Monday, January 6th 2025
AMD Debuts Ryzen AI Max Series "Strix Halo" SoC: up to 16 "Zen 5" cores, Massive iGPU
AMD at the 2025 International CES debuted the Ryzen AI Max 300 series of mobile processors. These chips are designed to go up against the Apple M4 Pro, or the chip that powers the Apple MacBook Pro. The idea behind it is to provide leadership CPU and graphics performance from a single package, minimizing the PCB footprint from having a discrete GPU. In stark contrast, the Intel Core Ultra 200V "Lunar Lake," is designed more to go against the Apple M4, or the chips that power the latest MacBook Air but not quite the MacBook Pro. What sets "Strix Halo" functionally apart from "Lunar Lake" or even the M4 Pro, is that the AMD chip doesn't have memory-on-package (MoP), it relies on discrete LPDDR5X memory chips.
The "Strix Halo" processor is "Fire Range" on steroids. There are one or two "Zen 5" CCDs, for up to a 16-core/32-thread core configuration. Each of these "Zen 5" cores are unlike the ones on "Strix Point," in that they feature a fully unlocked AVX512 hardware pipeline (512-bit FP). The CCD shares a lavish 32 MB of L3 cache among 8 "Zen 5" cores. This is hardly the star attraction. Unlike "Fire Range," which features the small 6 nm client I/O die from "Granite Ridge," The new "Strix Halo" features a massive SoC die built on the 5 nm EUV foundry node. This packs the star attraction of the processor, it's oversized iGPU that has a massive 40 compute units (2,560 stream processors).
The iGPU of "Strix Point" is based on the RDNA 3.5 graphics architecture. RDNA 3.5 has numerous improvements over RDNA 3, including optimization for the LPDDR5X memory type. The iGPU has 40 CU, which works out to 2,560 stream processors 80 AI accelerators, and 40 Ray accelerators, besides 160 TMUs, and 64 ROPs. AMD runs the iGPU engine clock at speeds of up to 2.90 GHz. AMD is looking to achieve gaming performance similar to that of the desktop Radeon RX 6750 XT with this iGPU.
Besides the oversized iGPU, the SoC die features the same Ryzen AI XDNA 2 NPU found on "Strix Point," which belts out 50 AI TOPS, and meets Microsoft Copilot+ AI PC logo requirements. This NPU, combined with the AI acceleration capabilities of the iGPU, makes the "Strix Halo" the first PC processor capable of running a 70B parameter LLM entirely on one package.
To keep the up to 16 "Zen 5" CPU cores, the large iGPU with 40 CU, and the 50 TOPS NPU fed with sufficient memory bandwidth, AMD has given the "Strix Halo" a quad-channel (256-bit) LPDDR5X-6400 memory interface. This should give the processor a memory bandwidth of 256-bit. The iGPU talks to the system over the core switching fabric (Infinity Fabric) of the SoC die.
AMD is launching as many as 7 processor models based on "Strix Halo," three of these are for the consumer segment, and four of these are AMD PRO branded models for the commercial notebook segment.
The lineup is led by the Ryzen AI Max+ 395, with a 16-core/32-thread CPU that runs at 3.00 GHz base, 5.10 GHz maximum boost. The iGPU is the maxed out variant labeled Radeon 8060S, with all 40 CU being enabled, and a 2.90 GHz iGPU engine clock. The Ryzen AI Max+ PRO 395 has essentially the same specs, but with the added AMD PRO commercial feature-set. Both chips come with a configurable TDP range of 45 W to 120 W.
Next up are the Ryzen AI Max 390 and Ryzen AI Max PRO 390. These feature a 12-core/24-thread CPU (all "Zen 5" cores), with a 3.20 GHz base frequency, and 5.00 GHz max boost. The iGPU is cut down, is branded Radeon 8050S, and comes with 32 CU (2,048 stream processors, 64 AI accelerators, 32 Ray accelerators, 128 TMU, 64 ROPs), and a 2.80 GHz engine clock. The cTDP is unchanged at 45 W to 120 W.
A step-down from these are the Ryzen AI Max 385 and Ryzen AI Max PRO 385. These feature a single "Zen 5" CCD for an 8-core/16-thread CPU configuration, which ticks at 3.60 GHz base frequency, and 5.00 GHz maximum boost. The iGPU is the same Radeon 8050S as the Ryzen AI Max 390, with 32 CU.
At the very entry level is the Ryzen AI Max PRO 380. There is no consumer variant of this chip. It comes with a 6-core/12-thread CPU config (all "Zen 5" cores), 3.60 GHz base and 4.90 GHz max boost; and at truncated 16 MB of shared L3 cache on the CCD. The iGPU is heavily truncated, too. It is branded Radeon 8040S, and comes with just 16 CU (1,024 stream processors, 32 AI accelerators, 16 Ray accelerators, 64 TMU, 32 ROPs).
Below are AMD's performance claims for the Ryzen AI Max series.
AMD Ryzen AI Max 300 series processors should begin appearing as ultraportable gaming notebooks such as the ASUS ROG Flow Z13; the HP ZBook Ultra G1a mobile workstation, or the HP Z2 Mini G1a mini-PC workstation, in the first half of 2025.
The "Strix Halo" processor is "Fire Range" on steroids. There are one or two "Zen 5" CCDs, for up to a 16-core/32-thread core configuration. Each of these "Zen 5" cores are unlike the ones on "Strix Point," in that they feature a fully unlocked AVX512 hardware pipeline (512-bit FP). The CCD shares a lavish 32 MB of L3 cache among 8 "Zen 5" cores. This is hardly the star attraction. Unlike "Fire Range," which features the small 6 nm client I/O die from "Granite Ridge," The new "Strix Halo" features a massive SoC die built on the 5 nm EUV foundry node. This packs the star attraction of the processor, it's oversized iGPU that has a massive 40 compute units (2,560 stream processors).
Besides the oversized iGPU, the SoC die features the same Ryzen AI XDNA 2 NPU found on "Strix Point," which belts out 50 AI TOPS, and meets Microsoft Copilot+ AI PC logo requirements. This NPU, combined with the AI acceleration capabilities of the iGPU, makes the "Strix Halo" the first PC processor capable of running a 70B parameter LLM entirely on one package.
To keep the up to 16 "Zen 5" CPU cores, the large iGPU with 40 CU, and the 50 TOPS NPU fed with sufficient memory bandwidth, AMD has given the "Strix Halo" a quad-channel (256-bit) LPDDR5X-6400 memory interface. This should give the processor a memory bandwidth of 256-bit. The iGPU talks to the system over the core switching fabric (Infinity Fabric) of the SoC die.
AMD is launching as many as 7 processor models based on "Strix Halo," three of these are for the consumer segment, and four of these are AMD PRO branded models for the commercial notebook segment.
Next up are the Ryzen AI Max 390 and Ryzen AI Max PRO 390. These feature a 12-core/24-thread CPU (all "Zen 5" cores), with a 3.20 GHz base frequency, and 5.00 GHz max boost. The iGPU is cut down, is branded Radeon 8050S, and comes with 32 CU (2,048 stream processors, 64 AI accelerators, 32 Ray accelerators, 128 TMU, 64 ROPs), and a 2.80 GHz engine clock. The cTDP is unchanged at 45 W to 120 W.
A step-down from these are the Ryzen AI Max 385 and Ryzen AI Max PRO 385. These feature a single "Zen 5" CCD for an 8-core/16-thread CPU configuration, which ticks at 3.60 GHz base frequency, and 5.00 GHz maximum boost. The iGPU is the same Radeon 8050S as the Ryzen AI Max 390, with 32 CU.
At the very entry level is the Ryzen AI Max PRO 380. There is no consumer variant of this chip. It comes with a 6-core/12-thread CPU config (all "Zen 5" cores), 3.60 GHz base and 4.90 GHz max boost; and at truncated 16 MB of shared L3 cache on the CCD. The iGPU is heavily truncated, too. It is branded Radeon 8040S, and comes with just 16 CU (1,024 stream processors, 32 AI accelerators, 16 Ray accelerators, 64 TMU, 32 ROPs).
Below are AMD's performance claims for the Ryzen AI Max series.
26 Comments on AMD Debuts Ryzen AI Max Series "Strix Halo" SoC: up to 16 "Zen 5" cores, Massive iGPU
They said they left their CPU at 55W, against a 30W CPU that's already a stupid model - that i9 has a 30W TDP vs 17W of the other LNL CPUs to achieve a 100MHz higher boost clock. Let's not forget it's a 16c/32t CPU vs a 4p4e one.
Their comparison against the M4 Pro was also awful, they did not disclose the TDP used, and compared it to another 30W CPU, with not so impressive MT gains (apart from v-ray). 4x the power (assuming they used the 120W cTDP) for minor gains, impressive.
Adding to that, the comparison against the 4090 was also extremely non-sense. They used a 70B model that requires more than 35GB of memory on a 24GB GPU, which has to offload many layers to the CPU. And no comparisons against no other GPU whatsoever in any other scenarios.Some stuff seems to be wrong here. AMD mentioned a bandwidth of 256GB/s, so that would mean they're using LPDDR5X-8000 with their 256-bit bus.
Consumers, tech reporters and companies that are going to use these parts need presentations with pretty numbers. And that's what companies are giving them.
We here will wait for reviews no matter if these charts are true or exaggerated or misleading, we will wait for reviews. These presentation charts are NOT for us.
But in this specific case, it doesn't even matter if those graphs are true or not, they're comparing products that have nothing to do with one another. It'd be the same as comparing an N100 to a 7950x, it just doesn't make sense.
www.notebookcheck.net/Small-powerhouse-with-Strix-Halo-HP-ZBook-Ultra-14-G1a-launches-with-Ryzen-AI-Max-Pro.942121.0.html
Fan part. When the first Ryzen series was out, the Ryzen 1000, intel fans where comparing one AMD model to 5 Intel models to make it look like it was losing in all 5 categories!!!
Yeah, weird how this product is underclocking the memory a bit. Could it be strix halo's fault, or just an HP thing?I didn't deny that, and people also complained about those Intel graphs as well.