HP ZGX Nano G1n AI Station Evaluate: A Safe, Sustainable Desk-Facet AI Node

The DGX Spark platform is acquainted territory for us at this level. We’ve reviewed the Dell, ASUS, Acer, and Gigabyte takes on NVIDIA’s GB10 Grace Blackwell reference design, and the core elements are constant throughout all of them: 1,000 TOPS of FP4 compute, 128GB of unified LPDDR5x reminiscence, and twin 200GbE networking in a 150mm chassis. HP’s ZGX Nano G1n AI Station builds on that basis, however the best way HP has constructed round it units this unit other than the remainder of the Spark subject.

Essentially the most seen variations are in supplies and development. HP wraps the ZGX Nano in a chassis constructed from as much as 75% recycled aluminum and 20% recycled metal, with packaging that carries as much as 93% recycled content material. The interior format splits the chassis into higher and decrease halves, making it simpler to entry elements just like the SSD and coin-cell battery than on a number of of the Spark models we’ve examined. Thermally, HP charges the system at 22 dBA idle and 27.6 dBA beneath intensive workloads, quiet for a system dissipating roughly 780 BTU/hr at peak.

Safety is the place HP pushes furthest previous the reference platform. The ZGX Nano ships with TPM 2.0 working in FIPS 140-2 licensed mode, meets Frequent Standards EAL4+, and consists of BIOS-level safe boot and PXE controls. Storage is factory-installed as a self-encrypting OPAL NVMe drive. Taken collectively, HP is positioning this unit not solely as a developer desk-side AI node but in addition as a system that may function inside regulated environments the place provide chain certifications, encryption at relaxation, and tamper resistance matter for procurement.

Construct and Design

The HP ZGX Nano G1n takes a noticeably completely different method to the DGX Spark design in contrast with the opposite techniques we’ve checked out to this point (see our Dell/ASUS/Acer/Gigabyte critiques). As an alternative of the extra widespread construct, the place the internals really feel tucked right into a high cowl, HP splits the chassis into higher and decrease halves, making the interior format simpler to know as soon as inside. What first seems extra difficult seems to be pretty sensible, with easy entry to components just like the coin-cell battery and SSD after eradicating only a handful of screws. That extra thought-about inner construction additionally carries over to the outer construct, the place HP locations better emphasis on how the system is constructed and the supplies used all through.

That mentioned, HP wraps it in a modern black case with a 150mm-square footprint and depends closely on recycled supplies. Particularly, the construct makes use of as much as 75% recycled aluminum, 20% recycled metal, and vital quantities of post-consumer recycled plastics. Even the packaging displays this dedication. Corrugated supplies comprise as much as 93% recycled content material, and plastic packaging incorporates not less than 30% recycled content material.

Thermally, the system depends on forced-air cooling. This can be a notable engineering alternative given the density of the NVIDIA GB10 Grace Blackwell Superchip. Regardless of its compact footprint, HP specifies a full thermal envelope. Underneath most load, the system dissipates as much as roughly 780 BTU/hr, relying on configuration. Peak system energy draw reaches roughly 228W. Moreover, HP advertises comparatively low noise ranges, rated at 22 dBA at idle and 27.6 dBA beneath intensive workloads.

Bodily, the unit measures 5.9 x 5.9 x 2.01 inches with out toes, firmly inserting it in ultra-compact territory. HP explicitly states that the unit is just not rack-mountable, reinforcing its function as a desk-side AI node slightly than conventional information heart infrastructure. Serviceability is minimal by design. Customers want a #1 Phillips screwdriver to entry inner elements, and most elements, together with reminiscence, are non-user-replaceable.

Internally, the ZGX Nano makes use of NVIDIA’s reference board design, as do many different OEMs constructing on the DGX Spark platform. The LPDDR5x reminiscence is soldered on to the board and runs at as much as 8533 MHz. General, the platform prioritizes effectivity and density over modularity.

Safety and Upgradability

HP locks down the ZGX Nano G1n by design. It options an built-in TPM 2.0 module that operates in FIPS 140-2-certified mode, meets Trusted Computing Group specs, and is Frequent Standards EAL4+ licensed. BIOS-level protections embody safe boot controls, PXE-based distant boot capabilities, and the flexibility to disable boot from detachable media fully.

From a {hardware} standpoint, HP is express: this technique is just not upgradeable. The 128GB of LPDDR5x unified reminiscence sits soldered on to the board. Moreover, patrons should choose storage on the time of buy. Whereas the only M.2 slot helps PCIe Gen5 x4 electrically, manufacturing unit configurations ship with PCIe Gen4 x4 NVMe SSDs. These are available in 2TB or 4TB capacities and are all self-encrypting OPAL drives.

HP notes that spare components will stay obtainable for as much as 5 years after manufacturing ends. Nonetheless, that is basically an appliance-style system slightly than a modular workstation.

I/O and Growth

The entrance of the unit is minimalist, that includes solely an influence button and a standing LED. On the again, the system presents a dense array of high-performance connectivity choices. HP delivers energy through a normal NVIDIA-recommended 240W USB-C adapter and warns that third-party adapters might trigger degraded efficiency or instability.

Three USB 3.2 Sort-C ports present USB connectivity, every working at 20 Gbps and supporting DisplayPort 1.4a Alt Mode. A devoted HDMI 2.1a port offers extra show output. For networking, the system consists of each a Realtek RTL8127-CG 10GbE controller and an NVIDIA ConnectX-7 controller, offering twin 200GbE QSFP112 ports, every with 200 Gbps throughput.

The networking stack helps a variety of enterprise options. These embody PXE boot, Wake-on-LAN, VLAN tagging (802.1Q), time synchronization (802.1as/1588), and full-duplex operation throughout all supported speeds. Moreover, a Wi-Fi 7 (802.11be) 2×2 module with Bluetooth 5.4 offers wi-fi connectivity and helps MU-MIMO, WPA3 safety, and operation throughout the two.4GHz, 5GHz, and 6GHz bands.

Graphics and Audio

The built-in NVIDIA Blackwell GPU within the GB10 Superchip handles all graphics duties. The system helps as much as 8K output at 60Hz through USB-C DisplayPort 1.4a and 8K at 30Hz through HDMI 2.1a. HP recommends utilizing direct cable connections for 8K output, as adapters or docks might trigger instability or degrade sign high quality.

Audio runs over HDMI, with no devoted analog audio outputs. This aligns with the system’s positioning as a compute node slightly than a conventional multimedia workstation.

Thermals Testing

CPU Temperature

Throughout CPU thermal testing, the HP ZGX Nano G1n reached a peak temperature of 77.3°C in the course of the workload’s extra intense bursts. This locations HP beneath the most well liked techniques within the comparability stack throughout peak transitions, as different models climbed into the 90°C vary. Because the workload transitioned into Equal ISL/OSL after which Decode Heavy, CPU temperatures stabilized slightly than persevering with to rise sharply.

On the decrease finish, the CPU recorded a minimal temperature of 36.4°C throughout light-load circumstances. This implies the HP has efficient warmth dissipation when the system is just not beneath heavier computational stress. General, the ZGX demonstrated managed burst CPU thermal habits with secure sustained-load efficiency.

GPU Temperature

GPU thermals adopted an analogous sample. During times of heavy acceleration, the GPU reached a most temperature of 69°C. This positions HP on the cooler facet of the comparables throughout peak burst circumstances, with a number of different techniques (just like the Dell, ASUS, and Founders Version) working noticeably hotter on the high finish. As exercise shifted into Equal ISL/OSL and Decode Heavy phases, GPU temperatures leveled off and remained secure.

The GPU recorded a minimal temperature of 34°C throughout lighter phases, indicating strong idle thermal capabilities.

NVMe Temperature

In the course of the Equal section, the NVMe drive reached roughly 42°C, displaying solely a gradual rise from its resting baseline. Because the workload shifted to Prefill Heavy, the storage temperature rose noticeably, starting from 42°C to 47°C. In Decode Heavy, the drive operated in its warmest vary, 47°C to 54°C, the place it peaked, but remained noticeably beneath most different Spark techniques.

NIC Temperature

In the course of the Equal section, NIC temperature ranged from 39°C to 52°C, displaying a gradual climb, indicating average thermal buildup as community exercise ramps up early within the run.

In Prefill Heavy, NIC thermals elevated, starting from 48°C to 64°C, as a result of this section locations far more sustained strain on the networking subsystem. Throughout Decode Heavy, NIC temperature was in its warmest vary, 52°C to 68°C, the place the height was reached. Nonetheless, thermal habits remained secure all through the take a look at.

GPU Energy Consumption

In the course of the Equal section, GPU energy consumption ranged from 2.86W to simply over 40W, inserting the HP ZGX Nano G1n in the course of the pack.

In Prefill Heavy, GPU energy began at roughly 37W, dipped to as little as 35W, and spiked to as excessive as 69W, making this probably the most power-intensive section of the run.

Throughout Decode Heavy, GPU energy consumption settled right into a decrease, extra secure vary of 35W to 46W, indicating that energy demand eased because the workload shifted away from the extra aggressive burst habits.

Thermal Abstract

Underneath load, the ZGX Nano G1n operates inside a tightly managed thermal envelope. Most system energy consumption is roughly 228W, and warmth dissipation is roughly 780 BTU/hr. Against this, idle energy draw stays low at roughly 36–38W, which signifies environment friendly energy scaling when the system is just not lively. The forced-air cooling answer maintains secure operation inside HP’s specified vary of 5°C to 30°C.

HP ZGX Nano AI Efficiency Testing

To judge the HP ZGX Nano with GB10, we examined Spark models utilizing the vLLM On-line Serving benchmark, probably the most extensively adopted high-throughput inference and serving engine for giant language fashions. The vLLM on-line serving benchmark simulates real-world manufacturing workloads by sending concurrent requests to a working vLLM server and measuring key metrics, together with complete token throughput (tokens per second), time to first token, and time per output token, throughout various load circumstances.

Our testing spanned a variety of fashions, together with dense architectures and micro-scaling information sorts, and evaluated efficiency throughout three workload situations: Equal ISL/OSL, Prefill Heavy, and Decode Heavy. These situations characterize distinct real-world serving patterns, from balanced enter and output hundreds to compute-intensive immediate processing and memory-bandwidth-bound token era.

Along with the HP ZGX Nano with GB10, we benchmarked different OEM techniques from Dell, ASUS, Acer, and Gigabyte. This allowed us to position HP’s outcomes inside the broader aggressive panorama and perceive the place it leads, retains tempo with the pack, or trails throughout completely different fashions and workloads.

GPT-OSS-120B

With GPT-OSS-120B, the HP ZGX Nano G1n posts its strongest ends in Prefill Heavy, the place throughput climbs from 304.5 tok/s at batch 1 to 2773.3 tok/s at batch 64. Equal ISL/OSL additionally scales steadily, rising from 69.6 tok/s to 722.9 tok/s throughout the sweep. Decode Heavy is far lighter by comparability, beginning at 183.7 tok/s in batch 1, dipping barely in batch 2, then recovering to 262.9 tok/s by batch 64.

GPT-OSS-20B

With GPT-OSS-20B, HP’s highest numbers come from Prefill Heavy, however the scaling is much less linear than with the opposite fashions. Prefill begins at 1626.6 tok/s at batch 1, climbs to 1980.3 tok/s at batch 2, drops sharply to 1120.3 tok/s at batch 4, then recovers to 4345.1 tok/s by batch 64. Equal ISL/OSL scales extra easily from 92.6 tok/s to 1550.6 tok/s, and Decode Heavy rises from 94.4 tok/s to 670.4 tok/s.

Qwen3 Coder 30B A3B FP8

For Qwen3 Coder 30B A3B (FP8), HP once more excels in Prefill Heavy, with throughput rising from 432.2 tok/s at batch dimension 1 to 2069.4 tok/s at batch dimension 64. Equal ISL/OSL rises from 104.2 tok/s to 1274.4 tok/s, whereas Decode Heavy improves from 55.9 tok/s to 480.4 tok/s. That is amongst HP’s stronger general outcomes.

Qwen3 Coder 30B A3B Base

On Qwen3 Coder 30B A3B (Base), HP delivers regular progress throughout all three phases, though the topline stays within the Prefill Heavy section. That section will increase from 258.6 tok/s at batch 1 to 1629.4 tok/s at batch 64. Equal ISL/OSL scales from 60.3 tok/s to 690.3 tok/s, whereas Decode Heavy rises from 33.0 tok/s to 331.8 tok/s.

Llama 3.1 8B Instruct FP4

With Llama-3.1-8B-Instruct (FP4), HP reveals a transparent step up in throughput. Equal ISL/OSL climbs from 76.4 tok/s at batch 1 to 2774.1 tok/s at batch 64, making it the strongest of HP’s three phases on this mannequin. Prefill Heavy additionally scales aggressively, rising from 316.8 tok/s to 2397.1 tok/s at batch 32 earlier than slipping to 2270.4 tok/s at batch 64. Decode Heavy will increase from 40.7 tok/s to 547.6 tok/s throughout the sweep.

Llama 3.1 8B Instruct (Base)

On Llama-3.1-8B-Instruct (Base), the HP ZGX Nano G1n scales cleanly throughout all three phases. In Equal ISL/OSL, throughput rises from 28.2 tok/s at batch 1 to 1298.6 tok/s at batch 64. In Prefill Heavy, HP will increase from 123.2 tok/s to 1759.5 tok/s, with features remaining sturdy all through the sweep earlier than tapering barely on the high finish. Decode Heavy is far lighter by comparability, rising from 15.5 tok/s at batch 1 to 366.4 tok/s at batch 64.

GPU Direct Storage

How GPU Direct Storage Works

Historically, when a GPU processes information from an NVMe drive, the info should first go by the CPU and system reminiscence earlier than reaching the GPU. This course of creates bottlenecks as a result of the CPU acts as a intermediary, including latency and consuming system sources. GPU Direct Storage eliminates this inefficiency by permitting the GPU to entry information immediately from the storage machine over the PCIe bus. This direct path reduces information motion overhead, enabling sooner, extra environment friendly transfers.

AI workloads, particularly these involving deep studying, are extremely data-intensive. Coaching giant neural networks requires processing terabytes of knowledge, and any delay in information switch results in underutilized GPUs and longer coaching occasions. Accordingly, GPU Direct Storage addresses this problem by delivering information to the GPU as rapidly as attainable, minimizing idle time and maximizing computational effectivity.

As well as, GDS advantages workloads that stream giant datasets, similar to video processing, pure language processing, and real-time inference. By decreasing CPU reliance, GDS accelerates information motion and frees CPU sources for different duties, additional enhancing general system efficiency.

GDSIO Learn Throughput 16K

Taking a look at GDSIO Learn Throughput 16K, the HP ZGX Nano G1n begins at 0.70GiB/s with 1 thread, inserting it among the many stronger low-thread performers within the group. It dips to 0.41GiB/s at 2 threads, then climbs again to 0.86GiB/s at 4 threads, displaying the identical small early-thread inconsistency seen in a number of of those techniques. From there, scaling turns into far more constant. Throughput rises to 1.6GiB/s at 8 threads and a pair of.2GiB/s at 16 threads, then continues upward to three.0GiB/s at 32 threads. On the greater queue depths, the HP retains gaining floor, reaching 3.9GiB/s at 64 threads and peaking at 4.6GiB/s at 128 threads.

GDSIO Learn Common Latency 16K

Taking a look at GDSIO Learn Common Latency (16K), the HP ZGX Nano G1n begins at roughly 0.02ms with 1 thread and stays low by 2 threads (0.08ms) and 4 threads (0.07ms). Latency edges up barely at 8 threads (0.08ms) and 16 threads (0.11ms), then will increase extra noticeably at 32 threads (0.16ms) and 64 threads (0.25ms). At 128 threads, latency reaches 0.42ms, nonetheless a bit beneath the best ends in the group whereas monitoring the system’s regular throughput scaling throughout the take a look at.

GDSIO Write Throughput 16K

Taking a look at GDSIO Write Throughput 16K, the HP ZGX Nano G1n begins at 0.84GiB/s on 1 thread, rises to 1.4GiB/s on 2 threads, and reaches 2.2GiB/s on 4 threads. Efficiency continues to scale strongly at 8 threads (3.0 GiB/s) and reaches 3.3GiB/s at 16 threads, the place it successfully ranges off. From there, throughput stays almost flat at 3.3GiB/s with 32 and 64 threads, then eases barely to three.2GiB/s with 128 threads, indicating the platform reaches its write ceiling comparatively early and sustains that degree persistently by the remainder of the sweep.

GDSIO Write Common Latency 16K

Taking a look at GDSIO Write Common Latency (16K), the HP ZGX Nano G1n begins at roughly 0.02ms with 1 thread and stays very low by 2 threads (0.02ms) and 4 threads (0.03ms). Latency rises modestly at 8 threads (0.04ms) and 16 threads (0.07ms), then jumps at 32 threads (0.15ms) and 64 threads (0.30ms). At 128 threads, latency reaches 0.61ms, nonetheless pretty nicely managed general, although the upward pattern aligns with the purpose the place write throughput has already flattened at greater thread counts.

GDSIO Learn Throughput 1M

Taking a look at GDSIO Learn Throughput 1M, the HP ZGX Nano G1n begins at 3.2GiB/s on 1 thread and rises to 4.1GiB/s on 2 threads. Efficiency continues to climb at 4 threads (5.2GiB/s) and eight threads (5.5GiB/s), after which the platform successfully reaches its ceiling. Throughput then holds basically flat at 5.5GiB/s for 16, 32, and 64 threads, earlier than easing barely to five.3 GiB/s at 128 threads, indicating a powerful early ramp adopted by a really secure high-thread plateau.

GDSIO Learn Common Latency 1M

Taking a look at GDSIO Learn Common Latency (1M), the HP ZGX Nano G1n begins at roughly 0.31ms with 1 thread and stays comparatively low at 2 threads (0.47ms) and 4 threads (0.76ms). Latency will increase with concurrency, rising to 1.4ms at 8 threads, 2.9ms at 16 threads, and 5.9ms at 32 threads. The pattern continues at 64 threads (12.8ms) and reaches 27.2ms at 128 threads, monitoring the upper queue depths although throughput had already flattened a lot earlier within the sweep.

GDSIO Write Throughput 1M

Taking a look at GDSIO Write Throughput 1M, the HP ZGX Nano G1n begins at 3.1GiB/s with 1 thread and rises to three.5GiB/s with 2 threads, then holds that degree at 4, 8, and 16 threads. Efficiency dips barely to three.3GiB/s at 32 threads earlier than returning to three.5GiB/s at 64 threads. At 128 threads, throughput will increase to three.7GiB/s, indicating a principally flat write profile throughout the sweep with solely minor variation and a small uptick on the highest thread rely.

GDSIO Write Common Latency 1M

Taking a look at GDSIO Write Common Latency (1M), the HP ZGX Nano G1n begins at roughly 0.31ms with 1 thread, rising to 0.57ms with 2 threads and 1.1ms with 4 threads. Latency continues to climb as concurrency will increase, reaching 2.2ms with 8 threads, 4.4ms with 16 threads, and 9.4ms with 32 threads. The upward pattern continues at 64 threads (17.7ms) and reaches 37.3ms at 128 threads, reflecting steadily rising queue strain although write throughput itself stays pretty flat by a lot of the sweep.

Conclusion

HP’s ZGX Nano G1n carries the DGX Spark platform’s anticipated efficiency profile and provides engineering selections that set it other than the opposite Spark techniques within the subject. In our testing, CPU temperatures peaked at 77.3°C and GPU temperatures at 69°C, each on the cooler facet of the Spark models we’ve benchmarked. vLLM efficiency was strongest in Prefill Heavy workloads throughout all six fashions we examined, with scaling that held cleanly by greater batch sizes. GPU Direct Storage learn throughput reached 4.6 GiB/s at 16K and 5.5 GiB/s at 1M block sizes, and write throughput plateaued early however held that degree persistently throughout the remaining thread counts.

The place the ZGX Nano G1n separates itself from the remainder of the Spark subject is within the work HP did across the reference design. The recycled-materials content material, the higher/lower-chassis cut up that improves inner serviceability, and the acoustic envelope that holds at 27.6 dBA beneath load all replicate deliberate engineering selections past what the GB10 platform itself requires. The safety stack follows the identical sample. TPM 2.0 in FIPS 140-2 mode, Frequent Standards EAL4+, and SED OPAL storage push this unit previous a developer equipment and towards a system that may clear procurement in regulated environments.

Like different Sparks, this isn’t a general-purpose workstation, and HP doesn’t place it as one. For builders, small groups, and organizations that want native AI compute with credible sustainability and safety tales behind the acquisition, the ZGX Nano G1n is a transparent differentiated choice inside the Spark lineup. For retailers the place these standards don’t apply, the underlying platform is the fixed throughout all 5 OEM techniques we’ve reviewed, and the choice comes all the way down to ecosystem, assist, and worth.

Product Web page – HP ZGX Nano G1n AI