CES 2025: 3D V-Cache and AI on the new Ryzens

CES 2025: 3D V-Cache and AI on the new Ryzens

AMD has added to its current product portfolio, and within this they also presented quite interesting designs.

At this year’s CES, AMD focused on supplementing the existing product portfolio, so many new desktop and mobile processors were presented. In terms of desktop solutions, the Ryzen 9000 series has been available for quite some time, and we have already prepared a test for each model, including the Ryzen 7 9800X3D version aimed at gamers. The latter model has been expanded with two companions, as the development including 12- and 16-core 3D V-Cache chips has also arrived, the parameters of which are detailed in the table below:

AMD’s new Ryzen 3D V-Cache processor generation
Type Core/Turbo Clock SMT L2 cache L3 cache Radeon
seeds
number		 Consumption
(TDP)
9 9950X3D (16 mag) 4,3/5,7 GHz van 16 x 1 MB 128 MB 128 170 W
9 9900X3D (12 mag) 4,4/5,5 GHz van 12 x 1 MB 128 MB 128 120 W

(+)

Socket AM5 socket processors will arrive in the first quarter of the year and are also compatible with existing motherboards after a BIOS update. Their performance will be particularly high due to the extra cache, and in this area AMD competes mostly with the previous generation Ryzen 7000X3D processors, which the newer models can obviously beat.

Rowing on the mobile line, AMD will offer the HX series for a higher performance level, these are actually versions of the desktop designs with a mobile casing, codenamed Fire Range, and their parameters are detailed in the table below:

AMD’s new Ryzen HX mobile processor generation
Type Core/Turbo Clock SMT L2 cache L3 cache Radeon
seeds
number		 Consumption
province
(TDP/cTDP)
9 9955HX3D (16 mag) 2,5/5,4 GHz van 16 x 1 MB 128 MB 128 54/55-75 W
9 9955HX (16 mags) 2,5/5,4 GHz van 16 x 1 MB 64 MB 128 54/55-75 W
9 9850HX (12 mag) 3/5,2 GHz van 12 x 1 MB 64 MB 128 54/45-75 W


(+)

Notebooks based on Fire Range designs and intended primarily for mobile workstations will arrive in the first half of the year.

The mass market will be targeted more by the chip codenamed Krackan Point, which can be understood as a skinned version of the already available Strix Point. The advantage of this is that it can be produced more cheaply, since the chip itself is smaller, but in return it does not have as much performance. Specifically, Krackan Point offers 4 Zen 5 and 4 Zen 5c cores, as well as an 8 CU, RDNA 3.5 IGP. The parameters of the specific models are detailed in the table below:

AMD Crackan Point – Ryzen AI 300 mobile generation with Zen 5 and 5c cores
Type Clock signal / Turbo clock signal SMT L2 cache L3 cache Radeon cores
real / effective number		 IGP Max. hour mark Consumption range (TDP/cTDP)
7 350 (8 mag) 2/5 GHz van 8 x 1 MB 16 MB 512/1024 3 GHz 28/15-54 W
7 340 (6 mag) 2/4,8 GHz van 6 x 1 MB 16 MB 256/512 2,9 GHz 28/15-54 W
AMD Crackan Point – Ryzen AI PRO 300 mobile generation with Zen 5 and 5c cores
Type Clock signal / Turbo clock signal SMT L2 cache L3 cache Radeon cores
real / effective number		 IGP Max. hour mark Consumption range (TDP/cTDP)
7 PRO 350 (8 mag) 2/5 GHz van 8 x 1 MB 16 MB 512/1024 3 GHz 28/15-54 W
7 PRO 340 (6 mag) 2/4,8 GHz van 6 x 1 MB 16 MB 256/512 2,9 GHz 28/15-54 W

Business and end-user notebooks will arrive for each of the new models in the first quarter.

We have come to one of the most interesting developments, which is arguably the Strix Halo design, which was primarily designed to counter Apple’s specifically beefy mobile series. It offers 16 Zen 5 cores in a more serious package and provides a 256-bit memory bus for LPDDR5X memories. The AI ​​engine offers a speed of over 50 TOPS, which is typical for the entire Ryzen AI 300 series, and the real interesting part is the RDNA 3.5 IGP, which hides 20 multiprocessors, and also includes 32 MB of Infinity Cache. The parameters of the arriving models are detailed in the table below:

AMD Strix Halo – Ryzen AI 300 Max mobile generation with Zen 5 cores
Type Clock signal / Turbo clock signal SMT L2 cache L3 cache Radeon cores
real / effective number		 IGP Max. hour mark Consumption range (TDP/cTDP)
Max+ 395 (16 mag) 3/5,1 GHz van 16 x 1 MB 64 MB 2560/5120 2,9 GHz 55/45-120 W
Max 390 (12 mag) 3,2/5 GHz van 12 x 1 MB 64 MB 2048/4096 2,8 GHz 55/45-120 W
Max 385 (8 mag) 3,6/5 GHz van 8 x 1 MB 32 MB 2048/4096 2,8 GHz 55/45-120 W
AMD Strix Halo – Ryzen AI 300 PRO Max mobile generation with Zen 5 cores
Type Clock signal / Turbo clock signal SMT L2 cache L3 cache Radeon cores
real / effective number		 IGP Max. hour mark Consumption range (TDP/cTDP)
Max+ PRO 395 (16 mag) 3/5,1 GHz van 16 x 1 MB 64 MB 2560/5120 2,9 GHz 55/45-120 W
Max PRO 390 (12 mag) 3,2/5 GHz van 12 x 1 MB 64 MB 2048/4096 2,8 GHz 55/45-120 W
Max PRO 385 (8 mag) 3,6/5 GHz van 8 x 1 MB 32 MB 2048/4096 2,8 GHz 55/45-120 W
Max PRO 380 (6 mag) 3,6/4,9 GHz van 6 x 1 MB 16 MB 1024/2048 2,8 GHz 55/45-120 W

Strix Halo is a very unique development, as it has never been similar among Microsoft Windows notebooks. According to AMD’s measurements, the top-of-the-line model easily beats the Apple M4 Pro options in rendering workflows, and it’s the first desktop platform capable of running Llama 3.1 70B-Q4 LLM locally at appreciable speeds, thanks to its 128GB memory can be associated with the processors, and the IGP can receive 96 GB of this.

(+)

In fact, the biggest advantage of the entire platform is that the integrated graphics controller can access much, much more memory than, for example, a dedicated GPU, so it becomes possible to run tasks that are extremely memory intensive. Mobile and desktop solutions for business and end users based on the Ryzen AI Max series will come in the first half of the year.

Finally, AMD is updating the handheld console market as well, as the Ryzen Z2 series is coming. Their parameters can be found in the tables below.

AMD Strix Point – Ryzen Z2 mobile generation with Zen 5 and Zen 5c cores
Type Clock signal / Turbo clock signal SMT L2 cache L3 cache Radeon cores
real / effective number		 Consumption range (TDP/cTDP)
Z2 Extreme (8 mag) 2/5 GHz van 8 x 1 MB 16 MB 1024/2048 28/15-35 W
AMD Hawk Point – Ryzen Z2 mobile generation with Zen 4 cores
Type Clock signal / Turbo clock signal SMT L2 cache L3 cache Radeon cores
real / effective number		 Consumption range (TDP/cTDP)
Z2 (4 mags) 3,3/5,1 GHz van 4 x 1 MB 16 MB 768/1536 28/15-30 W
AMD Rembrandt – Ryzen Z2 mobile generation with Zen 3+ cores
Type Clock / Turbo clock SMT L2 cache L3 cache Radeon cores Consumption range (TDP/cTDP)
Z2 Go (4 mag) 3/4,3 GHz van 4 x 512 kB 8 MB 768 28/15-30 W


(+)

Here you can see that the Z2 series hides three different chips and also has a different IGP architecture. The basic Z2 offers RDNA 3, the Z2 Extreme RDNA 3.5, and the Z2 Go RDNA 2 design. Due to the specific target market, the NPU is inactive in all models. Z2 models will begin shipping in the first quarter.

Source: prohardver.hu