AMD "Ryzen 9000G" Desktop APU Series Tipped For Q4 2025 Launch

[T0@st]

Successors to AMD's current-generation lineup of Ryzen 8000G desktop APUs are reportedly in the pipeline—according to the latest HXL/9550pro predictive declaration, finalized units could arrive at retail later this year. They propose that an "AMD AM5 New APU" family could arrive alongside an unannounced MSI Unify-X enthusiast-grade motherboard design, within the final quarter of 2025. Press interpretations of this inside track information point to possible upcoming "Ryzen 9000G" processors, utilizing Team Red's Zen 5 and RDNA 3.5 technologies. This potent combination already exists, albeit in mobile form—namely within Team Red's stable of Ryzen AI "Strix Halo, Strix Point," and "Krackan Point" APUs.

Industry experts opine that AMD will most likely deploy high-end "Strix Point" silicon to desktop, or more fancifully: "Gorgon Point." The latter codename turned up via leaks last week. Around early 2024, we witnessed Team Red's transfer of "Phoenix"—from original mobile formats—to their AM5 desktop platform. TechPowerUp's W1zzard evaluated the Ryzen 5 8500G "Phoenix 2" APU last summer; this plucky budget-friendly model sports Zen 4 and Zen 4c cores. Theoretically a flagship "Ryzen 9000G" SKU could emerge with twelve processor cores (4x "Zen 5" + 8x "Zen 5c"), a Radeon 890M iGPU, and an XDNA 2 NPU.



View at TechPowerUp Main Site | Source

 

[Mrgravia]

Good to hear. I'd love to build some APU setups for some of my family who don't need discreet GPU, but want to play some light gaming.

 

[Franzen4Real]

Good to hear. I'd love to build some APU setups for some of my family who don't need discreet GPU, but want to play some light gaming.

I have an old DanCase A4 v.2 that has been sitting in the closet for quite a while that would be great for use with one of these APU's. The GPU side of the case could then be used to mount a small AIO, and still have the PCIe riser cable available for an add in card of some sort.

 

[Wirko]

Let's see how much Microsoft screws up the Windows scheduler. Two types of cores with different performance levels, each of them with SMT. That will be a rough ride.

 

[Mrgravia]

Also makes the Jonsbo N10 a more interesting purchase. Strong APU and lots of room for added storage would be quite useful.

I'm also eyeing some 3d printed cases, like the Modcase Evolution's APU case.

 

[Timbaloo]

Desktop Strix Point seems so incredibly unsexy considering that 8700G has 8 full cores and was already completely memory bandwidth bound GPU wise.

 

[blinnbanir]

I was actually shocked by the performance of the 8600G. I was also blown away by the 9900X3D. AMD have so many compelling CPUs and APUs.

Desktop Strix Point seems so incredibly unsexy considering that 8700G has 8 full cores and was already completely memory bandwidth bound GPU wise.

40 Compute Units vs 10

 

[Timbaloo]

40 Compute Units vs 10

1. You're referring to Strix Halo with the 40 CUs, not Strix Point (890M, 16 CUs)
2. 8700G has a 780M which has 12 CUs
3. As written before, if the 780M was already memory bandwidth limited (and significantly so) then neither 16 CUs of the 890M (Strix Point) nor the 40CUs (Strix Halo) would gain any relevant performance in GPU usage scenarios.

 

[Wirko]

Desktop Strix Point seems so incredibly unsexy considering that 8700G has 8 full cores and was already completely memory bandwidth bound GPU wise.

A lot will depend on the chip's memory OC abilities. CUDIMM support must also be there of course.

 

[ymdhis]

Let's see how much Microsoft screws up the Windows scheduler. Two types of cores with different performance levels, each of them with SMT. That will be a rough ride.

It's not like the P/E core nonsense with Intel. They only differ in cache and clockspeed, so handling things should be as simple as the "preferred" cores in older Ryzens.

I'm more worried about how many PCIE lanes will they have, since they keep reducing that for APUs. I run a machine with iGPU for gaming and bifurcation on the primary PCIE slot to add extra NVME storage, so limiting the PCIE lanes would be a problem.

 

[JustBenching]

It's not like the P/E core nonsense with Intel.

Intel has it's own hw scheduler

 

[Wirko]

Intel has it's own hw scheduler

Not really, it's more like a performance counter/analyser that the software scheduler can get useful info from. AMD may or may not have something similar, but they prefer to keep secrets for themselves.

I'm more worried about how many PCIE lanes will they have, since they keep reducing that for APUs. I run a machine with iGPU for gaming and bifurcation on the primary PCIE slot to add extra NVME storage, so limiting the PCIE lanes would be a problem.

You can infer some information from Ryzen AI diagrams in the TPUCPUDB. Some PCIe lanes are shared with video, and there's 4 lanes less for the chipset link.

 

[igormp]

Intel has it's own hw scheduler

As said above, it's not really a scheduler, it provides more information so the actual OS scheduler can work better.

AMD may or may not have something similar, but they prefer to keep secrets for themselves.

At least going from the activity with linux patches, AMD doesn't have something like that, they just have static tables with information from individual cores (such as if a core is a high frequency, high cache or low power one), which the scheduler can make use of assigns stuff depending on the power mode.

 

[medi01]

Cool stuff, but with OEMs being too afraid to use it in actual laptops, rather frustrating.

 

[sudothelinuxwizard]

Theoretical "9700G" based on r7 350 would be really darn power efficient. I'm excited for the prospect of a "9900G" too, assuming the price is right (spoiler: it probably won't be)

 

[Deleted member 245263]

I was hoping for this Zen/Zen-C hybrid. If they release it at a reasonable price, I'll probably buy this CPU and one more AM5 board.