Product Info for this review focuses on the HP ZBook Ultra G1a (2025), a premium 14-inch workstation-class laptop that typically lands in the $4000+ range depending on configuration. The big reason it’s still drawing attention nearly a year after release is simple: it brings AMD’s most ambitious “Strix Halo” APU design into a compact, travel-friendly chassis—exactly the kind of thin-and-light form factor many people associate with Intel’s latest flagship mobile platforms.
After spending time with an earlier mini PC powered by the same Ryzen AI MAX+ 395, the next obvious question was how that silicon behaves inside a real-world laptop: thermals, sustained power behavior, portability, and whether it can truly serve as an all-in-one machine for productivity, creative work, light-to-serious gaming, multitasking, and on-device AI. The HP ZBook Ultra G1a is built to answer that question, making it a natural contender in the “14-inch flagship” showdown space.
At the heart of the ZBook Ultra G1a is the AMD Ryzen AI MAX+ 395 APU, the top chip in the Strix Halo family. This is positioned less like a typical ultrabook processor and more like a compact workstation platform—designed to blend high CPU throughput, unusually strong integrated graphics, and a modern NPU for AI acceleration.
The CPU portion features 16 cores and 32 threads built on TSMC’s 4nm process. Instead of the mixed-core approach seen in some other mobile designs, this Strix Halo configuration focuses on high-performance Zen cores across the full 16-core layout. Internally, those cores are split into two core complexes, each with 8 cores and 32MB of L3 cache, adding up to 64MB of L3 cache total, alongside 16MB of L2 cache. Base clock is listed at 3.0GHz, with boost up to 5.1GHz. Power is notably flexible: a default 55W rating, configurable down to 45W and up to a very aggressive 120W depending on the system vendor’s tuning. In this particular laptop, the configuration runs around 60W and can briefly boost to roughly 70W for short bursts.
Where Strix Halo really separates itself from typical thin-and-light laptop chips is the integrated GPU. AMD uses RDNA 3.5 here, an updated and efficiency-tuned evolution of RDNA 3. The iGPU is extremely large by integrated standards, boasting 40 compute units—massively higher than the 12 CUs found on the top-end Ryzen AI 300 “Strix” parts. Peak clock reaches up to 2900MHz, and feature support includes modern graphics APIs and a full set of AMD upscaling and frame-generation options such as FSR 2 and FSR 3, plus frame generation and latency-reduction features like Anti-Lag 2. In practical terms, it’s positioned as one of the fastest iGPUs currently available, aiming to reduce the need for a discrete GPU in many workflows.
On the AI side, the Ryzen AI MAX+ 395 carries an XDNA 2 NPU rated at up to 50 TOPS, with support for modern AI frameworks. Combined with the GPU’s AI capabilities, the platform’s total AI compute is quoted at 126 TOPS. That matters for people shopping specifically for “AI PCs” who want stronger local performance for supported apps and AI-enhanced workflows without immediately relying on cloud processing.
Memory is another defining part of this machine’s identity. The ZBook Ultra G1a uses LPDDR5x running at 8000 MT/s and—crucially—connects it over a 256-bit wide interface. That enables up to 256GB/s of memory bandwidth, which is especially important because a powerful integrated GPU lives and dies by memory throughput. Configurations come pre-soldered from 32GB up to a massive 128GB. The wider memory bus and high bandwidth help feed the 40-CU RDNA 3.5 iGPU, which only has 32MB of cache onboard and therefore leans heavily on system memory performance.
This memory design also turns into a major advantage for AI workloads, especially large language models. Unlike typical integrated systems that feel constrained by fixed VRAM, Strix Halo allows dedicating a large pool of system memory to the GPU. Out of the box, this ZBook configuration allocates 32GB of LPDDR5x to the iGPU, and it can be pushed much further via BIOS options—up to 112GB reserved for graphics when using a 128GB configuration, leaving 16GB for the rest of the system. That kind of flexibility can be a game-changer for AI experimentation, larger models, and workloads that would otherwise run into VRAM ceilings on many discrete GPUs.
Storage on the reviewed unit is a 2TB PCIe Gen4 SSD. Expansion is straightforward but limited: there’s a single NVMe 2280 M.2 slot, with factory options ranging from 512GB up to 4TB.
Connectivity is well suited for workstation use and multi-display setups. Ports include Thunderbolt 4 over USB-C with power delivery and DisplayPort 2.1 support, additional 10Gbps USB-C ports (also with PD and DP2.1), HDMI 2.1, and a standard headphone/mic combo jack. There’s also a security lock slot, reinforcing the business and professional target audience.
For the display, HP offers two 14-inch options. The more basic choice is a WUXGA anti-glare panel rated at 400 nits with a 60Hz refresh rate. The higher-end option, included on the reviewed unit, is a 2.8K OLED touch panel also rated at 400 nits, but capable of up to a 120Hz refresh rate—an attractive pairing for both creator-friendly visuals and smoother everyday use.
Despite the workstation-class internals, the ZBook Ultra G1a keeps the premium thin-and-light promise. The chassis is about 18mm thin and weighs roughly 1.57kg (3.46 lbs). Battery capacity is 74.5Wh, and charging is handled by a 140W USB-C power adapter, which is convenient for travel and docking setups.
One software/firmware note mentioned in testing is that BIOS options are fairly minimal, but there are still a couple of settings that matter for performance tuning. A power mode selector allows switching between performance-focused and quieter profiles based on target wattage. There are also graphics-related options for adjusting UMA frame buffer allocation, which ties directly into the “reserve more memory for the iGPU” advantage that makes this platform so compelling for graphics-heavy and AI-heavy workloads.
Moving into early CPU benchmark impressions, a 3DMark CPU Profile comparison highlights an important architectural reality: even when two competitors advertise “16 cores,” the thread count and how those cores are utilized can shift outcomes. In the referenced results, both chips trade blows at the highest thread test, while the competing 16-thread design shows a noticeable edge in the 16-thread portion of the benchmark. It’s an early sign that performance leadership may vary depending on whether your workloads scale better with high thread counts, stronger per-thread throughput, or sustained power behavior in a thin chassis.
Overall, the HP ZBook Ultra G1a stands out as a premium 14-inch laptop that blends workstation ambitions with real portability. Its key selling points are crystal clear: a 16-core/32-thread flagship AMD APU, one of the most powerful integrated GPUs available, unusually high memory bandwidth, and the ability to allocate massive memory pools to graphics and AI workloads—features that can make a meaningful difference for creators, developers, and power users who want serious performance without hauling a bulky 16-inch machine.Intel’s upcoming Panther Lake and AMD’s Ryzen AI MAX+ 395 “Strix Halo” are shaping up to be one of the most interesting laptop CPU face-offs in years, because the results don’t point to a single, universal winner. Instead, the performance story depends heavily on what you actually do on your laptop: quick everyday responsiveness, heavy multi-threaded production work, workstation-class 3D and viewports, or AI-accelerated tasks.
Single-core performance: Panther Lake edges ahead where responsiveness matters most
If you care about snappy feel in lighter workloads—app launching, UI responsiveness, general productivity, and tasks that lean on one fast core—Panther Lake comes out looking strong. In multiple tests, the Panther Lake-based Core Ultra X9 388H holds a small lead in single-core results. One note from the testing highlights that Panther Lake’s single-core performance is also about 5% ahead of Strix Halo.
That advantage shows up clearly in Cinebench single-core, where Panther Lake posts 130 points, giving it an 18% uplift over Strix Halo. In Geekbench 6, the two chips are extremely close overall, but Panther Lake still comes out slightly faster in single-core.
Multi-core performance: Strix Halo pushes ahead when all cores get to work
Once you move into heavier multi-threaded workloads, Ryzen AI MAX+ 395 starts flexing. In CPU-Z, Strix Halo’s single-core score is roughly on par with Panther Lake, but its multi-core score jumps ahead with a 23% uplift. Cinebench flips the script as well: while Panther Lake leads in single-core, Strix Halo takes a 10% win in multi-threaded performance.
In other words, if your workflow regularly scales across many cores—batch jobs, code compilation, big compute tasks, or sustained creation work—Strix Halo has the momentum.
Memory and cache behavior: huge bandwidth for Strix Halo, but latency tells another story
Strix Halo’s memory subsystem is a major part of its identity. Thanks to a 256-bit wide channel design, it delivers very strong memory and cache performance, surpassing 200 GB/s write bandwidth as advertised. That kind of throughput can be valuable in bandwidth-hungry workloads.
The trade-off is latency. The same testing notes that memory latency on Strix Halo is roughly double that of the Panther Lake system. So while the raw bandwidth numbers are impressive, some tasks that are sensitive to latency rather than throughput may not benefit as much as the bandwidth headline suggests.
For 3D rendering in Blender, Strix Halo clearly separates itself from Panther Lake. The Ryzen AI MAX+ 395 delivers a 42% boost over the Core Ultra X9 388H in the Classroom render, a 45% boost in Junkshop, and a massive 55% boost in Monster. Those are not minor differences—if Blender is a key part of your work, Strix Halo looks like the more compelling option based on these numbers.
Office productivity: essentially even
In Procyon Office, the HP Zbook Ultra G1a with Ryzen AI MAX+ 395 lands around on par with the Panther Lake Core Ultra X9 388H laptop. The takeaway is simple: both platforms offer strong performance for typical office-focused tasks, and you’re unlikely to feel a dramatic difference in day-to-day document work, spreadsheets, or general business apps.
Compression and everyday heavy lifting: WinRAR is a big Strix Halo win
WinRAR’s compression benchmark is another standout for AMD’s chip. Strix Halo ends up 30% faster than the Core Ultra X9 388H. If you compress and archive large files often—or run workflows that resemble this kind of integer-heavy throughput—this is a meaningful advantage.
Workstation graphics and pro viewports: Strix Halo dominates SPECviewperf
One of the biggest statements in the data comes from SPECviewperf 15.0.1, where the Ryzen AI MAX+ 395 “demolishes” the Core Ultra X9 388H in content creation and rendering workloads. Even specialized workloads like medical and science tests come out as clear wins for Strix Halo, pointing to serious compact workstation potential for laptops built around this platform.
AI performance: mixed results depending on the test and accelerator path
AI benchmarks show a more nuanced split.
In Geekbench AI, Strix Halo leads the CPU-only ONNX tests by around 6% to 13% over the Core Ultra X9 388H. But with DirectML, the gap can expand dramatically—40% to 200%+—because the integrated GPU on the SoC plays a much larger role.
However, UL Procyon tells a different story. In those AI workloads, Intel’s Core Ultra X9 388H and its AI accelerators (notably the NPU and GPU options) are described as delivering better overall performance capabilities than the Ryzen AI offerings. In short: depending on whether your AI work is CPU-bound, GPU-accelerated via certain frameworks, or tuned for specific NPUs, the “best” chip can change.
What this means if you’re choosing a laptop in 2026
Panther Lake (Core Ultra X9 388H) looks like the better pick if you prioritize top-tier single-core performance and strong results in certain AI benchmark suites that lean on Intel’s accelerator stack.
Ryzen AI MAX+ 395 “Strix Halo” looks like the stronger option if you do serious multi-core work, Blender rendering, compression-heavy tasks, and especially if you want compact workstation-style performance for professional visualization workloads.
The most important takeaway is that these aren’t two chips separated by small, forgettable margins. The differences can be minor in office work and some CPU tests, but they can become dramatic in content creation, workstation benchmarks, and certain GPU-accelerated AI scenarios—so the right choice depends on your workload, not brand loyalty.AMD is quietly making a strong case that “AI PC” features and real gaming performance don’t have to be limited to bulky laptops with power-hungry discrete graphics. A good example is the HP ZBook Ultra G1a configuration built around the Ryzen AI MAX+ 395 and its integrated Radeon 8060S GPU, which shows just how far modern iGPUs have come—especially when paired with smart software features and generous unified memory.
Alongside the performance push, AMD has also introduced a new AI Bundle inside its latest Radeon Adrenalin driver. The idea is simple: give users a ready-to-install kit of popular local AI tools without forcing them to piece everything together themselves. It’s optional, but it’s not small—expect a 30+ GB container if you choose to install it. Once enabled, it can provide access to a mix of local chatbot and content-creation utilities, including ComfyUI, Amuse AI, LM Studio, and Ollama, with support for widely used AI LLM and SLM models across text, image, and video workflows.
In quick testing, LM Studio was run with a basic Mistral-7B model for text generation and delivered fast responses thanks to GPU acceleration. For users who want something more visual, Amuse AI leans into image generation, filters, and custom design tools, including a feature where patterns you draw can be interpreted by a vision engine to generate images. Amuse requires downloading different models depending on the style and quality you want, but on Strix Halo-based systems it can also tap into the NPU, adding up to 50 TOPs of extra AI capability. That NPU acceleration helps power features such as XDNA Super Resolution and XDNA 2 Stable Diffusion.
One of the more interesting takeaways is how unified memory can change the AI story on a laptop. With enough memory available—figures like a 112 GB allocation are referenced—it becomes possible to run extremely large models (even up to 120B parameters) locally with speed levels that are difficult to match on many top-end discrete GPUs simply because those cards are limited by VRAM capacity. For example, even flagship-class consumer GPUs with 32 GB of memory can become constrained in these scenarios.
Still, most people looking at a mobile workstation like the ZBook Ultra G1a also want to know the bottom line for graphics performance. Before getting into game frame rates, synthetic benchmarks help show how the Radeon 8060S and competing integrated solutions stack up. It’s worth noting that raw FLOPs don’t automatically translate into real performance because Intel and AMD are using very different GPU architectures here. Even so, the results paint a consistent picture.
In 3DMark Speed Way, a ray tracing-focused benchmark, the Radeon 8060S comes out on top—leading the Arc B390 by 85% and landing nearly 4x higher performance than the Radeon 880M, which has 12 compute units compared to the 8060S’s 40.
In 3DMark Steel Nomad, the Radeon 8060S posts a 16.5% uplift over the Arc B390 and measures 3.67x faster than the Radeon 880M.
In 3DMark Port Royal, another ray tracing test, the Radeon 8060S is 27% faster than the Arc B390 and holds a 3.42x advantage over the Radeon 880M.
In 3DMark Time Spy, the lead expands further: the Radeon 8060S is 34% ahead of the Arc B390 and 3.20x faster than the Radeon 880M.
DX11 performance still matters because a lot of PC games continue to rely on it, and Fire Strike is a useful proxy there. In that test, the Radeon 8060S shows a 49% lead over the Arc B390 and a 3.26x lead over the Radeon 880M.
Finally, in 3DMark Night Raid, the Radeon 8060S maintains its advantage with a 28% lead over the Arc B390 and 2.80x higher performance than the Radeon 880M.
Synthetic scores are one thing, but playable results in real games are what most buyers care about—especially at common laptop resolutions like 1200p. In Cyberpunk 2077 at 1200p using the Medium preset, the Radeon 8060S delivered a 20% uplift over the Arc B390 when using balanced upscaling, and it was 2.80x faster than the Radeon 880M. Turn on frame generation and the 8060S keeps stretching its advantage, holding a 31% lead over the Arc B390 and a 3.32x lead over the Radeon 880M.
Forza Horizon 5 also favors the Radeon 8060S. Tested at 1200p with the Medium preset and quality upscaling, it runs 40% faster than the Arc B390 and 2.64x faster than the Radeon 880M—strong enough results that the game can be pushed much higher while still keeping frame rates feeling smooth.
F1 24 is where the gap becomes dramatic. The Radeon 8060S delivers an 81% uplift over the Arc B390, and it’s more than 4.5x faster than the Radeon 880M in the same kind of comparison. With frame generation enabled, performance climbs even further into territory that’s hard to associate with integrated graphics.
In Horizon Forbidden West, the Radeon 8060S is about 12% faster when using upscaling, but enabling frame generation increases the lead to 34%. Notably, both GPUs referenced here can stay above 60 FPS with frame generation on, and hover around 60 FPS with upscaling alone—making it a practical, playable setup rather than a “benchmark-only” win.
Taken together, the message is clear: AMD’s latest integrated Radeon solution in systems like the HP ZBook Ultra G1a isn’t just about checking an “AI PC” box. It’s combining local AI tooling (via an easy-to-install driver bundle), meaningful NPU acceleration for certain creative features, and iGPU performance that can compete surprisingly well in modern games at 1200p—especially when you use upscaling and frame generation wisely. If you’re shopping for a thin-and-light workstation or premium laptop that can handle both creation and play without relying on a discrete GPU, this is exactly the kind of configuration worth paying attention to.Testing across several modern games shows just how far integrated graphics have come, especially on AMD’s Ryzen AI MAX+ 395 with the Radeon 8060S inside the HP Zbook Ultra G1a. In multiple 1080p scenarios, this iGPU doesn’t just look competitive—it often looks like the clear step up versus Intel’s Arc B390 in the ASUS Zenbook Duo 2026, and it absolutely separates itself from older integrated Radeon solutions like the Radeon 880M.
In Horizon Zero Dawn running the “Favor Quality” preset, the setup used FSR 2 upscaling on Balanced. Under those conditions, the Radeon 8060S posted a big leap: roughly a 44% uplift over the Arc B390, and an enormous 2.82x uplift over the Radeon 880M. That kind of gap highlights what a newer, higher-tier integrated GPU can do when paired with the right memory subsystem and tuning.
Metro Exodus with ray tracing enabled is one of the toughest real-world checks for any integrated solution, and results there make the landscape pretty clear. The Radeon 8060S and the Arc B390 were the only integrated SoC graphics options in this set that could reach the 60 FPS range with RT on. Everything else clustered much lower, stuck around the 20–30 FPS range. In other words, if ray tracing is non-negotiable on an iGPU laptop, the list of realistic options narrows fast.
Resident Evil 9 Requiem spotlighted another strong win for AMD’s Radeon 8060S. Here, it delivered nearly double the performance of the Arc B390—about a 92% advantage—showing that the gap isn’t limited to one engine or one style of game.
The Callisto Protocol added more context to the stack. In this title, the Radeon 8060S again came out ahead, with a clear lead over the Arc B390 and a massive 2.57x uplift versus the Radeon 880M. (The overall takeaway remains consistent: Radeon 8060S is operating in a different tier than last-generation iGPUs.)
To push things further, the testing also included a “max settings at 1080p” pass across several demanding games, including Cyberpunk 2077 (with and without frame generation), F1 24, Forza Horizon 5, Horizon Zero Dawn, Metro Exodus Enhanced Edition, and The Callisto Protocol. The goal here wasn’t just to chase averages—it was to see which platform stays viable when you stop optimizing for the iGPU and start playing like you would on a traditional gaming system.
Power, thermals, and battery life tell an equally important part of the story, especially in thin 14-inch designs where performance is often limited by sustained wattage and cooling. Out of the box, the Ryzen AI MAX+ 395 platform runs a higher power profile (around 60–70W) compared with a Ryzen AI 9 365 configuration at 28W. Both machines referenced in the comparison used a default “Balanced” preset, with a “Performance” mode available through BIOS. These profiles guide behavior, but real usage can still vary by workload.
In practice, both laptops hit peaks around 70W. During gaming while plugged in, the Strix Halo system (Ryzen AI MAX+ 395) typically landed around 45–50W, while the Panther Lake system sat closer to 30–40W. For heavier application tests like SPEC15, the Strix Halo machine often pushed to about 60W with bursts up to 70W, while Panther Lake hovered more steadily around 50W.
Thermals also split in an interesting way. The Panther Lake laptop ran hotter overall in this thin-and-light format, and thermal throttling showed up during AIDA64 stress testing. Gaming was better behaved, but temperatures still lived in the 70–80°C range. The Strix Halo system, by monitoring data, ran notably cooler—peaking around 81°C versus the 96–100°C range observed on the Panther Lake platform. In gaming, the Strix Halo chip sat around 61°C, which is impressive for the class. There was a trade-off, though: while the Zenbook Duo stayed cooler to hold and didn’t ramp up as erratically, the HP Zbook Ultra G1a could get quite warm on the bottom and the fans became clearly audible under full load.
Battery life is where Intel’s platform held a strong advantage. Across the tested everyday use cases, the Panther Lake laptop delivered close to double the runtime versus the Strix Halo system. Battery capacity likely plays a role too, since the Zbook’s battery is 74.5Whr, while the Panther Lake machine uses a larger 99Whr pack.
The broader conclusion is that this new wave of high-performance SoCs is making “fusion-style” computing feel real in ways earlier generations never quite achieved. On the CPU side, Ryzen AI MAX+ 395 brings 16 Zen 5 cores and 32 threads, which creates a meaningful advantage in heavy productivity and multi-threaded workloads compared with a 16-thread competitor. That translates well beyond office tasks into serious workstation workflows like content creation and rendering.
On the AI side, the Strix Halo memory setup is positioned as a major strength, especially for larger models and heavier GPU memory needs. One standout detail is the ability to dedicate a very large pool of memory to the GPU—up to 112GB. That flexibility can matter for both AI workloads and gaming, since it reduces the typical iGPU pain point of being boxed into a small fixed VRAM amount. The LPDDR5X-8000 256-bit configuration also helps keep bandwidth in a healthier range for an integrated design.
For people looking at the Ryzen AI MAX+ 395 and Radeon 8060S as a gaming-capable workstation platform, the practical expectations break down like this:
At 1080p in AAA games, max settings are often within reach, though some titles will benefit from upscaling set to Quality.
At 1440p, high settings are generally the target, sometimes paired with upscaling and frame generation on Quality.
At 2160p, medium-to-high settings can be workable when leaning on upscaling (Quality/Balanced) along with frame generation.
All of that is powered by what’s essentially a workstation-first laptop approach. The graphics performance is boosted by professional driver support, and the overall package aims to deliver consistent performance and stability for work while still being surprisingly capable in games. The HP Zbook Ultra G1a also brings a premium build direction, including a 120Hz 2880×1800 OLED display, slim bezels, and strong external display support through USB-C with multiple DisplayPort outputs and Thunderbolt 4 connectivity.
The biggest caveat is price. With configurations reaching $4000+ in the US, this is firmly in premium mobile workstation territory. Still, the combination of high-end CPU throughput, unusually strong integrated GPU performance, and a design that fits it into a compact 14-inch chassis is exactly why this class of “AI PC” hardware is drawing so much attention right now.
