ASUS puts local AI in a mini PC. What does a $3,000 premium actually buy?
The ASUS ProArt GR1X has reached retail, with a US 64GB listing at $4,499.99. In one independent coding test against another 128GB desktop, the pricier ASUS review unit generated output about 26% faster. A separate rendering test shows why that advantage does not carry across every application.
A roughly 26% lead in one coding benchmark came with a price gap of about $3,000. ASUS announced retail availability for the ProArt GR1X on November 18; its global press page carries a November 19 header and a November 18 dateline. The US Micro Center page lists the 64GB, 1TB version at $4,499.99 for in-store pickup only. Best Buy lists a 64GB, 2TB version at $4,999.99. In mainland China, IT Home reported a November 18 on-sale schedule and prices of RMB 34,999 for 64GB and 2TB, and RMB 54,999 for 128GB and 4TB. Those figures describe different configurations, not inventory at every store or proof of deliveries.
This is a new desktop form of a platform that reached Windows laptops in October, not the invention of on-device AI. AMD-powered compact desktops already gave shoppers the option of keeping sizable models in local memory. Tom’s Guide tested two machines with 128GB: its $6,999 ASUS review unit and a Geekom A9 Mega with an AMD chip that cost about $3,000 less. At that point the interesting question is how much less waiting you get from the more expensive machine, not which box is the first to fit the model.
ASUS makes the CPU and GPU share a pool of memory instead of making the model fit entirely within a separate graphics card’s memory. A machine that stays on the desk also has more leeway than a laptop. Windows Central put the desktop and Surface Laptop Ultra chip power limits at 140W and 80W respectively. Those are thermal design figures, not measured power at the outlet. In its Stable Diffusion XL test, the two 128GB machines were not equally quick: the ASUS took 20.6 seconds per image and the Surface took 25.05. However, Italian outlet DDay reported that its ASUS system was about 4% slower than the Surface in a native Blender test. The additional headroom does not help every application in the same way.
On the coding task Tom’s Guide ran with the same Qwen model in LM Studio, the ASUS produced 30.03 tokens per second compared with 23.77 from the Geekom, and took 0.55 seconds rather than 1.26 to start responding. That may matter to someone who repeats the workflow all day. It is not a measured 26% improvement in the time to finish a job, much less a guaranteed saving on cloud bills. Teams with repetitive local work and large creative files have the clearest reason to measure it for themselves. Occasional users would still be paying for a machine that spends much of its time idle. Neither the lowest retail price nor the fastest single benchmark describes that purchase on its own.
iHolyshit review
I would judge this as a second workstation for a studio, not a box that automatically replaces the household’s subscriptions. Over the next six to twelve months, I’ll look for independent comparisons that run the same model and production tasks for a month, tracking completed jobs, power drawn at the outlet, and how often the work still falls back to the cloud. A faster first reply is nice. What matters is how much more usable work the same budget buys by the end of the week.
Share this article
What do you think of this story?
Anonymous feedback, up to 5 reactions per article.
Reader comments
Comment without signing in, or sign in to display your member name.
Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.
No comments yet. Be the first to share.