When Intel announced its Xe family of graphics processors in 2020, it introduced up to four microarchitectures. This eventually led to large delays, Xe-HP data center GPU cancellations, and driver issues. The company now says it has learned its lesson. The Xe 2 ‘Battlemage’ family has fewer microarchitectures to offer. Still, there are his Xe2-LPG and Xe2-HPG microarchitectures for different kinds of GPUs.
“We have Xe, we have Xe 2, and in that Xe 2 generation we have Xe-LPG, we have HPG (…), we have small variations (…), and this is our big lesson,” said Tom Peterson. said. In an interview with an Intel Fellow hardware lux“The idea was that we had to optimize for each segment, build separate chips, do separate verifications. I think it’s better to think of it like a hard IP business.”
With the Battlemage generation of GPUs, Intel will stick to the Xe2-LPG and Xe2-HPG microarchitectures. It’s unclear if Xe2-HPC is still planned, but officially Intel’s next-generation GPU for HPC, codenamed Realto Bridge, will feature ‘Enhanced Xe-HPC’ cores, not Xe2-HPC cores. Based on
“As we progressed through the roadmap, we realized that the QA process and segmentation were very costly. We knew we needed to differentiate our IP and customize it for each segment,” said Peterson. increase. “[…] Take only one thing and go anywhere without changing it. That’s the strategy we’re looking at going forward. Because that’s the only way to reuse the IP and make it actually work. ”
Intel’s original Xe family contained four microarchitectures. Xe-LP for integrated and low-end standalone GPUs, Xe-HPG for discrete desktop graphics cards, Xe-HP for cloud data centers, and Xe-HPC for high performance computing. Developing a vastly different GPU microarchitecture has advantages in terms of performance and die size. For example, with Intel selling so many client CPUs, a slightly smaller iGPU without features like ray tracing could translate into hundreds of millions of dollars in cost savings. Meanwhile, the redesigned Xe-HPC core leads to improved performance.
But Intel had to design, validate, and manufacture up to nine different GPU variants to deliver on that promise. That’s a pretty huge number for one product family, even for a giant like Intel.
Ultimately, the company decided that the Xe-HPG and Xe-HPC GPUs could handle the data center verticals, so the Xe-HP had to be phased out. However, the company spent valuable time developing his Xe-HP hardware and software stack. Additionally, the Xe-LP and Xe-HPG are so different that the company had to tune the integrated and standalone GPUs separately and develop different drivers for the embedded and discrete graphics processors. As a result, the Xe-LP iGPU was in time for 2020, while the Xe-HPG and Xe-HPC GPUs were delayed by at least a year.
The company says it learned a lesson with its first-generation Xe graphics processors, but it remains to be seen if the Xe2 Battlemage GPUs will be ready to compete with the best graphics cards from AMD and Nvidia.
