Wednesday, December 6th 2023
AMD Announces XDNA 2 NPU Architecture for Next Gen "Strix Point" Mobile Processors Arriving in 2024
AMD in its Ryzen 8040 series "Hawk Point" mobile processors announcement made the first mention of XDNA 2, its next-generation on-chip neural processing unit (NPU) architecture. Above all, the XDNA 2 NPU is expected to introduce an over 3 times improvement in performance over the first generation XDNA NPU powering the Ryzen 7040 series "Phoenix" processor. XDNA 2 is making its debut with AMD's next-generation Ryzen "Strix Point" mobile processor that the company looks to launch in 2024. While "Phoenix" offers 10 TOPS of NPU performance, AMD mentions an "over 3 times" performance improvement, which probably puts this figure at 32 TOPS for "Strix Point."
The "Strix Point" mobile processor is rumored to debut faster "Zen 5" CPU cores, a possible CPU core count increase to 12, and a much more powerful iGPU based on the updated RDNA 3.5 graphics architecture, with some SKUs expected to feature CU counts as high as 32, and designed to square off against the iGPU of the Apple M3 Max processor. Besides "Zen 5" CPU cores and RDNA 3.5 iGPU, we now know that even the NPU gets an overhaul with this XDNA 2 announcement, and a possible 32 TOPS NPU performance.
The "Strix Point" mobile processor is rumored to debut faster "Zen 5" CPU cores, a possible CPU core count increase to 12, and a much more powerful iGPU based on the updated RDNA 3.5 graphics architecture, with some SKUs expected to feature CU counts as high as 32, and designed to square off against the iGPU of the Apple M3 Max processor. Besides "Zen 5" CPU cores and RDNA 3.5 iGPU, we now know that even the NPU gets an overhaul with this XDNA 2 announcement, and a possible 32 TOPS NPU performance.
10 Comments on AMD Announces XDNA 2 NPU Architecture for Next Gen "Strix Point" Mobile Processors Arriving in 2024
What's an NPU - is that like a CPU or GPU?
What are are TOPS - are they like FLOPS, ROPS, or OPS?
I'm guessing we're playing Marketing Bullshit Bingo here, in "hey, let's ride this AI-hype gravy train" land so uh, "BINGO!"
What did I win?!
(Seriously, all anyone cares about is "how fast does it load webpages" and "how many frames a second"?)
The NPU is going to be more efficient at ML stuff so that Windows laptop battery life will be less embarrassing vs a Macbook when you use software that makes use of it. And that list of software is growing.
Current CPUs provide enough performance for loading webpages and GPUs provide enough performance for simple gaming (but definitely not enough for true VR).
What has been missing for all these years in computers and software is useful INTELLIGENCE, CLEVERNESS, WISDOM and INTUITION. I don't care if a website loads in 1 second or in 0.0000001 second, really.I've literally been waiting 20 years for OSes to finally get artificially intelligent. There were books in the 90s about this and it has not happened yet, possibly because of performance reasons, but we are finally getting to a point when performance is gonna be high enough.
My bet is Deck 2 will use a 6 core Zen 4 (2 x Zen4 + 4 x Zen4C) with 12 RDNA 3.5 CUs, built on TSMC's N5, and 2 years later we'll see a refresh on N4, just like we saw recently.