Intel has had trouble getting its chips in the hands of customers on time, but is providing the next best thing – to try out those chips in the cloud.
Delayed chips such as Sapphire Rapids server processors and Habana Gaudi 2 AI chip will be available on a platform called the Intel Developer Cloud, which was announced at the Intel Innovation event being held in San Jose, California.
The chipmaker is pitching the cloud service as a “try and buy” service for a range of server chips, including AI chips and GPUs. The Developer Cloud also hosts the company’s developer tools, libraries and toolkits, and sample code.
“It won’t just be the Xeon Scalable, it’ll be the Xeon Scalable with high-bandwidth memory, you’ll get access to Xeon-D processors, the Habana Gaudi 2 processors… and our GPUs,” said Lisa Spelman, corporate vice president and GM for Intel Xeon products, during a press briefing ahead of the show.
The announcement may feel like a distraction from the multiple delays of Sapphire Rapids, which is expected to start shipping early next year. The delays of Sapphire Rapids – which has support for new technologies that includes PCIe 5.0 – has caused cloud companies and server makers to reshape product roadmaps.
But Intel had been talking about the developer cloud service behind closed doors at the Vision conference in May. The goal behind the cloud service is also to provide access to the latest hardware so developers can prepare applications in time of volume shipments for chips.
Intel’s customers will be able to simulate a heterogeneous computing environment with a range of chips in the cloud environment. Developers will get access to tools like OneAPI and the SYCL layer so standard C++ applications can be deployed without calling specific GPUs or AI chips for acceleration.
Intel is following in the footsteps of Nvidia, whose Launchpad service offers access to the company’s latest chips such as Hopper GPUs, on which developers write and test AI applications. Nvidia has built data centers around the world for Launchpad.
For both Intel and Nvidia customers, the offerings are a cheap and quick way to access the hardware. In many cases, developers don’t have the hardware, or can’t afford it, especially GPUs.
More development environments are turning to the DevOps model, with many iterations and software releases on a daily basis. The cloud-native deployments on an Intel platform could help port final versions of applications easily to other public and private cloud services using Intel chips.
The Intel Developer Cloud goes into beta this week, and will include Sapphire Rapids as an early offering. Intel is introducing unique tweaks in Sapphire Rapids for applications like AI and databases, and developers will be able to test those features in the cloud.
One new on-chip feature in Sapphire Rapids called the memory analytics accelerator provides hardware acceleration for analytics applications. Customers will be able database analytics workloads faster, while improving power efficiency.
“This accelerator lightens the load from CPU cores to accelerate both compression and decompression for common database use cases,” Spelman said.
Another new feature is a matrix extension called AMX to improve performance for applications such as artificial intelligence that rely on matrix calculations
“This allows developers to boost their productivity and their deployment pace and really kind of increases the simplicity and using auto tuning functionality of the neural coder and it allows them to take advantage – have that performance boost that’s built into the hardware acceleration,” Spelman said.
Intel already has AI-specific features like DLBoost – or deep-learning boost – on its Xeon chips.
CPUs are generally not tuned for AI-style computing, which is deterministic. CPUs are considered peripherals, with the responsibility to execute instructions and put it into the memory, with the heavy lifting carried out by AI chips such as GPUs or ASICs.
Intel executives didn’t provide additional details on the actual shipment dates for Sapphire Rapids. But the chip is going through a rigorous validation cycle, said Ronak Singhal, senior fellow at Intel.
“When we find issues sometimes then we’ll push out the schedule, but we think that’s the right trade-off for our customers and what they’re expecting from us,” Singhal said.
Intel also shared some further details about its CPU core subscription service for customers based on their computing needs. The subscription model is available through partners that include HPE, Lenovo and PhoenixNAP.
“We’re going to build on that and announce what we’re calling the ‘activation model’ through Intel-on-Demand, which will be available on fourth-gen Intel Xeon Scalable Processors,” Spelman said.
The activation model will allow customers to turn on additional accelerators, security, and other features beyond the base configuration of the original SKU.
“This gives them greater flexibility and choice for when it’s needed,” Spelman said.
Additional OEMs such as Supermicro, Lenovo and H3C are testing out the new services.
“We’ll continue to add more providers and invest in features, software and services to support the industry,” Spelman said.
Intel also has a program to offer subscription services for software running on its server chips. Spelman didn’t provide details on whether its software and hardware subscription services are joint or separate offerings, or how they would interoperate.
“The focus is going to be around specific technologies, allowing customers to essentially opt in when they know that they will value and get benefit from those technologies. You can imagine how this can be expanded into the future for things like cores on and off,” Singhal said.