First impressions: ASUS ProArt RTX Spark PCs, and the 128 GB number I cannot stop chewing on
Most of what I see coming out of IFA 2026 is last year’s laptop with a different sticker. Then ASUS put up the ProArt P16, the ProArt P14 and a small box called the ProArt GR1X, and the CPU in all three carries an NVIDIA name. Grace does the general work, Blackwell RTX takes care of the graphics, wired together over NVLink-C2C.
The pitch underneath the branding is memory. Up to 128 GB of it, one pool, either half of the chip free to reach into it, with 6,144 CUDA cores and as many as 20 CPU cores quoted alongside up to 1 petaflop of FP4 performance. ASUS describes the P14 as the world’s first Windows PC purpose-built for personal agents, which is a mouthful, though I understand what the marketing is reaching for.
So I’m curious and a bit wary at once. A ProArt laptop giving up its x86 chip isn’t a small move, whatever the spec table says.
The memory pool is the whole argument
Last year’s ProArt P16 went out the door with an RTX 5090 Laptop GPU and up to 24 GB of GDDR7. The one ASUS presented at IFA 2026 claims 128 GB, shared.
ASUS reckons the P14 can run a 120-billion-parameter model with up to a million tokens of context. The 128 GB configuration also gives 3D artists room for scenes above 90 GB, with OptiX handling the workload. It can decode 12K 4:2:2 footage, too. I haven’t seen enough projects to know how well ASUS’ numbers hold up, but 128 GB is a very different proposition from 24 GB.
Somehow the chassis got thinner while all of that happened. ProArt P16 comes in at 12.9 mm and 1.77 kg with a 99 Wh battery, which ASUS puts at 14% thinner and 9% lighter than the outgoing model. P14 is lighter still, 1.48 kg, 13.9 mm, 90 Wh. Both panels are Lumina Pro OLED, with Delta E under 1, up to 1,600 nits, 120 Hz with VRR, and an anti-reflection coating on top.
GR1X is the one I’d put on my own desk. Palm-sized, 150 x 150 x 51 mm, cooled by a dual-fan stack ASUS counts at 218 blades with seven levels of fan control, and rated for running flat out around the clock. It has 10 GbE, Wi-Fi 7, and drives four 4K displays. Acer brought its own RTX Spark small-form-factor machine to IFA as a design with nothing pinned down, so ASUS at least turned up with measurements.
MuseTree puts FLUX.2 on the local machine, so you can use it without a token meter running in the background. StoryCube is aimed at the media you accumulate along the way. ASUS is showing validated ComfyUI workflows beside its Zenni Claw agent.
Where I get twitchy
Grace is Arm. I’d worry about the plugin that sits inside a studio’s main app and hasn’t seen an update in years. If it breaks on Arm, the rest of the machine’s performance doesn’t matter much. ASUS said nothing about emulation.
Soldered unified memory also means the configuration you buy is the configuration you keep, forever, no second chance at 128 GB in year three.
And every figure above came out of ASUS or NVIDIA marketing with an “up to” attached, measured somewhere I cannot see. GR1X doesn’t have a launch date, only a notify-me form, which tells me the mini PC trails the laptops. Worth watching if you keep slamming into a VRAM ceiling, worth sitting out if your day depends on obscure Windows software.
128 GB has a narrow audience
The local-model crowd has a pretty obvious reason to care. If you’re running a model at home, a long fine-tune or agent job can tie up a machine for hours. ComfyUI jobs are another case where having 128 GB of shared memory changes what you can keep loaded at once. For 3D work, the P16 makes more sense if your scene files are already pushing past 90 GB.
Video editors, less so. If your RTX 50 Series laptop already handles your Premiere and Resolve jobs, I don’t see much reason to gamble on compatibility just for a different architecture.
Related: Best Mac for video editing in 2026, from the $899 Mac mini to the M5 Ultra Mac Studio
My take for now
While I like the FLUX.2 angle, I’m holding off on the hardware verdict because nobody has had time with the machines outside a hotel showroom in Berlin. Running FLUX.2 on-device without a token limit is a solid reason to keep AI work on the machine.
I want to see the shared memory pool under load. Give me an hour-long render that stays cool and a plugin-heavy Premiere project that works without drama, and I’ll be convinced.
