Deep Learning: Intel launches its new processors for data centers


Intel took advantage of its Intel Vision event to announce on Tuesday the launch of its new deep learning processors for training and inference, Habana Gaudi2 and Habana Greco, making AI more accessible and more valuable for data center operators. data. At its Intel Vision event, the chipmaker also shared details about its IPU and GPU portfolios, all aimed at business customers.

These new deployments are not trivial for Intel. “AI is driving the data center,” said Eitan Medina, COO of Habana Labs, Intel’s data center team focused on deep learning AI processor technologies. “It’s even the largest and fastest growing application. But different customers use different blends for different applications,” he said.

Diverse use cases explain Intel’s investment in a variety of data center chips. Habana processors are designed for customers who require deep learning computations. The new Gaudi2 processor, for example, can improve vision modeling for applications used in autonomous vehicles, medical imaging and defect detection in manufacturing. As a reminder, Intel bought Habana Labs, a programmable chipmaker based in Israel, for around $2 billion in 2019.

Chips engraved in 7 nm

The second generation Gaudi2 and Greco chips are both etched in 7 nanometers, an improvement over the 16 nanometers of the first generation. They are built on Habana’s high-efficiency architecture. According to Intel, Gaudi2 offers twice the training throughput of Nvidia’s A100-80GB GPU for the ResNet-50 computer vision model and the BERT natural language processing model.

“Compared to the A100 GPU, implemented in the same process node and of roughly the same size, Gaudi2 offers significantly higher learning performance, as demonstrated by the apples-to-apples comparison on major loads of work”, we explain on the side of Habana Labs. “This deep learning acceleration architecture is fundamentally more efficient and backed by a strong roadmap.”

Compared to the first generation Gaudi, Gaudi2 offers up to 40% better price performance in the AWS Cloud with Amazon EC2 DL1 instances and on-premises with the Supermicro X12 Gaudi Learning Server. The chip introduces an integrated media processing engine for compressed media and host subsystem offloading. Gaudi2 triples the built-in memory capacity from 32 GB to 96 GB of HBM2E with a bandwidth of 2.45 TB/sec. This chip integrates 24 x 100GbE RoCE RDMA NICs, on-chip, for scaling and scaling using standard Ethernet.

The new Arctic Sound GPU is unveiled

Gaudi2 processors are now available for Habana customers. Habana has partnered with Supermicro to bring the Supermicro Gaudi2 training server to market this year. Meanwhile, the second-generation Greco inference chip will be available to select customers in the second half of this year. The second generation of Greco includes increased on-board memory, which increases bandwidth fivefold and increases on-chip memory from 50 to 120 MB. It also adds additional decoding and media processing performance, while offering a smaller form factor for computational efficiency.

“Gaudi2 can help Intel customers train increasingly large and complex deep learning workloads with speed and efficiency, and we anticipate the inference efficiencies that Greco will bring.” Sandra Rivera, executive vice president of Intel, in a statement.

Intel also on Tuesday unveiled an expanded roadmap for its infrastructure processing unit (IPU) portfolio. Intel originally built IPUs for cloud computing giants — hyperscalers like Google and Facebook — but it’s now expanding those units’ access to other customers. Intel will ship two IPUs next year: Mount Evans, Intel’s first ASIC IPU, and Oak Springs Canyon, Intel’s second-generation FPGA IPU that ships to Google and other service providers .

In 2023 and 2024, Intel plans to launch its third-generation 400GB IPUs, codenamed Mount Morgan and Hot Springs Canyon. In 2025 and 2026, Intel plans to ship 800 GB IPUs to customers and partners. Intel also gave news of its data center GPU, codenamed Arctic Sound. Designed for media transcoding, visual graphics and inference in the cloud, Arctic Sound-M (ATS-M) is the industry’s first discrete GPU with an AV1 hardware encoder. It offers a performance of 150 trillion operations per second (TOPS). The ATS-M will be available in two form factors and more than 15 system designs from partners including Dell, Supermicro, Inspur and H3C. Its launch is scheduled for the third quarter of this year.

Source: ZDNet.com





Source link -97