافیک NVIDIA H200 Tensor Core 141GB
NVIDIA H200 Tensor Core
Core Architecture
Process Node: TSMC 4N (Custom 4nm)
Transistor Count: ~80 billion
Streaming Multiprocessors (SMs): Up to 144 (similar to H100, but with memory enhancements)
CUDA Cores: Not explicitly disclosed (Hopper focuses on Tensor Core efficiency)
Tensor Cores: 4th Gen with new FP8 acceleration
Thread Block Clustering: Allows better workload distribution across SMs
