Just when you thought you had a moment away from any more AI-focused hardware news, AMD is leaping into the “AI PC” arena with its latest mobile laptop chips. The new Ryzen AI 300 series boasts better performance than either Intel or Qualcomm, plus neural processing capabilities.
The chips industry has always been a game of one-upmanship. Now more than ever, chipmakers are trying more than ever to compare their CPUs and GPUs not just on power but on the future promise of ultimate PC performance thanks to the proliferation of AI. AMD doesn’t have to fight against its longtime rival Intel for the consumer-end PC market, but Qualcomm, mainly thanks to the ARM-based Snapdragon X Elite and X Plus in the latest Copilot+ PCs.
AMD is mainly focused on hyping up its two new chip series. One is the new version of its Ryzen CPUs with the Ryzen 9000 series, and the other is the Ryzen AI 300 series stuffed with a new NPU in the form of XDNA 2. On laptops, the two chips will be the Ryzen AI 9 365 and the beefier Ryzen AI 9 HX 370. It’s technically the company’s third-gen AI-centric CPU, but this latest series is differentiated by its massive upgrade in neural processing.
Microsoft says it needs NPUs with at least 40 TOPS to mark them for Copilot+ PCs. Like the recent Snapdragon chips, the HX 370 and the 365 have the same NPU running at 50 TOPS. It’s one of the bigger boasts of AI performance from this past year, but despite the company’s claim it’s there to run more complex AI models, we still have to see if there will be any software worth these new neural components.
The 370 comes with 12 cores, 24 threads, and a 5.1 GHz max boost speed, while the 365 sits at ten cores, 20 threads, and 5.0 GHz max speed. The chips also have the RDNA 3.5 built-in GPU for some mobile graphics work or gaming. During the Taiwan Computex conference, several big OEMs promoted their first PCs that will sport the AI 300 chips. This includes Acer with its Swift series of laptops that are slated for later this year. That company had previously revealed a Swift 14 with a Snapdragon X chip and logos that glow when you’re using the NPU. Asus is also coming out the gate with AMD-powered Zenbook S 16 as well as the ProArt P16 laptop and the ProArt PX13 2-in-1.
These new chips sport the new centralized architecture from AMD, namely Zen 5, on the CPU end. The chipmaker claimed Zen 5 is a big update compared to Zen 4, which is supposed to handle twice the bandwidth of the last generation. What does this mean for PCs? AMD promises you’ll see up to 19% better benchmark performance in Geekbench 6 or 13% better in 3DMark’s physics tests, but that will depend on your PC’s exact chip and other architecture.
Zen 5 is different from the XDNA 2 NPU architecture. You can break up the TOPS speed in the NPU into a whole bunch of other categories, but AMD claims XDNA 2 is two times as power efficient and many times the total neural computing capacity of the previous gen’s 10 or 16 TOPS.
For those who could not care less about the productivity machines centering on the AI 300 chips, all you want to know is how the Ryzen 9000 series stacks up compared to the last generation and Intel’s latest. That includes the Ryzen 5 9600X, Ryzen 7 9700X, Ryzen 9 9900X, and at the tippy top end is the Ryzen 9 9950X. Most boast slightly higher clock speeds, but several, like the 9900x and 9700X, are far more power efficient with better TDP.
To pick on the big boy, that 9950X has 16 cores, 32 threads, and up to 5.7 GHz clock speeds. That’s technically the same specs as the Ryzen 9 7950X3D from the last gen. AMD is trying to hit Intel’s Core i9-14900K by claiming you’ll see marginally better framerates in games like Cyberpunk 2077 or F1 2023 and far better bandwidth for multitasking thanks to the new Zen 5 architecture.
All those gaming-centric CPUs should be arriving in July this year. There’s good news for anybody with the motherboard supporting the AM5 socket. AMD promises to support AM5 through 2027, so if you’re considering upgrading, you’ll have a chance in the next few years. After over eight years of running, AMD plans to end support for AM4 sometime in or after 2025. Zen 5 will continue to be specific to AM5.
The chipmaker said pricing is not set for the series 9000 chips, but we should know more closer to release in July.
Related Stories
A "battle of the giants" is unfolding in the market for chips for real-time artificial intelligence systems:
A separate "front line" in this confrontation is the development and implementation of SoM [System on Module] with Programmable Logic. This post is dedicated to one small "battle," on the example of which we want to show why, in our opinion, China could win this "war."
Since the announcement in 2019 by Xilinx (which then bore this name without a proud three-letter prefix), Versal ACAP (Adaptive Compute Acceleration Platform) chips were inaccessible to developers—the first development boards cost tens of thousands of U.S. dollars, and the difficulty of developing your own board for this chip would scare off anyone other than Tony Stark.
A lot of water flowed, and a lot of developers' tears were shed, but a silicon Versal is just as unavailable as The Palace of Versailles: the cheapest kit from AMD–VEK280 is sold by the official suppliers for $7K, excluding delivery and customs clearance. The classic argument in the style of "if you don't have money for an iron door, you don't need it" does not always work in the field of R&D—a rare developer will refuse to study a top-end chip at the expense of his employer, but even with this approach, the cost is too high.
[...] The problem is that the announcement of AMD Xilinx has so far remained an announcement, but the developers from Alinx, the Chinese company, did not waste any time. This company already is known for its inexpensive development boards with Zynq‑7000 and Ultrascale+ on board, not much different from SoM. Now they not only promised, but also mass-produced the SoM V100 with the XCVE2302-SFVA784-1LP-E-S chip (Versal AI Edge family) for $750 [1].
[...] There is, of course, a fly in the V100 ointment. The developers from Alinx were so inspired by Kria that they also used "legendary" Samtec connectors "well-liked" by all designers and engineers. Who among us hasn't drilled them from the side with the thinnest drill, forgetting to route that very necessary pin right in the middle in the inner row? However, to achieve the required transmission speeds with a compact size, there is hardly an alternative to Samtec connectors.
V100 SoM specs: 4 GBytes DDR4 (64-bit data-bus), 64 MBytes QSPI FlashROM, 8 GBytes eMMC, Gen4 ×8 PCI-Express, 8 x GTY up to 12.5 Gbps, 53 (for ARM cores) + 106 (for FPGA part) input/output lines, two Samtec ADF6-40-03.5-L-4-2-A-TR connectors with 160 pins each, single supply voltage 12V, and the dimensions are 65 x 60 mm.
Related:
- $119+ BeagleV "Powerful", Open-Hardware RISC-V Linux SBC Targets AI Applications
- AMD Confirms Ryzen AI 300 PCs Won't Feature Copilot+ Features At Launch
- AMD is Convinced its Latest CPUs will Crush Qualcomm AI Chips
(Score: 2, Insightful) by Anonymous Coward on Wednesday June 05 2024, @05:42PM
What a coincidence. People stopped upgrading their machines when performance flattened
so let's come up with some 'Do Not Want' bullshit that ups memory and processor requirements
(Score: 2) by looorg on Wednesday June 05 2024, @05:44PM (1 child)
Seems to be valid for most companies I would guess. They all seem to be convinced, at least publicly, that their latest product launch will crush all competition, after all their stock options and yearly bonuses depend on it.
After all they have crammed all the current buzzwords in there. "AI PC" is the next thing, I have seen it in many articles over the last month or so. From how we won't be able to live without it and all computers sold within a year or so will have special AI hardware in them etc. No longer just CPU, GPU and memory. Now your special AI circuit will matter to ... Riight.
(Score: 5, Interesting) by RamiK on Wednesday June 05 2024, @06:25PM
Thing is, if you want to run an AI model on your laptop instead of on the cloud, having an NPU for inference instead of using the GPU is the difference between hours and minutes of battery life.
Of course, that's if you actually want to run something of the sort...
Anyhow, NPUs are also useful for all sorts of low latency FFTs in general so they're fairly appropriate for video/audio processing in games: https://www.mdpi.com/2076-3417/14/1/405 [mdpi.com]
Of course, it will be years before worthwhile killer features come out.
compiling...
(Score: 0) by Anonymous Coward on Wednesday June 05 2024, @05:52PM
https://www.antipope.org/charlie/blog-static/2024/06/is-microsoft-trying-to-commit-.html [antipope.org]
The comments bring this all on-point
This is all a ploy to force you to upgrade, AGAIN