'A Major Engineering And Architecture Transformation'
Intel is looking to give customers and partners confidence in its next-generation products with a new “transistor-resilient design” approach that will impact everything from its forthcoming Tiger Lake mobile processors to multiple discrete GPU products it plans to release within the next year.
The Santa Clara, Calif.-based company made several new product and technology disclosures at its Intel Architecture Day pre-briefing event on Tuesday, including news that discrete GPUs for laptops and servers will ship later this year and that it has delivered the largest single intranode enhancement with the new version of its 10-nanometer process that will power Tiger Lake.
With the disclosures, the company is hoping to show that it is investing in next-generation design methodologies that will help it overcome any issues in the future with internal manufacturing capabilities, which recently prompted the company to delay the launch of 7nm products by six months.
“We at Intel are going through a major engineering and architecture transformation that sets us up for the next decade and reimagines our design methodology, architecture and the way we solve problems for our users and customers,” said Raja Koduri, an AMD veteran who is now Intel‘s chief architect and who leads the Architecture, Graphics and Software Group.
What follows are six big announcements from Intel Architecture Day 2020, made public Thursday, that range from Intel‘s big discrete GPU and Tiger Lake news to its 10nm desktop processor plans and how it plans to take a building-block approach for client processors in the future.
Intel's First Discrete GPUs For Servers And Laptops To Launch This Year
Intel seems intent on delivering on the promise it made in 2018 when the company said its first discrete GPUs would come out in 2020. At this week‘s Intel Architecture Day, the company said it will launch its DG1 GPU for laptops and a newly disclosed SG1 GPU for servers this year.
Revealed at CES 2020 at the beginning of the year, DG1 is in production and on track to ship later this year, according to Intel, but the company did not say how many laptops will use it.
SG1, which combines the power of four DG1 GPUs, is based on Xe-LP, Intel‘s low-power GPU microarchitecture, like the DG1, and it and will come in a small form factor for servers targeting low-latency, high-density Android cloud gaming and video streaming. The company said SG1 will enter production soon and ship later this year.
Intel also revealed two other GPU types, one based on the company‘s Xe high-performance GPU microarchitecture for data centers and another based on the newly disclosed Xe high-performance gaming GPU microarchitecture for desktop PCs.
The company said its first Xe-HP GPU for servers—which will ship next year and is already sampling with customers—represent the “industry‘s first multi-tiled, highly scalable, high-performance architecture” that will provide GPU scalability, artificial intelligence optimization and media performance. With its multi-tile nature, the Xe-HP GPU will come in three configurations—one tile, two tiles and four titles— the latter two of which Intel said will make it function like a multi-core GPU.
The company said Xe-HP can also deliver on the compute side, providing the most single-precision floating-point, or FP32, performance on a single GPU package. The company said it can also deliver high performance for media workloads and demonstrated a single Xe-HP tile transcoding 10 full streams of 4K video at 60 frames per second and multiple Xe-HP tiles.
“The single-precision FP32 performance numbers of over 40 teraflops is already impressive and respectable, considering that the best GPU in the market today is the 20 teraflops range,” said Raja Koduri, Intel‘s chief architect and head of the Architecture, Graphics and Software Group.
Intel‘s Xe-HPG GPU for desktop PCs is also set to start shipping next year, and it will feature a new memory subsystem based on GDDR6 and hardware-accelerated ray tracing. The company said the Xe-HPG GPU will incorporate “building blocks” from Xe-LP for efficiency, Xe-HP for scalability and Xe-HPC for optimized compute frequency. Xe-HPC is the basis for the company’s previously announced Ponte Vecchio GPU for high-performance computing and AI workloads.