Many of you might know GIGABYTE from PC gaming components, but they also make high-performance computing (HPC) servers and other servers for industrial use. In the latest news, the brand has announced its innovation in high-performance computing (HPC) and artificial intelligence (AI) through its subsidiary Giga Computing.
Welcome the new GIGABYTE G383-R80 and two GIGABYTE G593 series servers. Each of them is a powerhouse in its own right. These new servers are designed to harness the unparalleled capabilities of AMD Instinct MI300 Series accelerators.
Let’s dive into the technological specifications and innovation of these cutting-edge servers that are set to propel the world of HPC and AI into new dimensions.
G383-R80 Series
At the forefront of this groundbreaking release is the GIGABYTE G383-R80. This is a technical marvel in itself. The server series seamlessly marries CPU and GPU technologies to deliver exceptional advancements in HPC.
This server hosts four LGA6096 sockets designed specifically for the AMD Instinct MI300A APU. This is a mighty chip that intricately integrates a CPU boasting twenty-four AMD Zen4 cores with a potent GPU constructed with AMD CDNA 3 GPU cores.
The unique chiplet design facilitates resource sharing. One notable thing is the 128GB of unified HBM3 memory. So, it can handle AI tasks, especially the large AI models that are gaining a lot of popularity these days.
The G383 server is engineered with expansion in mind. So GIGABYTE has added twelve PCIe Gen5 slots for networking, storage, or additional accelerators. The front of the chassis, which offers eight 2.5-inch Gen5 NVMe slots, is a standout feature.
With its expertise in managing workloads that are sensitive to latency, this configuration is specifically designed to handle demanding tasks like real-time big data analytics, meeting the growing demands of the banking and telecom sectors.
G593 Series
Then, we have the GIGABYTE G593 series. The GIGABYTE G593-ZX1 and G593-ZX2 servers mark a significant leap in the world of exascale computing. These servers use the AMD CDNA accelerator architecture for top performance. These servers, which include the most recent CDNA 3 architecture, are aimed to provide significant performance improvements over the MI250X GPU, which was their predecessor.
Again, these servers are made for AI-related tasks, and when we talk about server level, it has a lot to do with large language models. For that, eight MI300X GPUs, each with a powerful 192GB of unified HBM3 memory, are the center of the G593 servers; this arrangement is perfect for processing huge language models.
These robust MI300X-based servers are powered by two AMD EPYC 9004 processors. With twelve expansion slots to accommodate GPUs, networking cards, or storage devices, the servers provide a high degree of versatility. Hence, this configuration is good for scalability and adaptability to various computing requirements.
Key Specifications
G383-R80
- Platform: AMD Instinct MI300A APU
- CPU: 4 x AMD Instinct MI300A
- DDR5 memory: Not Applicable
- Storage bays: 8 x 2.5-inch Gen5 NVMe
- GPU: 4 x MI300A
- Expansion slots: 12 x FHFL
- Power supply: 3 + 1 2200W
G593 Series
- Platform: AMD Instinct MI300X GPU
- CPU: 2 x AMD EPYC 9004 processors
- DDR5 memory: 24 x DIMMs, 1DPC
- Storage bays: 8 x 2.5″ Gen5 NVMe
- GPU: 8 x MI300X
- Expansion slots: 4 x FHHL & 8 x low-profile
- Power supply: 4 + 2 3000W
These cutting-edge servers are slated to hit the market in the first half of 2024, promising to redefine the landscape of HPC and AI.
If you are a business or a team working on large language models or need high computing power for AI-intensive tasks, these servers can provide more value.