The distinction now has an important practical consequence. AMD’s on-stage HP ZBook demonstration remains an unannounced prototype with no confirmed retail specification, price, or availability. Lenovo, however, has announced a concrete Gorgon Halo desktop: the ThinkCentre X Ultra. Its November 2026 timing is later than AMD’s earlier Q3 expectation for HP and Lenovo systems, but it supplies the first clear example of how an OEM’s actual configuration can differ from a processor family’s maximum capability.
RTX Spark and Gorgon Halo are competing approaches
NVIDIA says RTX Spark Windows PCs will arrive in October 2026. Its published laptop superchip options combine Grace CPUs with Blackwell RTX graphics. One configuration has a 20-core Grace CPU and a 6,144-core Blackwell RTX GPU; another has an 18-core Grace CPU and a 5,120-core GPU. NVIDIA also advertises Windows 11, DLSS 5, and up to 128GB of LPDDR5X memory on the higher configuration.
Those specifications make RTX Spark an NVIDIA-branded Windows platform centered on RTX graphics and NVIDIA’s software ecosystem. They do not yet settle retail device pricing or independently tested gaming and AI performance.
AMD’s Ryzen AI Max PRO 400 series addresses related local-AI and workstation needs, but it is not an RTX Spark platform by another name. AMD positions the processors for commercial PCs, mobile workstations, and compact desktop systems. The company says they combine Zen 5 CPU cores, RDNA 3.5 graphics, and an XDNA 2 neural processing unit.
For the flagship Ryzen AI Max+ PRO 495, AMD lists 16 CPU cores, 32 threads, 40 Radeon graphics compute units, 55 NPU TOPS, and up to 192GB of system memory. The overlap with RTX Spark is real: both seek to put unusually substantial compute and memory resources into personal Windows systems for local AI work. But their branding, architectures, graphics stacks, and eventual device designs differ. Buyers should compare specific systems and the applications they intend to run rather than assume a component-for-component equivalence.
Lenovo makes the memory distinction concrete
AMD’s headline platform figure is support for up to 192GB of unified system memory, with up to 160GB allocatable as graphics memory. A large shared pool can matter for local AI because a model, its context, development tools, and other Windows applications can coexist in system memory rather than depend solely on transfers between separate CPU and GPU memory pools.
That processor ceiling has never meant that every Ryzen AI Max PRO 400 system would ship with 192GB. OEMs choose the final memory options, and the announced Lenovo ThinkCentre X Ultra demonstrates why that distinction matters.
Lenovo has specified the ThinkCentre X Ultra with up to a Ryzen AI Max+ PRO 495, Radeon 8065S graphics, and up to 128GB of unified memory. Lenovo also says up to 96GB can be allocated to graphics memory. In other words, the desktop can offer a very large shared-memory configuration while still remaining below the processor’s 192GB platform maximum and AMD’s 160GB maximum graphics-memory allocation.
This is not a contradiction. It is the normal difference between what a processor supports and what a manufacturer chooses to build, validate, price, and sell. It also means that a shopper cannot infer a system’s memory capacity or graphics-memory allocation from “Ryzen AI Max PRO 400” alone.
Lenovo expects the ThinkCentre X Ultra to start at €3,100 and become available in November 2026. That is the currently announced schedule for this particular Lenovo system. It supersedes any impression that Lenovo’s Gorgon Halo hardware is simply due in the third quarter.
AMD’s May 20, 2026 announcement had said HP and Lenovo systems were expected in Q3 2026. That should be understood as dated vendor guidance, not a current shipping timetable for the ThinkCentre X Ultra. The Lenovo product announcement provides a more specific, later expectation for its announced desktop. It does not establish the timing of an HP system.
HP’s ZBook remains a prototype, not a retail promise
At IFA, AMD executive Jack Huynh showed early HP ZBook models reportedly carrying the “Sundae” codename and using a Ryzen AI Max 400-series processor. That is meaningful evidence that HP hardware exists in prototype form, but it is not a finalized product announcement.
HP has not confirmed a retail model name, final configuration, price, or availability for the ZBook prototype. Reporting described as much as 190GB of unified memory in the system shown, but that exact configuration remains unconfirmed by an HP product specification. AMD’s published maximum for the Ryzen AI Max+ PRO 495 is 192GB, not 190GB.
There is also limited public evidence that a Gorgon Halo HP configuration has been tested before retail launch. A pre-release Geekbench result reported in late July associated an HP ZBook Ultra G3a with a Ryzen AI Max+ PRO 495 and 96GB of memory. That sighting suggests a real system configuration was in circulation before IFA, but it cannot answer the questions that matter most to workstation buyers: sustained performance, cooling behavior, fan noise, battery life, graphics performance, application compatibility, or final pricing.
The practical conclusion is straightforward. Lenovo now gives buyers an announced product to assess, with a stated maximum of 128GB unified memory and an expected November arrival. HP’s “Sundae” remains a promising but unconfirmed prototype. Neither one should be represented as a 192GB retail laptop simply because the processor family supports that capacity.
The graphics names are different, though the broad design is close
The Ryzen AI Max+ PRO 495 uses Radeon 8065S graphics with 40 compute units. The earlier Ryzen AI Max+ 395, from the preceding Strix Halo generation, uses Radeon 8060S graphics, also with 40 compute units and RDNA 3.5 architecture.
The similar names and matching compute-unit count make confusion understandable, but the GPUs are not identically named. Nor do those headline details alone prove a large real-world graphics gain. Clock speeds, memory capacity, allocated graphics memory, firmware, cooling, and the application itself can all shape the result.
That is also why Acer’s IFA hardware should not be read as announced Gorgon Halo hardware. Acer specifies its Aspire G 3D 16 with a Ryzen AI Max+ 395, Radeon 8060S graphics, and up to 128GB of LPDDR5X memory. That is a Strix Halo configuration rather than an announced Ryzen AI Max PRO 400 machine. Acer separately described an NVIDIA design as an SFF RTX Spark concept, underlining that its AMD and NVIDIA systems are separate products.
For Windows buyers, broad labels such as “Ryzen AI Max,” “AI Halo,” “Gorgon Halo,” and “RTX Spark” are not enough. Check the full processor name, graphics designation, maximum installed memory, graphics-memory allocation, cooling design, ports, and device form factor.
Project Zenith gives high-memory Windows PCs a clearer purpose
Microsoft’s Project Zenith is the most consequential Windows-specific signal around this category. Announced on September 4, Project Zenith is described as a ready-to-code, distraction-free Windows experience for developer-class devices with at least 64GB of unified memory and 250GB/s of memory bandwidth. Microsoft says it will first become available with AMD Ryzen AI Halo, followed by OEM and silicon-partner devices.
That creates a more tangible software rationale for these high-memory designs than generic AI-PC marketing. Microsoft is setting requirements well above those of ordinary consumer laptops. For developers working with local AI-assisted tools, large codebases, or model-driven workflows, unified memory and bandwidth may be more important constraints than a short NPU demonstration suggests.
Yet the announcement does not identify every supported model, SKU, market, or release date. The stated memory and bandwidth levels should be regarded as entry requirements for Project Zenith, not a guarantee that all qualifying systems will provide the same experience. Buyers interested in Zenith should verify support for the exact PC they are considering.
A compact AI desktop is not a Halo Station
IFA also placed several dramatically different hardware scales under the local-AI umbrella. Ryzen AI Max PRO 400 systems and RTX Spark laptops or compact desktops are intended for desks, offices, and mobile-workstation use. Their success will depend on familiar Windows trade-offs: price, chassis size, power draw, cooling, display options, upgradeability, drivers, and whether important software uses the available CPU, GPU, or NPU effectively.
AMD’s Threadripper Halo Station belongs in a different category. The system shown at IFA was reported to include a 96-core Threadripper PRO processor, two liquid-cooled Instinct MI350P accelerators, 2TB of DDR5 memory, and 288GB of HBM3E in its displayed dual-accelerator configuration. AMD reportedly described a path to four accelerators and 576GB of HBM3E, while claiming the workstation can run models exceeding one trillion parameters.
This is specialized infrastructure for organizations and professionals with substantial budgets, power, cooling, and deployment requirements, not a comparison point for a compact Lenovo desktop or an HP notebook. AMD’s trillion-parameter statement remains a vendor claim that needs details on models, quantization, throughput, and independent testing before it can be treated as a broadly useful performance conclusion.
What Windows buyers should verify next
The evidence supports measured optimism. Ryzen AI Max PRO 400 could be compelling where a large unified-memory pool is more valuable than the usual thin-laptop priorities. Lenovo’s ThinkCentre X Ultra turns that possibility into a named product with a specific, if later, November 2026 expectation. But it also reinforces that OEM configurations, not just processor limits, determine what users can actually buy.
Before committing to any system in this category, look for four things:
- The final OEM specification. Confirm the exact processor, installed memory, graphics-memory allocation, storage, ports, wireless hardware, warranty, and regional price. A 192GB processor ceiling does not make a given machine a 192GB system.
- Independent sustained testing. Long AI-inference, compilation, rendering, gaming, thermal, acoustic, and battery tests matter more than isolated pre-release benchmarks.
- Application-level compatibility. Determine whether key tools favor AMD graphics, NVIDIA RTX hardware, an NPU, or a mixture of accelerators. The best paper specification has little value if a critical workflow is poorly supported.
- A specific Windows software plan. Project Zenith makes the 64GB-plus developer category more concrete, but users should confirm availability on the exact product rather than assuming it will be present at launch.
AMD’s IFA presentation showed that local AI hardware is moving beyond conventional AI-PC messaging toward devices with workstation-like memory budgets. Lenovo’s announced ThinkCentre X Ultra is now the clearest Gorgon Halo product example, even though its 128GB maximum is well below the processor family’s 192GB ceiling and its November availability is later than AMD’s original Q3 expectation. HP’s ZBook “Sundae,” by contrast, should remain classified as an unannounced prototype until HP provides final product details. The next test is not branding: it is whether finished systems deliver the performance, software support, thermals, and value that their ambitious specifications imply.