Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124

Ah, the teenage years of chip announcements: those periods when companies roll out modified versions of the big structural changes we’ve seen over the previous year (or earlier). In AMD’s case, some updates to its chipset lines were necessary. But since I did it Big announcements at last CESWhat should we realistically expect? Consumer Electronics Show 2026? The answer is to refresh the Ryzen AI lines that focus primarily on mobile processors, moving from the 300 series to the 400 series, and a bit more.
Don’t miss any of our unbiased technical content and lab reviews. Add CNET As Google’s preferred source.
The “more” comes in the form of an addition to the Ryzen X3D lineup among AMD’s gaming-focused 9000X series desktop processors. The new Ryzen 7 9850X3D slots up the Ryzen 7 9900X3D and is essentially the same processor, using slightly better-performing dies from the same batch and boosting the clock speed by 100MHz. In practice, the 9850X3D runs at 5.6GHz, compared to 5.5GHz for the 9900X3D. AMD says this translates to a performance improvement of about 7%.
The Acer Swift Go 16 AI includes a Ryzen AI 9 465 processor.
In the same vein, AMD has added two processors to the Ryzen AI Max Plus 300 series: the Ryzen AI Max Plus 392 and the Ryzen AI Max Plus 388, which slide above the old Ryzen AI Max 390 and 385, respectively. The only change is in the GPU: it jumps from 32 compute units to 40 compute units, effectively “adding” chips with 8090S graphics. This upgrade boosts the GPU’s performance for both gaming and AI and provides affordable alternatives to the Ryzen AI Max Plus 395. People don’t always notice when their CPU is slow, but gamers definitely notice when their GPU is slow.
| Cores (Zn 5c/Zn 5) | Maximum boost frequency (GHz) | Total topics | NPU | GPU | GPU compute units | Maximum GPU Frequency (GHz) | |
|---|---|---|---|---|---|---|---|
| Ryzen AE9HX475 | 12 (8/4) | 5.2 | 24 | XDNA 2 up to 60 pieces | Radeon 890M | 16 cubic metres | 3.1 |
| Ryzen AE9HX470 | 12 (8/4) | 5.2 | 24 | XDNA 2 up to 55 pieces | Radeon 890M | 16 cubic metres | 3.1 |
| Ryzen i9 465 | 10 (6/4) | 5 | 20 | XDNA 2 up to 50 pieces | Radeon 880M | 12 cubic metres | 2.9 |
| Ryzen AI 7450 | 8 (4/4) | 5.1 | 16 | XDNA 2 up to 50 pieces | Radeon 860M | 8 su | 3.1 |
| Ryzen i7 445 | 6 (4/2) | 4.6 | 12 | XDNA 2 up to 50 pieces | Radeon 840M | 4 su | 2.9 |
| Ryzen i5 435 | 6 (4/2) | 4.5 | 12 | XDNA 2 up to 50 pieces | Radeon 840M | 4 su | 2.8 |
| Ryzen i5 430 | 4 (3/1) | 4.5 | 8 | XDNA 2 up to 50 pieces | Radeon 840M | 4 su | 2.8 |
Finally, there’s the Ryzen AI 400 series, which are generally somewhat faster versions of their 300-series counterparts. The HX versions of the 300-series are older — Launched in June 2024 — but it keeps mostly the same specs, with only minor speed bumps, like the 100MHz tweak mentioned earlier. They’ve gotten a boost in NPU performance too, hitting 55 TOPS and 60 TOPS for the HX 470 and 475, respectively, up from the 50 TOPS found in the rest of the XDNA 2 chips. The remaining models get some minor clock speed improvements and slightly faster memory support.
Ryzen AI Halo
AMD has assembled its own embedded desktop system specifically for native AI development (up to 200 billion parameter models) called Ryzen AI Halo. It is configured with a Ryzen AI Max chip and 128GB shared memory and supports multiple operating systems; It will be pre-loaded with a set of open source tools and AMD’s ROCm AI API stack.