AMD's $4,000 AI dev kit runs a chip that's been on sale since spring
LTT Labs reviewed AMD's 'Ryzen AI Halo,' a roughly $4,000 AI developer box, then conceded in its own text that the chip inside offers nothing new.
The one thing actually new in the box is a set of AMD developer 'playbooks,' positioned as an answer to Nvidia's DGX Spark documentation.
Owners are blunter than the review: at the same price as a Spark, just buy the Spark, which is faster with better software.
A year-old chip in a $4,000 box
LTT Labs put its byline on the "AMD Ryzen AI Halo," a developer machine built on the Ryzen AI Max+ 395 (Strix Halo) with 128GB of unified memory. Then the review undercut its own subject. A reader lifted the admission straight from the text:
The AMD Ryzen AI Max+ 395 (Strix Halo) processor has been available since Spring 2025 and the Halo doesn't offer anything new on that front. It has the same 256 GB/s memory bandwidth limit as every board previously… you can go get a Framework Desktop for roughly the same price or a GMKtec EVO-X2 for a bit cheaper.
— a reader quoting the review directly
"Just buy a Spark," say the owners
At this price the box no longer undercuts Nvidia's DGX Spark, which runs about $4,700. An owner of the 395 hardware did the math:
These devices were great when they were cheaper than the DGX Spark. But when they cost the same price… there's no reason to buy this over a Spark. The Spark is literally a faster version of this, with better software support.
— an owner of a Ryzen AI Max 395 box
The hardware's identical; only the docs are new
If anything justifies the launch, it's the playbooks. An engineer who runs these boards named the addition and priced the rest:
The one thing that's new/worth pointing out are the https://developer.amd.com/playbooks/ … this is AMD's answer to Nvidia's playbooks… Hardware is the exact same as what used to be available for $2K last year (and is still $1K cheaper from Chinese OEMs).
— an engineer who runs these boards
AMD's software is the real liability
Those playbooks lead straight into AMD's weakest flank. A commenter who tracks amdgpu kernel regressions was blunt:
it's worth noting that AMD's software is universally weak and not worth any degree of reliance… every few months they merge a serious regression into amdgpu and sometimes even backport it into stable… just a few weeks ago they backported a null deference kernel oops into stable, it's still not fixed.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
A Claude trained to sabotage code betrayed its intent in a hidden "J-space"
On July 6, 2026, Anthropic reported that Claude contains an internal region it calls J-space, uncovered with a "Jacobian lens," that behaves like the brain's global workspace.