Intel Details XPU Strategy, Launches New Server GPU, OneAPI Gold

Intel made several announcements as part of its unveil of OneAPI Gold and a new GPU product intended for the video processing / Android cloud gaming market.

First up: the new Intel Server GPU. This new instantiation of the Xe-LP architecture has four separate graphics chips and work in tandem to process video streaming workloads, video transcoding workloads, and as a solution for Android game streaming. Xe-LP is the same architecture Intel is shipping in the Xe Max and inside of Tiger Lake. Performance reviews of Tiger Lake chips have put it somewhat ahead of AMD’s Ryzen 4000 integrated GPU, and the Xe Max is the same chip that’s inside TGL laptops, just running at a higher clock speed.

“We are on a journey from CPU to XPU,” Intel SVP and manager of pretty much everything (architecture, graphics, and software) Raja Koduri told wfoojjaec on a recent call. “Our CPU architecture has built Intel and it also played its part to enable the entire world of computing, but we know the workloads have evolved and we are striving for mastery over additional XPU architectures that are super-efficient for graphics, media, and AI, memory, security, and networking.”

Like Xe Max in laptops, the Xe-LP chip Intel is launching today is aimed at media workloads more than gaming, though in this case, the company is talking up the solution as an Android gaming platform. Apparently, Intel is working with several companies, including Tencent and GameStream, to bring Android streaming services to market. Intel’s argument is that the Xe-LP makes an attractive partner to Xeon, allowing customers to standardize on all-Intel solutions for these products.

Intel can support 120 streams in a two-GPU configuration and possibly up to 160 simultaneous players depending on the game in question. Since each card has 4 GPUs, Intel is supporting 15 Android gamers per GPU at 30fps. Android games are not nearly as hardware intensive as PC titles, but 15 streams per GPU at 30fps still works out to a cumulative total of 450 frames per second. It’d be interesting to know how the company distributes the disparate streaming workloads across the card and between the two GPUs. It’s also possible that the user claims here are a bit optimistic, of course.

OneAPI Goes Gold

Intel’s other major announcement concerns its OneAPI initiative. OneAPI is a cross-standard programming model intended to standardize code development between various system components, including GPUs, FPGAs, and CPUs. Intel has always offered its own libraries and code development tools, and the company is leveraging that expertise to create an overarching framework for product development. Support for features like AVX-512 and DLBoost is baked-in, as you’d expect.

Intel has been talking about OneAPI for several years now, but the project has only just gone gold, with toolkit availability expected in December. Free versions will be available both locally and in the cloud, and Intel will immediately transition its Parallel Studio XE and Intel System Studio tool suites to OneAPI products.

OneAPI has generated a fair bit of news for Intel over the past few years, but it isn’t very clear if it’s changed Intel’s overall fortunes in the AI market. Until its high-end Xe cards are available, Intel is working with one hand tied behind its back, as far as AI performance is concerned. Even with AVX-512 and DLBoost, a modern Xeon server cannot match the raw performance of a GPU.

With OneAPI, Intel is building an infrastructure backend it helps will attract programmers to its platforms as new GPUs — not to mention consumer CPUs with AVX-512 support — roll out in mobile and desktop through 2021 and beyond. For those of you concerned about CUDA support, there’s already a backend built into the Data Parallel C++ compiler that allows DPC++ code to run on top of CUDA GPUs. DPC++ is Intel’s unifying language implemented in OneAPI that can address FPGAs, CPUs, and GPUs and any other compatible accelerators anyone brings to market.

This announcement seems to conclude Intel’s plans for GPU launches in 2020, which means we’ll have to wait until 2021 to see what the company brings to the table in consumer cards. Tiger Lake has impressed in mobile, which hopefully implies good things about the larger cards Intel will ship in 2021.

Continue reading

Third-Party Repair Shops May Be Blocked From Servicing iPhone 12 Camera

According to a recent iFixit report, Apple's hostility to the right of repair has hit new heights with the iPhone 12 and iPhone 12 Pro.

Massive ‘Kilonova’ Explosion Shows First Observed Birth of a Magnetar

We've never seen a magnetar come into being, but a new high-energy event several billion light years away might be the first — a kilonova that signals the merging of two neutron stars.

The Iconic Arecibo Observatory Will Be Demolished Following Cable Failures

The Arecibo dish was damaged following a series of cable failures, and the National Science Foundation (NSF) has decided it would be too dangerous to repair.

Famed Arecibo Observatory Collapses Following Cable Failures

The National Science Foundation (NSF) previously expressed concern this could happen, which is why it decided last month the dish would be demolished rather than repaired. Gravity took care of that a bit quicker than expected as the 900-ton suspended platform plummeted into the dish overnight, completely destroying the iconic instrument.