AMD Ryzen Z3 Handheld Leak: 4 Plus 2 Cores and 12 RDNA4m CUs [2026]

By Indie Kings | October 1, 2026

Updated October 1, 2026: The reported Ryzen Z3 pairs 4 Zen 6 cores with 2 Zen 5 low power cores and 12 RDNA4m compute units in a Medusa Point derivative aimed at 15W for Z3 and 25W for Z3 Extreme per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. This remains a reported leak with no AMD confirmation and several details such as final core counts and cut down models are still unclear [NEEDS VERIFICATION].

Image: AMD header art. Credit: AMD.

The Leak

The Ryzen Z3 handheld leak arrived through secondhand reporting rather than through an AMD announcement per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. VideoCardz attributes the core configuration talk to Gotou_3rd and CaptainMcShotgun, with added context attributed to Kepler. That chain matters because readers should weigh a leak relay differently from a specification sheet. No AMD slide, press release, or product page backs the Z3 configuration at the time of this writing. The responsible reading is that this is a reported plan or reported engineering direction, not a confirmed product.

The headline configuration is a semi custom Medusa Point derivative with 4 Zen 6 cores plus 2 Zen 5 low power cores for a total of 6 cores and 12 threads, paired with 12 RDNA4m compute units, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. The same report notes that the configuration may have changed during development. That caveat is important because semi custom chips often shift core counts, clock targets, and enabled units between early planning and final silicon. Readers should treat 6C and 12T as reported, not locked. Any retailer listing or forum post that states it as final is moving faster than the source.

The package detail is unusually specific for an early handheld leak. The report describes an FF6 package at 25 by 25 mm, smaller than the Strix Point FP8 package, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Package size affects board layout, cooling design, and how small a handheld can be while still feeding the chip clean power. A smaller footprint does not automatically mean lower power, but it does give device makers more room for battery, fans, or memory. For handheld makers, every millimeter around the socket region influences thickness and weight [NEEDS VERIFICATION].

Power targets are also reported as split by tier. Z3 is described as optimized around 15W, while Z3 Extreme is described as optimized around 25W, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Those figures read as design points rather than hard limits, because handheld APUs normally operate across a configurable power range set by the device maker [NEEDS VERIFICATION]. A 15W optimization suggests efficiency tuning for longer battery life in a smaller chassis. A 25W optimization suggests higher sustained clocks for a larger performance handheld with stronger cooling.

One open question is whether both Z3 tiers share the same 12 compute unit GPU configuration. The report states that it is unclear whether both share the 12 CU config or whether lower models cut CUs, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. That uncertainty mirrors prior Z series segmentation, where Extreme and non Extreme models sometimes differed in GPU count, CPU clocks, or power limits [NEEDS VERIFICATION]. Until AMD confirms a stack, buyers should not assume that a base Z3 equals a Z3 Extreme with a lower power limit. The difference could be silicon level if AMD fuses off units for yield or segmentation.

The leak also frames Z3 as more customized than the Z1 and Z2 generations, with no old chip rebadges suggested so far, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. That claim will matter to handheld buyers who felt prior generations mixed new silicon with familiar parts [NEEDS VERIFICATION]. A ground up handheld optimization could improve efficiency, driver behavior, and sustained performance in ways that a simple rebadge cannot. At the same time, customization claims are hard to verify from outside without die shots, block diagrams, or AMD statements. Treat stronger customization as reported direction, not proven fact.

For context, the Z series table below places Z1 through Z3 in one view. It is presented as reported leak information combined with previously reported generation details, never as AMD confirmation. The Z1 row reflects Zen 4 with RDNA3 as stated in the assignment brief. Later rows mix reported details with items that still need verification. The second table in this article separates what the leak reports from what AMD has confirmed. Readers who only want the short version can use those two tables, then read the CPU, GPU, and package sections for the tradeoffs behind the numbers.

Attribution discipline stays strict through this article. Every paragraph that states the 4 plus 2 core layout, the 12 compute unit count, the FF6 25 by 25 mm package, the 15W and 25W optimization points, or the comparison against standard Medusa Point cites VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Everything else about performance, battery life, pricing, release timing, device wins, clock speeds, memory support, and process node gets a [NEEDS VERIFICATION] tag. That split lets search engines and AI summaries cite the leak correctly without turning analysis into false fact.

CPU Trade

The reported CPU tradeoff is straightforward on paper. Standard Medusa is described as 4 Zen 6 plus 4 Zen 6c plus 2 low power cores with 8 compute units, while Z3 is described as 4 Zen 6 plus 2 low power cores with 12 compute units, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. In plain terms, Z3 drops 4 Zen 6c cores in exchange for more GPU silicon and power. That is a handheld first tradeoff because games usually respond more to GPU headroom and memory bandwidth than to extra efficiency cores [NEEDS VERIFICATION].

Zen 6 is expected to carry the single thread and gaming critical work in this layout [NEEDS VERIFICATION]. Four full Zen 6 cores give handheld makers a modern performance cluster for game threads, operating system work, and background tasks. The report does not provide clock speeds, cache sizes, or power sharing behavior between CPU and GPU, so direct comparisons to desktop Zen 6 remain premature. What can be said is that keeping four full performance cores preserves responsiveness while removing silicon that a thin handheld might struggle to feed. That choice favors sustained frame pacing over multithreaded benchmark scores [NEEDS VERIFICATION].

The 2 low power cores use Zen 5 architecture rather than matching Zen 6, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Mixed generation low power islands are plausible when a semi custom program reuses proven efficiency blocks while updating the main performance cores [NEEDS VERIFICATION]. Those cores likely handle light background work, sleep behavior, and low intensity tasks where waking a full Zen 6 core would waste energy. The practical benefit in a handheld is standby efficiency and smoother idle behavior rather than higher game frame rates [NEEDS VERIFICATION]. Actual scheduling behavior will depend on firmware and operating system support.

Removing 4 Zen 6c cores reduces total multithreaded throughput relative to standard Medusa on paper. That loss matters less in handheld gaming than it would in a laptop used for rendering or compilation [NEEDS VERIFICATION]. Most current PC games scale strongly to a handful of fast threads and then show diminishing returns from additional efficiency cores [NEEDS VERIFICATION]. By dropping those cores, AMD can reclaim die area and power budget for graphics. The leak frames that as a deliberate exchange of CPU silicon and power for GPU silicon and power, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler.

Total thread count is reported as 6 cores and 12 threads, with the warning that the config may have changed, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Twelve threads remain comfortable for modern Windows gaming plus common background software such as launchers, voice chat, and recording tools [NEEDS VERIFICATION]. The tighter limit could show in heavy multitasking, such as gaming while streaming with CPU encoding or running large creative workloads on the same device [NEEDS VERIFICATION]. Handheld buyers who mainly play games should feel less impact than buyers who want one device for play and production.

Power sharing is the hidden part of this CPU tradeoff. A handheld APU lives inside a fixed thermal and battery budget, so CPU and GPU constantly negotiate for watts [NEEDS VERIFICATION]. Fewer CPU cores can reduce the chance that background threads steal power during a GPU heavy scene. It can also simplify voltage and frequency management at 15W, where every watt affects clock stability [NEEDS VERIFICATION]. The reported 15W optimization for Z3 and 25W optimization for Z3 Extreme suggests AMD tuned those sharing curves for two different cooling classes, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler.

Compatibility and software behavior should remain familiar if this leak holds. Zen 6 plus Zen 5 low power cores still present as a standard x86 processor to Windows and game launchers [NEEDS VERIFICATION]. No special game patches should be needed to boot existing PC libraries, although scheduling updates and chipset drivers often improve heterogeneous layouts after launch [NEEDS VERIFICATION]. Emulation, mods, launchers, and anti cheat systems generally key off instruction set and operating system behavior rather than core branding [NEEDS VERIFICATION]. That is why CPU core branding matters less to players than sustained clocks, thermals, and driver maturity.

The bottom line on CPU is focus over breadth. Standard Medusa covers more general compute with 4 Zen 6 plus 4 Zen 6c plus 2 low power cores and 8 compute units, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Z3 narrows the CPU to 4 plus 2 in order to widen graphics to 12 compute units, per the same VideoCardz relay. That is a defensible handheld tradeoff if battery life and frame stability are the goals [NEEDS VERIFICATION]. It would be the wrong tradeoff for a buyer who wants maximum multithreaded CPU rendering in a handheld shell, but that buyer was never the core audience for the Z series [NEEDS VERIFICATION].

GPU Trade

The GPU side is where the leak gets most interesting for players. Z3 is reported with 12 RDNA4m compute units, compared with 8 compute units for standard Medusa, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. That is a 50 percent increase in compute unit count on paper, before accounting for clocks, bandwidth, cache, or power [NEEDS VERIFICATION]. Compute unit count alone does not set frame rates, but a larger GPU gives architects more headroom to sustain resolution and detail settings at fixed handheld power. In a 15W to 25W envelope, extra units can sometimes allow lower clocks per unit while holding the same performance with better efficiency [NEEDS VERIFICATION].

RDNA4m is described as an RDNA 3.5 derivative with RDNA 4 class matrix ALUs and a native FSR 4 support path, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. That description positions RDNA4m as an evolution tuned for handheld efficiency rather than a clean sheet architecture. Matrix ALUs matter because modern upscaling, frame generation, denoising, and AI assisted graphics lean on matrix math throughput [NEEDS VERIFICATION]. A native FSR 4 path matters because vendor native upscaling usually beats generic scaling in both image stability and performance per watt [NEEDS VERIFICATION]. If the matrix claim holds, Z3 could age better than older handheld GPUs as more games use AI assisted rendering.

The report reserves RDNA 5 for the rumored PS6 handheld path rather than Z3, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. That separation helps set expectations. Z3 buyers should not expect a next generation console class feature set in a 15W PC handheld [NEEDS VERIFICATION]. Instead, Z3 looks like a bridge that brings RDNA 4 class AI blocks into a power efficient RDNA 3.5 foundation. Console handheld rumors belong in a separate story with separate sourcing, and this article does not treat them as Z3 confirmation [NEEDS VERIFICATION].

More GPU silicon only helps if memory can feed it. Handheld APUs share system memory between CPU and GPU, so bandwidth, latency, and cache behavior shape real gaming results [NEEDS VERIFICATION]. The leak does not specify memory type, bus width, Infinity Cache size, or clock targets for the 12 compute unit GPU. That missing context prevents honest FPS predictions. A 12 compute unit part starved of bandwidth can behave like a smaller GPU, while the same part with fast memory and good caching can stretch its legs [NEEDS VERIFICATION]. Readers should view CU count as potential, not as a benchmark.

FSR 4 support deserves careful framing. A native support path is not the same as guaranteed FSR 4 quality mode in every game at launch [NEEDS VERIFICATION]. Game integration, driver branches, operating system builds, and per title testing all affect whether an upscaler option appears and how clean it looks [NEEDS VERIFICATION]. The value of RDNA 4 class matrix units is that they give developers and AMD a stronger hardware target for machine learning based upscaling [NEEDS VERIFICATION]. Over a two to three year handheld ownership period, that hardware headroom can matter more than day one benchmark deltas [NEEDS VERIFICATION].

The unclear segmentation question returns here. Whether both Z3 tiers share the 12 CU config or lower models cut CUs is unclear, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. If AMD cuts units for a base model, the performance gap between Z3 and Z3 Extreme could be larger than the 15W versus 25W split suggests [NEEDS VERIFICATION]. If both keep 12 units and differ mainly in power tuning, then the gap will come mostly from clocks, thermals, and device cooling [NEEDS VERIFICATION]. Shoppers should wait for confirmed CU counts per SKU before comparing prices across devices.

Thermal density is another practical limit. Twelve compute units in a small FF6 package can concentrate heat in a smaller area than a larger laptop die [NEEDS VERIFICATION]. Good handheld cooling, liquid metal or high quality thermal interface, and sensible fan curves will decide how long boost clocks hold [NEEDS VERIFICATION]. A well cooled 25W Extreme device could outperform a thin 15W device by more than the wattage ratio implies, because sustained clocks depend on heat removal as well as power input [NEEDS VERIFICATION]. That is why device reviews matter more than chip headlines for handhelds.

In short, the GPU trade is breadth for efficiency. Standard Medusa pairs a wider CPU with 8 compute units, while Z3 narrows the CPU and widens graphics to 12 RDNA4m compute units, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. The RDNA 3.5 base plus RDNA 4 class matrix path suggests AMD wants better upscaling efficiency without jumping Z3 to RDNA 5, per the same VideoCardz relay. Real gains will depend on memory, clocks, cooling, and software support that remain unverified [NEEDS VERIFICATION]. Treat 12 compute units as promising handheld direction, not as a promised frame rate.

Package And Power

The FF6 package at 25 by 25 mm is smaller than the Strix Point FP8 package, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Smaller packages can simplify routing in compact handheld boards and leave more internal volume for batteries, speakers, storage, or cooling [NEEDS VERIFICATION]. They can also reduce motherboard complexity if pinout and power delivery requirements suit thin devices [NEEDS VERIFICATION]. None of that guarantees a thinner retail handheld, because screen, battery, controls, and chassis design also set thickness [NEEDS VERIFICATION]. It does give industrial designers a friendlier starting point.

Power optimization points are reported as around 15W for Z3 and around 25W for Z3 Extreme, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Those points likely reflect where AMD tuned voltage curves and sustained behavior, not strict minimum or maximum limits [NEEDS VERIFICATION]. Most handheld APUs support configurable TDP windows that let device makers pick battery life or performance profiles [NEEDS VERIFICATION]. A 15W tuned part can often run lower for indie games or higher for short boosts, while a 25W tuned part can often scale down for travel [NEEDS VERIFICATION]. Final behavior will vary by device firmware.

The 15W target fits the largest handheld use case, which is efficient 720p to 1080p gaming with upscaling [NEEDS VERIFICATION]. At 15W total package power, the GPU cannot hold high clocks indefinitely without strong cooling, so efficiency features matter more than peak theoretical throughput [NEEDS VERIFICATION]. This is where fewer CPU cores, RDNA 3.5 efficiency, matrix acceleration for upscaling, and careful power sharing can combine into smoother play [NEEDS VERIFICATION]. Battery life estimates remain impossible without battery capacity, screen efficiency, and game load details [NEEDS VERIFICATION]. No hour rating should be inferred from 15W alone.

The 25W Extreme target fits larger performance handhelds and docked play [NEEDS VERIFICATION]. Extra watts allow higher sustained GPU clocks, more stable CPU clocks during open world streaming, and better handling of ray tracing or high refresh displays [NEEDS VERIFICATION]. The cost is heat, noise, and battery drain, which is why Extreme handhelds tend to be heavier with larger fans [NEEDS VERIFICATION]. A 25W optimization does not mean the chip always uses 25W. It means the design expects cooling capable of handling that class of sustained load [NEEDS VERIFICATION]. Device makers will still offer lower power modes.

Board and battery design will translate these chip targets into real products [NEEDS VERIFICATION]. Memory choice is especially important because shared LPDDR bandwidth sets a ceiling on GPU scaling [NEEDS VERIFICATION]. Display choice matters because 1080p at 120 Hz needs far less GPU work than 1600p at 144 Hz [NEEDS VERIFICATION]. Cooling choice matters because a thin vapor chamber behaves differently from dual fan active cooling under long sessions [NEEDS VERIFICATION]. Two handhelds with the same Z3 chip can perform differently if one pairs it with faster memory and better thermals.

The reported versus confirmed table below keeps this section honest. It lists each major Z3 claim, what the leak reports, and whether AMD has confirmed it. As of this writing, the confirmed column is uniformly negative because no AMD confirmation is reported. That table is useful for AI summaries and quick readers who might otherwise mistake a leak relay for a specification sheet. It also shows where future updates should focus once AMD or device makers speak on record.

Regulatory and product timing details are absent from the sourced leak. There is no reported release date, price, or device list in the verified facts for this story. Any article that adds those details is either citing a different source or speculating [NEEDS VERIFICATION]. Handheld chips can appear in leaks many months before devices ship, and configurations can change during that window. The package and power numbers are useful for early analysis, but they are not a launch announcement. Readers should watch AMD statements and handheld maker announcements for confirmation.

What It Means For Handhelds

If the leak holds, Z3 signals stronger handheld specialization than prior Z generations. The report explicitly frames Z3 as more customized than Z1 and Z2, with no old chip rebadges suggested so far, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. For players, stronger customization can mean better performance per watt, cleaner handheld power management, and features chosen for small screens rather than carried over from laptops [NEEDS VERIFICATION]. For device makers, it can mean a cleaner story to sell against prior handhelds that reused familiar silicon [NEEDS VERIFICATION]. Both benefits depend on execution that cannot be judged from a leak alone.

The CPU to GPU rebalance fits how handhelds are actually used. Dropping 4 Zen 6c cores to grow from 8 to 12 compute units trades general multithreaded headroom for gaming graphics, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Most handheld play involves a single game in the foreground with light background tasks [NEEDS VERIFICATION]. That workload favors a capable GPU with efficient upscaling over extra CPU cores that sit idle [NEEDS VERIFICATION]. Content creators who render video on a handheld will prefer more CPU, but they are a minority of handheld buyers [NEEDS VERIFICATION].

FSR 4 readiness could be the longest lasting part of this rumor. RDNA4m is described with RDNA 4 class matrix ALUs and a native FSR 4 support path, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. On a small handheld screen, good upscaling can turn a 540p or 720p internal render into a clean 1080p or 1200p output with major power savings [NEEDS VERIFICATION]. Matrix acceleration helps those techniques run efficiently enough to matter at 15W [NEEDS VERIFICATION]. If developers adopt FSR 4 widely on PC, a Z3 handheld with native hardware support would be better placed than older handheld GPUs that rely on less advanced scaling [NEEDS VERIFICATION].

Segmentation remains the biggest shopping risk. It is unclear whether both Z3 tiers share 12 compute units or lower models cut units, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. A cut down base Z3 with fewer units and a 15W tune could still be efficient, but it would not match a full 12 unit Extreme part [NEEDS VERIFICATION]. Price gaps between tiers will then decide value [NEEDS VERIFICATION]. Until per SKU core counts are confirmed, shoppers should avoid assuming that Z3 branding alone guarantees 12 active compute units. Model numbers need specification sheets, not just family names.

Device diversity will shape outcomes as much as silicon. A 15W Z3 in a light 7 inch handheld with a 1080p screen serves a different buyer than a 25W Z3 Extreme in an 8 inch performance device with a high refresh panel [NEEDS VERIFICATION]. Battery size, fan noise, control feel, display brightness, storage speed, and dock support often decide satisfaction more than a few percent of chip performance [NEEDS VERIFICATION]. The smaller FF6 package gives designers flexibility, but retail success still comes from total device execution [NEEDS VERIFICATION]. Early chip leaks cannot predict which maker will tune firmware best.

Software support will also decide whether Z3 feels like a clean break from Z1 and Z2. Stronger customization claims imply closer work on drivers, power profiles, and handheld specific features, per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. Players should watch for day one driver branches, per game profiles, overlay tools, TDP controls, and frame pacing fixes from both AMD and device makers [NEEDS VERIFICATION]. A great chip with weak software can stutter, while a modest chip with mature software can feel smooth [NEEDS VERIFICATION]. Leak coverage cannot verify software maturity before devices ship.

Buying advice stays conservative while this remains unconfirmed. Do not delay a needed handheld purchase based only on Z3 rumors [NEEDS VERIFICATION]. Do not pay a premium for a device marketed on unverified Z3 specifications [NEEDS VERIFICATION]. Do compare current Z1 and Z2 devices on tested battery life, tested game performance, screen quality, controls, warranty, and price [NEEDS VERIFICATION]. If Z3 is real and as described, its advantages will be clearest in efficiency and upscaling rather than in raw CPU benchmarks [NEEDS VERIFICATION]. Wait for AMD confirmation and independent device reviews before treating it as a buying signal.

What to watch next is straightforward. First, watch for AMD acknowledgement of a Z3 program or Medusa Point handheld derivative [NEEDS VERIFICATION]. Second, watch for confirmation of the 4 plus 2 core layout and whether it changed during development [NEEDS VERIFICATION]. Third, watch for per SKU compute unit counts that resolve the shared versus cut question [NEEDS VERIFICATION]. Fourth, watch for memory, clock, and TDP windows from handheld makers rather than from leakers [NEEDS VERIFICATION]. Fifth, watch for FSR 4 game support lists that show whether the RDNA4m path delivers visible benefits on small screens [NEEDS VERIFICATION]. Each step moves Z3 from reported leak to testable product.

Sources: VideoCardz report by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. All core configuration, package, power, and architecture details in this draft are attributed to that VideoCardz relay. All performance interpretation, buying advice, device design discussion, and future outlook are analysis and carry [NEEDS VERIFICATION] where noted.

FAQ

What is the reported Ryzen Z3 configuration?
Per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler, it is a semi custom Medusa Point derivative with 4 Zen 6 plus 2 Zen 5 low power cores for 6C and 12T plus 12 RDNA4m compute units, although the config may have changed.

What are the reported power targets for Z3 and Z3 Extreme?
Per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler, Z3 is optimized around 15W and Z3 Extreme is optimized around 25W.

Do both Z3 models have 12 compute units?
It is unclear per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler whether both share the 12 CU config or lower models cut CUs.

What is RDNA4m in this leak?
Per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler, RDNA4m is an RDNA 3.5 derivative with RDNA 4 class matrix ALUs and a native FSR 4 support path, while RDNA 5 is reserved per rumors for a PS6 handheld path.

How does Z3 differ from standard Medusa Point?
Per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler, standard Medusa is 4 Zen 6 plus 4 Zen 6c plus 2 low power cores with 8 CUs, so Z3 drops 4 Zen 6c cores for more GPU silicon and power.

Has AMD confirmed Ryzen Z3?
No. This is a reported leak relayed by VideoCardz with no AMD confirmation, and buyers should wait for official specs and device reviews [NEEDS VERIFICATION].

Bottom Line

The reported Ryzen Z3 is a focused handheld tradeoff with 4 Zen 6 plus 2 Zen 5 low power cores and 12 RDNA4m compute units in a smaller FF6 package tuned around 15W and 25W tiers per VideoCardz by WhyCry Oct 1 2026 relaying Gotou_3rd and CaptainMcShotgun and Kepler. It trades 4 Zen 6c cores from standard Medusa for extra graphics, adds RDNA 4 class matrix support with an FSR 4 path without moving to RDNA 5, and promises deeper handheld customization than Z1 and Z2 with no rebadges suggested so far. Sharing of the full 12 CU config across tiers remains unclear, the configuration may have changed, and nothing here carries AMD confirmation. Wait for official specs, per SKU compute unit counts, memory and clock details, and tested device reviews before making a buying decision [NEEDS VERIFICATION].