AMD reveals how it plans to throw Intel (INTC) and Nvidia (NVDA) into the data center

AMD (AMD) is hungry to take advantage of a first-class production process in the growing server CPU and GPU markets.

At an event held in San Francisco on Tuesday morning, AMD revealed the first GPUs produced using the Taiwan Semiconductor (TSM) cutting-edge 7-nanometer (7nm) production process. The company also previewed its first 7nm server CPUs, and shared details of the Zen 2 CPU core architecture that will drive 7nm products on both the PC and server markets.

In addition, AMD announced that (AMZN) has joined the ratings from cloud infrastructure providers offering cloud computing instances that rely on AMD's Epyc server CPUs.

Announcements increased to AMD's shares, which sold heavily in October thanks to a technical correction and the easy Q4 guidance issued in the company's third quarter. AMD increased 3.9% in Tuesday trading to $ 20.68. AMD's first 7nm GPUs, the Radeon Instinct MI60 and MI50, are based on the Vega GPU architecture, launched in 201[ads1]7, and intended for a server AI and high performance computing (HPC) workload. As such, they aim at Nvidia's (NVDA) Volta-architecture server GPUs, which dominate the market for accelerators used to train AI / machine learning algorithms and also require a large proportion of the HPC accelerator market. MI60 is expected to deliver by year end, and MI50 by Q1 2019.

MI60, supported by a healthy 1 terabyte per second (TB / s) of memory bandwidth, is said to deliver 14.8 teraflops per second (TFLOPS) with single-precision (FP32) performance and 7.4 TFLOPS dual-precision (FP64) performance. Although FP32 performance is only slightly improved compared to AMD's MI25 GPU, made using an older 14nm production process, FP64 performance greatly improves. In addition, the MI60 chip has a smaller nozzle size than the MI25's, which provides power and production savings.

The MI60 official FP32 and FP64 performance is close to Nvidia's flagship Tesla V100 GPU, launched in 2017 and made with an older 12nm process. It would not be surprising to see Nvidia, which has lent on both performance and a strong developer ecosystem to dominate the AI ‚Äč‚Äčtraining market (and command premium prices), launch 7nm server GPUs in 2019.

AMD promises 7nm server CPUs, the code name Roma and for now just said to launch in 2019, will deliver an impressive 2x performance gain per CPU connector compared to AMD's first generation Epyc server CPUs (the code name of Naples), thanks to the use of a 7nm process and Zen 2 kernels . Roma CPUs will contain up to 64 nuclei (first genes Epyc peaks out to 32 kernels) and support twice as much memory bandwidth per channel as its predecessors. As the first Epyc processors, they will have modular architecture through which multiple CPU chips are placed inside the same package.

"I've been in the semiconductor industry for a long time. You do not get 2x," proclaimed CEO Lisa Su. She added that AMD is "deep" in the development work of Rome's successor, the code name of Milan.

AMD states that the use of 7nm makes it possible to double CPU core density versus older chips with its first generation of Zen CPU cores (they rely on 12nm and 14nm processes), and also make it possible to deliver 50% Reduction in power consumption when performance is unchanged. Apple's A12 Bionic system-on-chip (SoC), which runs the iPhone XS and XR, is also dependent on TSMC's 7nm process.

The company has not yet shared details about its first 7nm PC processors and GPUs. There is speculation that AMD will do it at the annual CES fair in January, where Su gives a key.

At the same time, the AWS partnership follows similar offers to deliver chips to the skate platforms for Microsoft (MSFT), Baidu (BIDU) and Tencent (TCEHY). Amazon, still the world's largest player of cloud infrastructure, rolls out three types of cloud computing instances with first-generation Epyc processors.

The AWS Agreement "is a testament to Epyc's abilities," says technical analyst Patrick Moorhead. He was impressed that AWS customers can get a 10% cost reduction by switching to Epyc from similar databases that rely on Intel INTC Xeon processors , and can do it without the need for any software or script changes.

AMD announcements come one day after Intel unveiled Xeon server CPUs based on its next Cascade Lake platform. The Cascade Lake chips, which begin shipping in the first half by 2019, supports up to 48 kernels per CPU, as well as memory modules with Intel's Next Gen Memory.

Like Intel's older Skylake CPUs, Cascade Lake bases itself on Intel's 14nm production process code, and a sequel platform called Cooper Lake, which later in 2019. After several delays, the first server CPUs are based on Intel's 10nm hub, which is considered competitive with TSMC's 7nm hub, is not expected until 2020. [19659002] Intel shares dropped 0.9% to $ 47.25 on Tuesday. Nvidia's shares closed 0.3% to $ 211.06.

Amazon and Microsoft are inventories in Jim Cramer's Action Alerts PLUS Member Club . Do you want to be notified before Jim Cramer buys or sells AMZN or MSFT? Learn more now.

Source link

Back to top button