Having worked at AMD twice, Mr. Koduri, senior vice president and general manager of the Accelerated Computing Systems and Graphics Group at Intel, announced at Intel’s investor meeting last week that Intel would launch an APU named Falcon Crest Shores in a couple of years. Koduri knows the inner workings of an APU and what it takes to get a harmonious relationship and mutual respect between the GPU and CPU concerning memory and I/O access. Getting the right balance of GPU shaders to CPU cores with the wattage budget of the package is no simple trick, and it has taken AMD several years to get the recipe right. The fact they got it right is evidenced in their gains in notebook market share (see How much is significant?).
Koduri said, “We are working on a brand-new architecture codenamed Falcon Shores,” Koduri said. “Falcon Shores will bring x86 and Xe GPU acceleration together into a Xeon socket, taking advantage of next-generation packaging, memory, and I/O technologies, giving huge performance and efficiency improvements for systems computing large datasets and training gigantic AI models.”
“We expect Falcon Shores to deliver more than 5× performance per watt, more than 5× compute density increase, and more than 5× memory capacity and bandwidth improvement [relative to current platforms], all in a single socket with a vastly simplified GPU programming model,” he continued. “Falcon Shores is built on top of an impressive array of technologies … including the Angstrom-era process technology, next-generation packaging, new extreme bandwidth shared memory being developed by Intel, and industry-leading I/O. We are super excited about this architecture as it brings acceleration to a much broader range of workloads than the current discrete solutions.”
Putting a GPU and CPU together is not a novel or new trick. Besides AMD, Intel, MediaTek, Nvidia Qualcomm, Samsung, and others have built SoCs with an integrated GPU. Intel is the world’s largest GPU supplier because of its integrated GPU. But there are GPUs, and then there are GPUs. AMD’s APU GPU is a monster compared to all the rest. Lots of shaders, lots of L3 cache, and an almost perfectly balanced memory manager, which is the key to performance.
And Intel has already been building an integrated Xe CPU; their 12th gen Alder Lake series reportedly uses Xe.
So what’s the big deal about Falcon shorts Shores. Chiplets son, CHIP-lets. Falcon Shore won’t be a monolithic hunk of silicon like all others; it will be two or more CHIPLETS and take advantage of Intel’s fabulous Foveros substructure.
It’s a clever idea. How do we know? Because AMD is going to do it too! See Back to the future—full-circle on GPU construction.
And, just for the record, let’s remember that Intel did an integrated GPU before AMD did, even though AMD announced one first. (see Famous Graphics Chips: Intel’s GPU History).
The big question, well, big for me at least, is when Intel brings out Falcon Shores, will they pull the Xe GPU from the 13th or 14th Gen CPUs? Or will those be entry-level CPU/GPUs? Presumably, Intel will charge more for tier APU than a regular CPU. AMD will love that; nothing like the smell of a little ASP rise in the morning. Intel’s former Senior VP at Intel Corporation, President of Intel Israel, Eden Mooly, used to say he hated GPUs. Why? Because he said, they eat up 60% of the dies, and we don’t get paid a dime for them. So Falcon Shores is an opportunity for Intel to get paid for the GPU they’ve been giving away. The shareholders are sure to love that.