#intelgaudi2

Intel Computex 2024

In order to make AI affordable and available to everyone, Intel CEO Pat Gelsinger announced at Intel Computex 2024 the newest technologies that combine state-of-the-art performance and power efficiency in data centres, acceleration, and the AI PC experience. Intel is dedicated to driving AI potential for partners and customers, while also empowering open ecosystems. Intel is still at the forefront of driving the industry towards a sustainable and scalable future with its increased processing power, cutting-edge power efficiency, and affordable total cost of ownership.

Gelsinger gave a keynote address at the June 4–7 event in Taipei, Taiwan, where he introduced the Intel Xeon 6 processors with Efficient-cores (E-cores), revealed the cost of the Intel Gaudi 2 and 3 AI accelerator kits, and unveiled the ground-breaking Lunar Lake client processor architecture a ground-breaking design that expands the AI PC category even further.

Lunar Lake

For its forthcoming Lunar Lake client computing processor, which was completely rebuilt to deliver industry-leading core and graphics performance with unparalleled AI, Intel unveiled the architecture specifics.

In comparison to the previous generation, the new Performance-cores (P-cores) and Efficient-cores (E-cores) give incredible performance at up to 40% reduced system-on-chip power. Compared to the previous generation, a new neural processing unit is up to 4 times quicker, allowing for equivalent advancements in generative AI. Moreover, performance in gaming and graphics is 1.5 times better with the new Xe2 graphics processing unit cores compared to the previous version.

Starting in the third quarter of 2024, Lunar Lake will power more than 80 new AI PC designs from more than 20 partners.

Lunar Lake Structure: A Motion Picture

An animated film demonstrates the various technological levels that comprise Lunar Lake. Intel unveiled Lunar Lake architecture at Intel Computex in Taipei on June 4, 2024. When compared to the previous generation, the thin-and-light PC processor offers notable gains in performance and efficiency in the neural processing unit, central processing unit, and graphics processing unit. Intel Corporation is credited.

Pictures of Lunar Lake

Image credit to Intel

Lunar Lake Provides the AI PC with a New Architecture

Intel unveiled the Lunar Lake architecture at Intel Computex 2024, revealing the flagship chip for the upcoming AI PC generation. Lunar Lake focuses on power-efficient compute performance for the thin-and-light segment while providing a significant boost in graphics and AI processing capabilities. Intel Corporation is credited.

Intel Core 6

The Intel Xeon 6 family of processors, which comes with E-core and P-core variants, was introduced by Intel to handle a wide range of workloads and use cases, including scalable cloud-native apps and high-performance computation requirements for artificial intelligence.

The first member of the series, code-named Sierra Forest, is the Intel Xeon 6 CPU with efficient cores, making its debut at Intel Computex 2024. When compared to 2nd Gen Intel Xeon processors on media transcoding workloads, it offers a performance per watt boost of up to 2.6x and a rack-level performance gain of up to 4.2x. Its high core density and remarkable performance also allow for rack-level consolidation of 3-to-1.

The third quarter of 2024 will see Intel Xeon 6 processors with P-cores increase AI, high-performance computing, image processing, and data analytics.

Xeon 6 Picture

Image credit to Intel

An Overview of Intel Xeon 6 Processors in 60 Seconds

Three key points concerning Intel Xeon 6 processors with Efficient-cores (E-cores) are outlined in this one-minute summary. Known by the code name Sierra Forest, they were introduced on June 4, 2024, at Intel Computex in Taipei, Taiwan. Designed with E-cores, they enhance performance and reduce power consumption for applications that require high densities and scaling. The Intel Xeon 6 processors with Performance-cores (P-cores), code-named Granite Rapids, will shortly trail this initial member of the family. Workloads requiring additional computation, such AI and computer vision, will be supported. Intel Corporation is credited.

Kulim, Malaysia; San Jose, Costa Rica: Intel Manufacturing (B-Roll)

The Intel Kulim Assembly Test (KuAT) plant in Kulim, Malaysia, and the Costa Rica Assembly Test (CRAT) site in San Jose, Costa Rica, are depicted in B-roll film from May 2024. It features footage of Intel Xeon 6 processors with efficient cores (code-named Sierra Forest) and Intel personnel inside the buildings. Intel Corporation is credited.

Gaudi Intel

With a price-performance advantage that offers customers options, quick deployment, and a lower total cost of operation, the Intel Gaudi architecture delivers the generative AI performance they desire.

Intel revealed at Intel Computex that a typical AI kit, which includes eight Intel Gaudi 2 accelerators and a universal baseboard (UBB), is expected to cost system providers about one-third as much as comparable competition systems. The kit is available to system providers for $65,000. Eight Intel Gaudi 3 accelerators with a UBB will be included in a kit that will retail for $125,000, which is roughly two-thirds less than comparable competitor platforms2.

Additionally, Intel disclosed that the launch of Intel Gaudi 3 systems is anticipated from six new system vendors. With plans to sell Intel Gaudi 3 systems, ASUS, Foxconn , the Gigabyte, Inventec, Quanta, and Wistron join Dell, HPE, Lenovo, and Supermicro.

Intel Tech Tour

Intel held its third-annual Technology Tour in Taiwan just a few days before Intel Computex, providing worldwide media and analysts with presentations on Xeon 6 and Gaudi accelerators as addition to an in-depth look at Lunar Lake architecture. Bringing next-generation technologies to life is the responsibility of business and technical leaders from throughout Intel, and these leaders delivered keynote addresses and technical deep dives over the two-day event.

Gaudi 2 Intel

For GenAI Performance, Intel Gaudi 2 is Still the Only Benchmarked Option Besides NV H100. The results of the industry-standard MLPerf v4.0 benchmark for inference were released by MLCommons. The company’s dedication to bringing “AI Everywhere” with a wide range of competitive solutions is reinforced by Intel’s results for its Intel Gaudi 2 accelerators and 5th Gen Intel Xeon Scalable CPUs with Intel Advanced Matrix Extensions (Intel AMX).

When it comes to generative AI (GenAI) performance, the Gaudi 2 Intel AI accelerator is still the sole benchmarked substitute for the Nvidia H100, offering excellent value for the money. Moreover, Intel continues to be the only manufacturer of server CPUs to provide MLPerf findings.

According to MLPerf Inference v3.1, the average improvement in Intel’s 5th Gen Xeon results over 4th Gen Intel Xeon processors was 1.42x. The Reason It Matters Intel’s MLPerf findings provide clients industry-standard standards to assess AI performance, building on its training and inference performance from prior MLPerf rounds.

Intel Gaudi 2 vs Nvidia NV H100 High-Performance AI Accelerator: This robust GPU is intended for usage in data centres for applications such as real-time deep learning inference, rapid data analytics, high-performance computing (HPC), and data analysis. Leading Performance: The H100 offers up to 7x better performance for HPC workloads than prior versions. LLM, or large language model Amicable: With its specialised Transformer Engine, which is intended to handle large LLMs, conversational AI activities may be performed up to 30 times quicker. Scalability and Security: To handle exascale workloads, the NVIDIA NVLink Switch System connects up to 256 H100 GPUs, and NVIDIA Confidential Computing, a built-in feature, secures data and applications. Intel Gaudi 2 The popular large language models (LLMs) and multimodal models that are covered by the Intel Gaudi software package are still growing. Intel submitted the Gaudi 2 accelerator results for the state-of-the-art models Llama v2-70B and Stable Diffusion XL for MLPerf Inference v4.0.

Hugging Face Text Generation Inference (TGI) has been quite popular, and Gaudi’s Llama work took use of this by using the TGI toolbox, which allows for tensor parallelism and continuous batching, thereby improving the efficiency of real-world LLM scaling. Gaudi 2 yielded 8035.0 and 6287.5 tokens-per-second for offline and server tokens, respectively, for Llama v2-70B. Gaudi 2 yielded 6.26 and 6.25 offline samples per second and server queries per second on Stable Diffusion XL, respectively.

Intel Gaudi 2 specs 7nm process technology Dual matrix multiplication engines (MME) and 24 programmable tensor processor cores (TPC) make up the heterogeneous compute architecture. Memory: 48 MB of SRAM and 96 GB of HBM2E onboard memory Networking: 24 on-chip integrated 100 Gbps Ethernet ports The Gaudi 2 Intel delivers outstanding speed and scalability and is optimised for deep learning training and inference applications. It has a heterogeneous computational architecture with twin matrix multiplication engines and 24 programmable tensor processor cores, and it is based on a 7nm manufacturing technology. The Gaudi 2 can effectively perform a range of deep learning tasks because to its design.

Additionally, the Gaudi 2 Intel has 96GB of HBM2E memory built-in, offering enough of capacity for data access. Moreover, the Gaudi 2 has 24 on-chip 100 Gbps Ethernet connectors, allowing several Gaudi 2 Intel accelerators to communicate at high speeds. Because of this, the Gaudi 2 is a good fit for deep learning clusters of any size.

Intel Gaudi 2 price These findings indicate that the Intel Gaudi 2 is still a competitive price. AI Will Be Everywhere in Intel Vision 2024

They are pleased to announce Intel Vision, which will be held in Phoenix, Arizona, on April 8–9, 2024. their flagship event, Intel Vision, brings together elite leaders in business and technology to discuss the most recent developments in the industry and solutions related to client, edge, data Centre, and cloud innovations.

Sign up now to take part in thought-provoking roundtables, captivating demonstrations, and cutting-edge AI insights with Intel executives and distinguished guests that will help you realise your technological vision.e/performance, a crucial factor to take into account when examining the total cost of ownership (TCO).

Concerning the Intel 5th Generation Xeon Outcomes: With advancements in both hardware and software, Intel’s 5th generation Xeon processors outperformed 4th generation Intel Xeon processors in MLPerf Inference v3.1, with a geomean increase of 1.42x. For instance, the 5th Gen Xeon entry demonstrated around 1.8x performance increases over the v3.1 submission for GPT-J with software optimisations including continuous batching. In a similar vein, MergedEmbeddingBag and further Intel AMX optimisations allowed DLRMv2 to provide around 1.8x speed increases with 99.9 accuracy.

Intel takes great pride in working together with OEM partners, like Quanta, Supermicro, Cisco, Dell, and WiWynn, to enable them to submit their own MLPerf results. Additionally, beginning in 2020, Intel has provided MLPerf results for four generations of Xeon CPUs; in many of these submissions, Xeon serves as the host CPU.

How to Utilise Intel Developer Cloud AI Solutions: The Intel Developer Cloud offers evaluations of 5th generation Xeon CPUs and Intel Gaudi 2 accelerators. Users may manage AI computing resources, execute training and inference production workloads at scale (LLM or GenAI), and do much more in this environment.

What to Watch For: Stay tuned for an update on Intel Gaudi 3 AI accelerators and additional information about Intel’s plan to deliver “AI Everywhere” at Intel Vision 2024.