CPUID has quietly shipped ARM64 builds of CPU-Z and HWMonitor that recognize NVIDIA’s RTX Spark platform, days before the compact AI desktop’s broader retail push. The update, flagged by VideoCardz, Notebookcheck, and Tom’s Hardware, gives reviewers and early buyers a way to verify and monitor the GB10 Grace Blackwell Superchip at the heart of the machine rather than guessing at what their new hardware is actually doing under load.
It is a small software release dressed up as a big signal. CPU-Z and HWMonitor do not ship ARM64 sensor support for hardware that is not close to shipping in volume. Their appearance now, on October 7, 2026, tracks with NVIDIA’s ongoing rollout of RTX Spark systems through partners including ASUS, Dell, HP, Lenovo, MSI, Gigabyte, and Acer, and it arrives at a moment when pricing and configuration options for the platform have shifted noticeably since its original debut a year ago.
What changed in CPU-Z and HWMonitor
The relevant update lands in the ARM64 1.06 packages of both utilities, which are distinct from the standard x86 Windows releases (CPU-Z 3.01 and HWMonitor 1.68) that most desktop users already run. CPU-Z’s ARM64 1.06 build adds improved detection of RTX Spark systems, surfacing the SoC identity, architecture, process node, core configuration, and maximum frequency by core cluster. That is meaningfully different from how CPU-Z reads a conventional x86 chip, because RTX Spark is not a CPU paired with a discrete GPU card. It is a single integrated superchip, and software needs to be taught how to parse that topology correctly.
HWMonitor’s ARM64 1.06 build goes further on the telemetry side, adding RTX Spark CPU and GPU sensor readouts alongside new disk metrics such as IOPS and average latency, plus refreshed interface themes. Together, the two updates mean someone running RTX Spark hardware can now pull live clock speeds, thermal readings, and storage performance numbers instead of relying on NVIDIA’s own system tools alone. CPUID also used the same release cycle to extend support for Qualcomm’s Snapdragon X2 Elite Extreme “EG” variant and other Snapdragon X-series chips, a reminder that the Windows-on-Arm ecosystem is expanding on multiple fronts at once, not just around NVIDIA’s silicon.
What has not shown up yet is a public CPU-Z Validator score or a documented HWiNFO benchmark run tied to a specific RTX Spark unit. Readers should treat any screenshot that lacks a validation link, a BIOS version, and a workload description as unverified. The monitoring tools recognizing the platform is a prerequisite for benchmarking, not a benchmark result in itself.
What NVIDIA RTX Spark actually is
RTX Spark is NVIDIA’s branding for a family of compact, ARM-based AI desktops and laptops built around the GB10 Grace Blackwell Superchip, a single package that fuses a Grace Arm CPU and a Blackwell GPU rather than pairing a conventional x86 processor with a separate graphics card. The design traces back to NVIDIA’s own DGX Spark line, and RTX Spark is effectively the consumer- and workstation-facing extension of that architecture, sold through PC partners instead of only as a boxed NVIDIA appliance.
The chip pairs a 20-core Arm CPU, reportedly co-designed with MediaTek, against a Blackwell GPU built with 48 streaming multiprocessors and 6,144 CUDA cores. NVIDIA rates the combination at up to roughly 1 petaflop of FP4 AI compute. Memory is unified: up to 128GB of LPDDR5X shared between CPU and GPU, linked by NVIDIA’s NVLink-C2C interconnect, with memory bandwidth around 273GB/s. The chip is manufactured on TSMC’s 3nm node. For networking, RTX Spark systems carry integrated ConnectX-7 connectivity rated up to 200Gb/s, and NVIDIA’s design supports clustering multiple units, with reporting describing configurations of up to four devices linked at 200Gb/s per device.
The memory and pricing shift since launch
Pricing has moved a lot since the original DGX Spark debuted. The first 128GB configuration, paired with a 4TB SSD, launched at $3,999 and began shipping around October 15, 2025. A year later, the same 128GB/4TB tier is being reported at roughly $6,950, a substantial jump tied to the broader memory and NAND price surge that has hit the entire PC industry through 2026. NVIDIA and its partners have also introduced a 64GB configuration, reported at approximately $4,999, which keeps the same GB10 chip, the 20-core Arm CPU, ConnectX-7 networking, the CUDA stack, and the DGX OS image, but halves memory and storage. The 64GB tier reportedly retains the same 273GB/s memory bandwidth as the 128GB model, since the bottleneck is architectural, not capacity-driven.
That pricing trajectory matters for anyone comparing RTX Spark against other local-AI hardware right now. A platform that launched as a sub-$4,000 developer box has effectively split into a cheaper, memory-constrained entry tier and a much pricier full-memory tier, which changes who the realistic buyer is. Exact commercial pricing for third-party RTX Spark laptops and desktops still varies by vendor and configuration, and NVIDIA has not published one universal MSRP across every partner SKU.
Why monitoring software support is the real story
It is tempting to read “CPU-Z adds a new chip” as a footnote. For RTX Spark specifically, it is not. RTX Spark is not a GPU upgrade for an existing PC, it is a new class of computer, and new classes of computers tend to surface problems that conventional monitoring tools were never built to catch. An integrated CPU-GPU package with unified memory can throttle, share thermal headroom, and manage power in ways that look nothing like a desktop tower with a CPU socket and a PCIe graphics card bolted in separately.
ARM-native builds of CPU-Z and HWMonitor matter because they avoid routing diagnostic reads through x86 emulation on Windows on Arm, which can distort sensor data or simply fail to expose it. With native support, reviewers and early adopters can verify that firmware and drivers correctly report the GB10’s architecture, confirm core topology and clock behavior under load, and catch thermal throttling or power-management bugs before the hardware reaches a wider retail audience. That is standard practice for any new silicon platform, from Apple’s first Arm Macs to Qualcomm’s Snapdragon X Windows laptops, and NVIDIA’s AI-focused chip is no exception.
There is also a software-ecosystem signal buried in the timing. CPUID typically does not build out sensor support speculatively; it does so when hardware partners or NVIDIA provide access to systems that are functionally close to final silicon. The appearance of RTX Spark support in CPU-Z and HWMonitor ARM64 1.06, arriving alongside Qualcomm Snapdragon X2 Elite Extreme support in the same release, suggests NVIDIA is working to make sure the low-level software stack other tools depend on is in place before OEM partners ship units in volume.
Ports, ecosystem, and OEM partners
NVIDIA has treated RTX Spark as a partner platform from the start rather than a single enclosure it builds and sells directly in large volume. The DGX Spark reference design carries two QSFP-class 200Gb/s networking ports, 10Gb Ethernet, Wi-Fi 7, and Bluetooth 5.4, though a full, authoritative port map across every third-party RTX Spark system does not exist yet, since USB, display, and storage I/O vary by manufacturer.
The vendor list has grown since the original announcement. The strongest current reporting ties Acer, ASUS, Dell, Gigabyte, HP, and MSI to GB10-based systems in the roughly $4,999 range, while earlier reporting also listed Lenovo and Microsoft Surface as fall 2026 launch partners. This site has previously covered a Surface Laptop Ultra leak pairing 128GB of RAM with a 20-core RTX Spark chip, and ASUS in particular has been active in the channel, having already reordered RTX Spark stock as eight OEMs target Q4 demand. The exact configuration and launch timing of a rumored Micro Center retail variant has not been confirmed by named sources.
RTX Spark pricing and specs at a glance
| Configuration | Memory | Storage | Reported price | Launch timing |
|---|---|---|---|---|
| Original DGX Spark | 128GB unified LPDDR5X | 4TB SSD | $3,999 | ~October 15, 2025 |
| Current 128GB tier | 128GB unified LPDDR5X | 4TB SSD | ~$6,950 | 2026, after memory price surge |
| New 64GB tier | 64GB unified LPDDR5X | ~2TB SSD | ~$4,999 | 2026 |
| CPU-Z ARM64 build | n/a | n/a | Free utility | Version 1.06 |
| HWMonitor ARM64 build | n/a | n/a | Free utility | Version 1.06 |
The jump from $3,999 to roughly $6,950 for the same 128GB/4TB configuration in about a year tracks closely with the broader DRAM and NAND cost spike that has rattled PC and server hardware pricing throughout 2026. NVIDIA’s response, introducing a cheaper 64GB tier rather than holding the line on the original price, mirrors how other vendors have handled the same memory crunch: trim capacity to protect a lower price point rather than let the whole lineup drift upward.
How RTX Spark compares to the rest of the local-AI desktop field
RTX Spark does not compete in a vacuum. Apple’s Mac Studio, Framework’s modular Framework Desktop, and AMD’s Ryzen AI Max 395-based mini PCs have all positioned themselves as compact machines capable of running large AI models locally with generous unified or shared memory pools. This site previously broke down the three-way matchup in DGX Spark vs Framework Desktop vs Mac Studio, and the core tradeoffs from that comparison still hold as RTX Spark’s retail rollout continues.
| Platform | CPU/GPU architecture | Unified memory | Primary differentiator |
|---|---|---|---|
| NVIDIA RTX Spark | 20-core Arm CPU + Blackwell GPU (GB10) | Up to 128GB LPDDR5X | Native CUDA stack, ConnectX-7 clustering |
| Apple Mac Studio | Apple silicon, unified memory architecture | Varies by configuration | Mature macOS ecosystem, efficiency |
| Framework Desktop | x86 modular design | Shared system memory | Upgradeability, open hardware approach |
| AMD Ryzen AI Max 395 systems | x86 mobile-derived CPU + integrated GPU | Large shared-memory configurations | x86 compatibility, broad software support |
NVIDIA’s pitch is straightforward: if a workload already runs on CUDA, RTX Spark gives it a path to local execution without the compatibility questions that come with Arm-based Apple silicon or the driver maturity gaps that can still show up on newer AMD integrated graphics stacks. The tradeoff is price and, for now, an ecosystem that is still filling in gaps, exactly the kind of gaps that CPU-Z and HWMonitor support is meant to help close.
Historical context: how we got from DGX Spark to RTX Spark
RTX Spark’s path has been unusually public for a NVIDIA hardware platform. The DGX Spark name first surfaced as a developer-focused appliance, with this site covering its initial 64GB launch at $4,999 amid memory price surges and earlier still, a direct size-up against Apple’s workstation line in DGX Spark vs Mac Studio, which compared 128GB against 512GB memory ceilings. From there, NVIDIA moved the branding toward a broader RTX Spark consumer and workstation identity, documented in coverage of the platform’s Windows-on-Arm PC launch at $2,899 with 128GB of RAM.
The rollout has not been entirely smooth. Early retail enthusiasm was strong enough that ASUS and MSI RTX Spark laptops reportedly sold out at around 100,000 units shortly after launch, and the platform’s arrival briefly rattled chip stocks, with reporting at the time describing Intel and AMD shares sliding roughly 4.4% on competitive concerns. Since those early milestones, pricing has moved substantially due to the memory market, and the ecosystem has kept expanding with new OEM commitments through the back half of 2026. The CPU-Z and HWMonitor update is the latest, smaller step in that same arc: a platform maturing from announcement, to early sellouts, to the unglamorous business of making sure third-party tools actually understand the hardware.
What reviewers and buyers should watch for next
With native monitoring support now available, expect the first wave of independent thermal and clock-behavior testing to follow quickly. Outlets that already cover enthusiast hardware validation, including Tom’s Hardware and Notebookcheck, are well positioned to publish early sensor logs once review units are in hand. Readers should look specifically for sustained-load clock behavior, since integrated CPU-GPU packages with shared thermal budgets can behave very differently under mixed CPU/GPU workloads than under a GPU-only AI inference task.
It is also worth watching whether CPUID or NVIDIA push out a dedicated CPU-Z Validator category for RTX Spark systems, which would make it far easier to compare clock speeds and configurations across different OEM implementations of the same GB10 chip. Right now, a validation database entry is the most reliable form of crowd-sourced confirmation this kind of platform gets, and its absence so far is one more sign that broad retail availability is still catching up to the announcement cycle.
Market impact: what this means for the AI PC category
The timing of this software update lands inside a larger story about local AI hardware pricing. Memory costs have been climbing across the PC industry through 2026, and NVIDIA’s decision to split RTX Spark into a 64GB and a pricier 128GB tier is a direct response to that pressure rather than a planned product strategy. For buyers evaluating whether to wait for RTX Spark, buy a Mac Studio, or build around a Ryzen AI Max 395 mini PC, the calculus now includes a wider price spread than existed at launch a year ago.
For component makers and monitoring-software vendors, the lesson is narrower but still relevant: every new integrated CPU-GPU platform, whether from Apple, Qualcomm, AMD, or now NVIDIA, requires its own round of low-level software validation before it can be trusted by reviewers and power users. CPUID’s willingness to build that support for RTX Spark ahead of a confirmed broad launch suggests the company sees enough commercial demand in the Windows-on-Arm AI PC space to justify the engineering time, alongside its parallel work supporting Qualcomm’s Snapdragon X2 Elite Extreme chips in the same release cycle.
Predictions: where RTX Spark goes from here
- Expect a dedicated CPU-Z Validator category for RTX Spark/GB10 systems within the next few months, once enough review and retail units are in circulation to generate a meaningful sample.
- More OEM partners, beyond the Acer, ASUS, Dell, Gigabyte, HP, and MSI names already reported, are likely to announce RTX Spark configurations as the memory market stabilizes and component costs become easier to forecast.
- The 64GB tier will likely become the volume seller given its lower roughly $4,999 price point, while the 128GB configuration increasingly gets positioned as a workstation-class option for larger local models.
- Competitive pressure from Apple’s Mac Studio line and AMD’s Ryzen AI Max platform will push NVIDIA and its partners to emphasize CUDA software compatibility and ConnectX-7 clustering as RTX Spark’s core differentiators, rather than competing purely on unified memory capacity.
- Further monitoring and diagnostic tool support, from HWiNFO to other sensor utilities, is likely to follow CPUID’s lead as RTX Spark reaches broader retail availability through Q4 2026.
RTX Spark vs the competition: the bottom line
None of the platforms in this category are directly interchangeable. RTX Spark’s bet is that developers and small teams already invested in CUDA will pay a premium for a compact machine that runs that software natively with unified memory and clustering support, rather than switching to Apple silicon or an AMD integrated platform purely for memory capacity per dollar. Whether that bet pays off at the new, higher 128GB price point depends heavily on how quickly memory costs ease industry-wide, something none of NVIDIA, AMD, or Apple fully control on their own.
Frequently asked questions
What is NVIDIA RTX Spark?
RTX Spark is NVIDIA’s platform branding for compact, ARM-based AI desktops and laptops built around the GB10 Grace Blackwell Superchip, which combines a 20-core Arm CPU and a Blackwell GPU with up to 128GB of unified LPDDR5X memory in a single integrated package.
Why do CPU-Z and HWMonitor need a special update for RTX Spark?
RTX Spark uses an Arm-based, integrated CPU-GPU architecture that differs structurally from a conventional x86 desktop with a separate graphics card. Native ARM64 builds of CPU-Z and HWMonitor are needed to correctly read the platform’s topology, clocks, and sensors without relying on x86 emulation.
How much does NVIDIA RTX Spark cost in 2026?
Reported pricing shows a 64GB configuration at approximately $4,999 and a 128GB/4TB configuration at approximately $6,950, up from the original $3,999 launch price for the 128GB tier in October 2025. Exact pricing varies by OEM partner and configuration.
Which companies make RTX Spark systems?
Reporting ties Acer, ASUS, Dell, Gigabyte, HP, and MSI to current GB10-based RTX Spark systems, with earlier reporting also naming Lenovo and Microsoft Surface as planned launch partners.
Is there a confirmed CPU-Z benchmark score for RTX Spark yet?
No. As of this report, there is no confirmed public CPU-Z Validator score or documented HWiNFO benchmark tied to a specific, identifiable RTX Spark unit. Any screenshot circulating without a validation link, BIOS version, and workload description should be treated as unverified.
How does RTX Spark compare to a Mac Studio or Framework Desktop?
RTX Spark’s main differentiator is native CUDA software compatibility and ConnectX-7 networking for clustering multiple units, while Mac Studio offers a mature macOS ecosystem and Framework Desktop offers x86 modularity and upgradeability. Each targets a somewhat different buyer depending on software stack and workflow needs.
Where can I download the updated CPU-Z and HWMonitor ARM64 builds?
Both utilities are distributed free by CPUID through its official CPU-Z and HWMonitor pages, which list separate ARM64 packages alongside the standard x86 Windows releases.
Does RTX Spark replace NVIDIA’s DGX Spark?
RTX Spark extends the same GB10 Grace Blackwell architecture introduced with DGX Spark into a broader lineup of OEM partner systems, rather than replacing it outright. DGX Spark remains NVIDIA’s own branded appliance within that same chip family.
