Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

MI300X is an AI/HPC orientated APU that is a traditional GPU accelerator.

The MI300X AI accelerator, the world's first datacentre APU, hopes to capitalise on the booming generative AI and HPC markets.AMD has used state-of-the-art manufacturing techniques to create the MI300 series and new technologies such as ‘3.5D’ packaging to produce two chips that are claimed to offer superior performance to NVIDIA in a wide range of AI workloads. AMD has not disclosed pricing for its new chips, but they are currently shipping to a wide range of OEM partners.

The Instinct MI300 is a game-changing design: the datacentre APU mixes a total of 13 small chips, many of which are 3D stacked, to create a chip with 24 Zen 4 CPU cores incorporating the CDNA 3 graphics engine and 8 stacks of HBM3. Overall, the chip has 153 billion transistors, making it the largest chip AMD has built to date. AMD claims the chip outperforms NVIDIA's H100 GPU by a factor of four in some workloads and claims twice the performance per watt.

The MI300X represents the pinnacle of AMD's small-chip-based design approach, combining eight 12Hi stacks of HBM3 memory with eight 3D stacks of 5nm CDNA 3 GPU small chips (known as XCDs) fused onto four underlying 6nm I/O chips that are connected using AMD's now-proven Hybrid Bonding technology.

The result is a 750W accelerator with 304 compute units, 192GB of HBM3 capacity and 5.3TB/s bandwidth. The accelerator also features 256MB of Infinity Cache, which is used as a shared L3 cache layer to facilitate communication between the smaller chips. AMD calls its packaging technology for tying the smaller chips together ‘3.5D,’ which means that the 3D stacked GPUs and I/O chips are fused together via hybrid bonding and are connected to the standard 2.5D chips in the rest of the module. The rest of the module is packaged in a standard 2.5D package (horizontally attached). We'll take a closer look at the architectural components below.

The MI300X accelerator is designed to work in groups of eight in AMD's generative AI platform, with throughput of 896 GB/s between GPUs interconnected via the Infinity Fabric. the system has 1.5TB of total HBM3 memory, delivering up to 10.4 Petaflops of performance (BF16/FP16). The system is built on the Open Compute Project (OCP) Universal Baseboard (UBB) design standard, simplifying adoption, especially for hyperscale enterprises.

• 1.Is the inventory and price seen on the platform accurate?

  A: all online goods on the shelves can be ordered online, but considering the large liquidity of spot inventory, it is not possible to achieve 100% accuracy at present. In case of any abnormality, you can contact our company online and give the corresponding solution.

• 2.Are all our products genuine?

  A: our self-supporting products are from cooperative domestic and foreign original factories or authorized agents, and the source can be traced to ensure the original and authentic products.

• 3.Can you provide the qualification certificate of the original factory or agent?

  A: at present, our self-employed brands TI，ST，ADI，NXP，LATTICE，CYPRESS，INFINEON，XILINX and other first-line brands. Other channels are the original factory and agency channels. If necessary, we can communicate and confirm according to the specific brand and model.

• 4.Can I make an inquiry through the company's website?

  A: You can inquire through the website, or by phone and email.

• 5.When can I deliver the goods after placing an order? How long will it take to get there?

  A: the delivery date is marked in most commodity information. You can estimate the delivery time of the commodity according to the delivery date. The specific arrival time depends on the warehouse where the commodity is located and the logistics mode you choose.

• 6.Can the company issue an invoice?

  A: you can issue ordinary invoices for individual users or special VAT invoices for enterprise users.