HP Zbook Ultra G1a “Ryzen AI MAX+ 395″ Laptop Review: Testing Top Strix Halo Vs Top Panther Lake In 14” Form Factor

Ever since I laid my hands on our first AMD Ryzen AI MAX+ 395 Mini PC, I wanted to try the same chip in a compact laptop. AMD's Strix Halo lineup already impressed me a lot in my previous review, and with the launch of Panther Lake, I was more curious to see how Strix Halo would do in a similar lightweight, slim & stunning design that Core Ultra Series 3 CPUs were shipping in.

Well, AMD heard my call, and they were able to arrange a Strix Halo laptop for me instantly. I do want to thank the team at AMD for the quick arrangement. The product I got is the HP Zbook Ultra G1a. It's not new, in fact, it's been out in the market for almost a year now, but I wanted this review to be a battle of the 14" flagships from AMD and Intel, to see who offers the best performance for Productivity, Gaming, multitasking, and AI in the most compact and most accessible laptop form factor. So two awesome machines, the fastest chips from each camp, and the battle setup across various benchmarks.

HP Zbook Ultra G1a Laptop - Specifications At A Glance

In terms of specifications, the HP ZBook Ultra G1a is equipped with the AMD Ryzen AI MAX+ 395 APU. This is the flagship processor within the Strix Halo lineup, which is purpose-built as a Mini Workstation platform, and has a lot to talk about, so let's get started.

The AMD Ryzen AI MAX+ 395 APU features 16 cores in total with 32 threads, and is based on TSMC's 4nm process technology. Unlike the standard Strix lineup, which mixes Zen 5 and Zen 5C cores, the Strix Halo family features the full-fledged, high-performance Zen 5C cores. These 16 cores are packaged within two Core Complexes or CCX chiplets, each housing 8 cores and 32 MB of L3 cache for a total of 64 MB & 16 MB of L2 on the CPU side.

The CPU has a base frequency of 3.0 GHz. The Zen 5 cores clock up to 5.1 GHz and have a TDP rating of 55W at default, which can be configured down to 45W and up to 120W. The SoC is featured on the FP11 platform and is BGA in design. In our testing, the Zbook Ultra G1a utilizes a 60W config that bursts up to 70W for a small amount of time.

2 of 9

For the iGPU, AMD is using its latest RDNA 3.5 architecture, which is a slightly upgraded & more efficient variant of the RDNA 3 architecture.

The AMD Ryzen AI MAX+ 395 features a massive integrated GPU core on the same interposer that is located within its IOD chiplet. This GPU is massive because it features 40 compute units, up from 12 CUs on the top Ryzen AI 300 "Strix" APU. The GPU clocks in at a maximum frequency of 2900 MHz.

The RDNA 3.5 iGPUs support all the latest APIs and AI Frameworks. Plus, RDNA 3.5 also supports the latest upscaling and frame generation features, such as FSR 2, FSR 3, FSR 3 Frame-Gen, and AFMF2, while adding advanced latency reduction technologies such as Anti-Lag 2. It's one of the fastest iGPUs on the market right now.

On the NPU side, the AMD Ryzen AI MAX+ 395 is equipped with an XDNA 2 NPU, which offers a peak TOPS of 50 and supports all the latest AI frameworks. This is the fastest NPU in terms of AI TOPS in the market right now, and the only thing that comes close is the 48 TOPS of the Lunar Lake lineup. But since there's also a massive GPU at disposal, the SoC offers a total AI compute of 126 TOPs.

AMD Ryzen AI MAX 300 "Strix Halo" APU Lineup:

SKU Name	Architectures	CPU Cores	Max Clock	Cache	GPU Cores	TDP
Ryzen AI Max+ 395	Zen 5 / RDNA 3.5	16 / 32	5.1 GHz	80 MB	40 CUs (Radeon 8060S)	45-120W
Ryzen AI Max+ 392	Zen 5 / RDNA 3.5	12 / 24	5.0 GHz	76 MB	40 CUs (Radeon 8060S)	45-120W
Ryzen AI Max 390	Zen 5 / RDNA 3.5	12 / 24	5.0 GHz	76 MB	32 CUs (Radeon 8050S)	45-120W
Ryzen AI Max 385	Zen 5 / RDNA 3.5	8 / 16	5.0 GHz	40 MB	32 CUs (Radeon 8050S)	45-120W
Ryzen AI Max+ 388	Zen 5 / RDNA 3.5	8 / 16	5.0 GHz	40 MB	40 CUs (Radeon 8060S)	45-120W
Ryzen AI Max 380	Zen 5 / RDNA 3.5	6 / 12	4.9 GHz	22 MB	16 CUs (Radeon 8040S)	45-120W

With the specs of the main SoC covered, let's talk about the rest of the specifications. First, we have the memory, which comes in the form of 128 GB of LPDDR5x. The laptop comes pre-configured and pre-soldered in 32 GB and up to 128 GB options. The LPDDR5x modules on the Zbook Ultra G1a operate at 8000 MT/s across a 256-bit wide bus interface.

This wider memory bus is available on AMD's Ryzen AI MAX+ APUs, allowing for up to 256 GB/s of bandwidth. This is essential since the GPU itself only has 32 MB of cache onboard, which means that the LPDDR5x subsystem is what will be used as the primary means of communication.

2 of 9

LPDDR5x is fast, but we expect that AMD will move to the newer LPDDR6 standards as soon as the next-gen Halo chips arrive to further offset the bandwidth requirements. The RDNA 3.5 architecture also has a decent memory compression algorithm, which can reduce some of the bandwidth needs.

The biggest advantage of Strix Halo APUs is that you can dedicate large pools of memory to the iGPU, making them a strong solution for LLMs. The HP Zbook Ultra G1a dedicates 32 GB of LP5X memory to the GPU, out of the box, and you can configure up to 112 GB by selecting the 128 GB mode within the BIOS, leaving 16 GB for the rest of the PC. This makes for a strong AI solution that can run massively sized LLMs without any issues or VRAM limitations that we expect on discrete GPUs, with no dynamic memory allocation options and an even better gaming platform where graphics memory limitations can be bypassed by simply dedicating more system memory to the iGPU.

Our review unit was equipped with a 2 TB PCIe Gen4 SSD. The laptop comes with a single NVMe 2280 M.2 slot. It can be pre-configured with 512 GB, up to 4 TB of storage.

2 of 9

IO includes a Thunderbolt 4 USB Type-C (PD+DP2.1) port, USB Type-C 10 Gbps (PD+DP2.1) port, HDMI 2.1 output, a headphone/mic combo port on the left, and another USB Type-C 10 Gbps (Data+Charge) port, security lock, a Thunderbolt 4 USB Type-C (PD+DP2.1) port on the right.

The laptop comes in two 14" screen options: a WUXGA "UWVA" anti-glare panel with 400 nits of brightness and a 60Hz refresh rate, and a higher-end 2.8K OLED "UWVA" panel with touch, 400 nits of brightness, and up to 120Hz refresh rate. Our unit was equipped with the higher-end OLED version of the screen. There are also four stereo speakers powered by Poly Studio (1W/8Ohm), discrete amplifiers, and an integrated dual array digital microphone.

The laptop is very compact, given its high-end workstation hardware, featuring an 18mm thin design, and weighing roughly 1.57 kg or 3.46lbs. The battery is a 74.5Whr Polymer design, which is charged by a 140W USB Type-C power adapter.

2 of 9

Talking a little bit about the BIOS, it is very barebones for the EVO X2. The first page lists down the various information about the Mini PC and has one important setting, the "Power Mode Select". This gives three options: Performance mode with a 120W target, Balanced mode with an 85W target, and Quiet mode with a 54W target. In the advanced menu, you can find the graphics options from where you can enable UMA to set a custom frame buffer size for the GPU.

HP Zbook Ultra G1a Laptop - CPU Performance

We start by comparing the 3DMark CPU Profile tests. Both the Ryzen AI MAX+ 395 and the Core Ultra X9 388H are 16-core chips, but the thread count is significantly different. Panther Lake maxes out at 16 threads, whereas Strix Halo has 32 threads. Both chips are equal in the max thread tests, but for the 16-16 thread test, Panther Lake edges out Strix Halo with 20% better performance. The single-core performance for Panther Lake is also 5% ahead of Strix Halo.

Strix Halo's memory subsystem and the 256-bit wide channel design allow the system to deliver strong memory and cache performance. It nets over 200 GB/s write bandwidth as advertised, but the latency is a hit, with double that of the Panther Lake system.

For Blender, the AMD Ryzen AI MAX+ 395 "Strix Halo" CPU delivers a 42% boost over the 388H in Classroom render, a 45% boost in Junkshop render, and a stunning 55% boost in Monster render.

In CPU-Z, the single-core score is about on par with the Panther Lake laptop, but the multi-core score sees a 23% uplift.

For Cinebench, Intel's Panther Lake secures the top spot in single-core performance with 130 points, an 18% uplift over Strix Halo. The opposite is seen in MT, though, with the Ryzen AI MAX+ 395 offering 10% higher performance.

In Geekbench 6, both the Ryzen AI MAX+ 395 and the Core Ultra X9 388H CPUs are neck and neck. The Panther Lake chip is slightly faster in single-core tests, while the Strix Halo matches it in multi-core tests.

In Procyon Office, the Zbook Ultra is around on par with the Panther Lake Core Ultra X9 388H CPU. Both CPUs offer respectable performance for office tasks.

WinRAR's compression test sees a huge win for Strix Halo, which ends up 30% faster than Intel's Core Ultra X9 388H.

In SPECviewperf 15.0.1, we can see that the AMD Ryzen AI MAX+ 395 demolishes Intel's Core Ultra X9 388H in content creation and rendering tasks. Even medical, science, and other high-performance tests are a clear win for Strix Halo. This goes on to show the tremendous workstation capabilities that Strix Halo packs, making this laptop a perfect usecase for compact workstation use.

HP Zbook Ultra G1a Laptop - AI Synthetic Performance

Next up, we have our AI benchmarks for the latest Intel and AMD CPUs. First up, we have the Geekbench AI benchmarks, where the Ryzen AI MAX+ 395 takes anywhere from 6% to 13% of the uplift cake from X9 388H on the CPU-only ONNX tests. With DirectML, the lead further swells to 40-200%+ thanks to the GPU on the SoC taking the charge.

For UL Procyon, we again see Intel's Core Ultra X9 388H and its various AI accelerators, such as the NPU and the GPU, offering better performance capabilities than the Ryzen AI offerings.

Trying Out AMD's AI Bundle!

AMD has also recently rolled out an AI Bundle in its latest Radeon Adrenalin driver release, which packs several useful AI tools such as a local chatbot, various text, image, and video generation utilities that utilize the most popular AI LLM/SLM models, and more. These come packaged in an optional 30+ GB container that users can opt to install, giving them access to ComfyUI, Amuse AI, LM Studio, Ollama, and more.

2 of 9

For LM Studio, we tested the most basic Mistral-7B model for text generation, which was able to deliver speedy results using GPU acceleration.

In Amuse, you can unlock your creative side with image generation, image filters, and custom designs, where you can draw patterns that the Vision engine will interpret and generate images from. Amuse requires the download of various models depending on the quality or the type of generation you want, but the software itself utilizes the NPU on Strix Halo for an added 50 TOPs of performance, which is used by the XDNA Super Resolution and XDNA 2 Stable Diffusion features.

2 of 9

And the aforementioned 112 GB memory allocation can be utilized by 120b models, for speedy results that aren't possible on even the highest-end discrete graphics cards, such as the RTX 5090, which packs just 32 GB of memory.

HP Zbook Ultra G1a Laptop - GPU Synthetic Performance

Now we are going to look at the GPU performance, and before we present to you the gaming numbers, we first have to see how the performance fares in synthetic benchmarks. For this purpose, we first want to outline the single-precision FLOPs each iGPU offers. Intel's Battlemage, Alchemist+, and AMD's RDNA 3.5 are entirely different architectures, and despite the FLOPS of the Radeon iGPU being higher, it doesn't necessarily mean that the Radeon iGPU will be faster.

But with that said, the following is how the chips compare:

In 3DMark Speed Way, a purely raytracing benchmark, we see that the Radeon 8060S GPU is the fastest among the bunch, leading over the Arc B390 by 85%, and almost 4x faster than the Radeon 880M, which features 12 compute units versus 40 on the 8060S.

For 3DMark Steel Nomad, the Radeon 8060S offers a 16.5% uplift over the Arc B390 and is 3.67x faster than the Radeon 880M.

In 3DMark Port Royal, the Radeon 8060S iGPU is 27% faster than the Arc B390, and also manages a 3.42x lead over the Radeon 880M.

In 3DMark Time Spy, the Radeon 8060S is 34% faster than the Arc B390, and 3.20x faster than the Radeon 880M.

For Fire Strike, Intel offers great performance even on DX11 APIs, which is a good showcase, as many games still run on DX11. The Radeon GPUs tend to do really well in Firestrike and the Radeon 8060S shows this with a 49% lead over the Arc B390, and a 3.26x lead over the Radeon 880M.

Lastly, we have 3DMark Night Raid, where the Radeon 8060S scores a 28% lead over the Arc B390, and a 2.80x lead over the Radeon 880M.

HP Zbook Ultra G1a Laptop - HD Gaming Performance

With the synthetic performance out of the way, we can start taking a look at pure gaming numbers, and we start off our testing spree with Cyberpunk 2077 running at Medium Preset at 1200P. With Balanced upscaling, the Radeon 8060S GPU achieved a 20% uplift over the Arc B390 and a 2.80x uplift over the Radeon 880M. With Frame-Gen enabled, the Radeon 8060S retains a 31% lead over the Arc B390, and a 3.32x lead over the Radeon 880M.

In Forza Horizon 5, we ran the game using Quality Upscaling at the Medium Preset at 1200P, and the Radeon 8060S is 40% faster than the Arc B390 and 2.64x faster than the Radeon 880M. You can crank Forza to the max, and still get a fast framerate.

F1 24 sees a massive uplift on the Radeon 8060S, acheiving 81% and more than 4.5x performance versus the Arc B390 / Radeon 880M iGPUs, respectively. With frame-gen enabled, this leads to further climbs up to insane figures.

In Horizon Forbidden West, the Radeon 8060S is around 12% faster using just upscaling, but the lead swells to 34% when frame-gen is enabled. Both GPUs can maintain over 60 FPS with frame-gen enabled, and around 60 FPS with just upscaling mode.

In Horizon Zero Dawn at the "Favor Quality" preset, we used the FSR 2 upscaling set to Balanced. Here, the Radeon 8060S delivers a 44% uplift over the Arc B390 and a 2.82x uplift over the Radeon 880M.

The Radeon 8060S and the Arc B390 are the only integrated SoC solutions that can deliver a 60 FPS range in Metro Exodus with RT enabled, as the other iGPUs are stuck in the 20-30 FPS range.

In Resident Evil 9 Requiem, the Radeon 8060S GPU offers almost double or 92% higher performance versus the Arc B390 iGPU.

Lastly, we have The Callisto Protocol, where the Radeon 8060S offers a 25% uplift over the Radeon 8060S, and a 2.57x uplift over the Radeon 880M.

Lastly, we chose to see how the Radeon 8060S and the Arc B390 perform when we crank up the settings to the max in each of the tested titles at 1080p.

HP Zbook Ultra G1a Laptop - Thermals, Power & Battery

A major factor of today's laptops is their power consumption, and in that regard, the AMD Ryzen AI MAX+ 395 has a higher out-of-the-box power profile set at 60-70W versus the 28W of AMD's Ryzen AI 9 365. Both laptops are set to their "Balanced" preset by default, but you can opt into the "Performance" mode through the BIOS. These are just power profiles, and actual power consumption can be very different.

While the peak power of both laptops was seen at 70W, the gaming power was slightly different. The Strix Halo laptop was close to 45-50W when plugged in, while the Panther Lake laptop was around 30-40W, also when plugged in. As for application power, in SPEC15, the Strix Halo power was hitting 60W with periodic bursts of up to 70W while Panther Lake was mostly static around 50W.

Panther Lake does get a bit hotter than the other chips. This thin and light design from ASUS is rated at 45W, so we did see thermal throttling in AIDA64's stress test. That wasn't the case while gaming, but even then, the laptop gets warm around 70-80 °C.

As for Strix Halo, ran remarkably cooler as per the HWmonitor stats, with a peak of 81C, which is nowhere close to the 96-100C we saw on Panther Lake. Even in gaming, it ran around 61C which is very decent. But one difference that I felt was that the chassis of the HP Zbook Ultra G1a got really warm at the bottom & you could hear the fans spin when it was being pushed to 100%. That wasn't the case with the ZenBook Duo, which was much cooler to hold and also didn't operate as erratically.

One thing where Intel clearly has the lead is the battery times. In all our usecases, the Panther Lake laptop lasted us almost twice as much run time versus Strix Halo. The Strix Halo laptop does come with a smaller 74.5Whr battery versus the 99Whr battery on the Panther Lake system.

Conclusion - Workstation Superpowers In a Compact 14" Design

The Future is Fusion, which used to be AMD's motto a decade ago when they first ventured into APUs. Back in the day, I used to believe that this Fusion was a reference to combining a powerful CPU with a high-end GPU, all on a single chip. The exascale APU patents from several years ago brought us closer to that dream, but it would take years of engineering and a solid framework to build a chip that offered the best of both worlds. To me, Strix Halo or Ryzen AI MAX+ is the full realization of that motto.

AMD Strix Halo Achieves Next-Level SoC Performance

The 16 Zen 5 cores offer tremendous productivity and multi-threaded performance. The Core Ultra X9 388H is a triumph for Intel, but its 16 threads can only do so much. That's where the 32 threads of the Ryzen AI MAX+ 395 showcase their absolute raw crunching prowess, whether it be office-based use or high-end workstation usage, such as content creation or rendering, the 395 was superb across all our tests.

2 of 9

The memory subsystem of the Strix Halo SoC also makes it a potent solution for AI, handling chonkier LLMs with ease, with these chips dedicating up to 112 GB to the GPU alone. This also works in favor of games where you're no longer restricted to measly 8 GB dedicated memory; instead, you can select a pool size from 512 MB, up to 112 GB, and run games perfectly even at higher resolutions. The LPDDR5X-8000 256-bit config also provides a decent amount of bandwidth, but I can already see where AMD is going to go next.

For gaming, we have the following recommendation for Strix Halo users:

• 1080p AAA Gaming: Playable With Max Settings (Some with Upscaling set to Quality Mode)
• 1440p AAA Gaming: Playable With High Settings (Some with Upscaling+Frame-Gen set to Quality Mode)
• 2160p AAA Gaming: Playable With Med/High Settings (Upscaling Set To Quality/Balanced With Frame-Gen)

The Radeon 8060S shines on the Ryzen AI MAX+ 395 SoC, delivering ample amounts of horsepower. Do note that this is a workstation-first laptop, but it can also game superbly well. That's thanks to the Radeon Pro driver and AMD's support that retains similar levels of optimizations across its consumer and professional driver branches. With that said, you are looking at a stunning 4x uplift over the older 12 compute unit Radeon iGPUs, & even against Intel's latest and greatest, the Radeon 8060S sets it apart, offering near RTX 5070 levels of performance backed by a gigantic shared memory pool.

Now for thermals and power, the HP Zbook Ultra G1a doesn't give us a lot of tuning options for power like the Mini PCs we tested with the same chip, which let us tune the power limit up to 120W. The SoC on the Zbook is configured between 60-70W, which is still very decent, and makes sense for the 14" chassis. The good part is that despite the limited chassis, we didn't encounter any thermal throttling and the chip ran much cooler than Intel's Panther Lake SoCs in a similar 14" design.

As for the overall look and feel of the Zbook Ultra G1a, HP did a nice job with the color scheme, giving it that silver color, and you have really tiny bezels on the OLED display. It comes with a 120Hz 2880x1800 resolution display, which is solidly driven by the SoC, and really makes for a great experience. The multi-DP outputs through USB Type-C ports and TB4 connectivity make this a formidable workstation powerhouse in such a small form factor.

Overall, the HP Zbook Ultra G1a is a portable workstation powerhouse, all powered by AMD's disruptive Strix Halo "Ryzen AI MAX+" SoCs, which have reshaped the AI PC segment with unseen levels of CPU and GPU performance. At $4000 US+, it is expensive but also expected of today's memory and storage shortages.

Follow Wccftech on Google to get more of our news coverage in your feeds.