Product1 distinct publisher3 min readUpdated
The scheduler change is throughput you get for the cost of an upgrade. The more durable news is that late-cycle patch volume is now permanently larger, and the maintainer has stopped fighting it.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
Linus Torvalds released Linux kernel 7.2 over the weekend, with Cache Aware Scheduling as the marquee addition [1][2]. For anyone packing many threads onto one socket, that is a throughput gain for the price of an upgrade; the more consequential line is elsewhere in his notes, where Torvalds describes AI-generated bug reports as a permanent condition of the release cycle rather than a spike [3].
The scheduler work is not new. The patch series has been in progress for several years [2]. Previously the load balancer spread threads across cores without checking whether threads on separate cores shared memory, so those cores re-fetched the same data and the system paid repeatedly for the same lines, a pattern known as cache bouncing [4]. Torvalds describes the goal as aggregating "tasks sharing data to the same LLC cache domain, thereby reducing cache bouncing and cache misses, and improving data access efficiency" [5]. The guardrails are what make this a multi-tenant feature rather than a benchmark result: the code avoids overloading a cache domain, and stops moving threads once a process consumes more than 25% of CPU time or more than one-third of a domain's load [6]. According to ZDNET, multithreaded servers, AI and ML compute jobs, and some games stand to benefit most [7].
The release note is blunter about process than about performance. Torvalds wrote that the last week of the release was "once again -- bigger than I would have wished for," adding that with "the whole 'new normal' thing," delaying releases for that reason would mean never releasing at all [8]. At release candidate seven he had already defined the term: "the new normal with a lot of fixes, many of them due to review by various AI tools" [3]. These are largely not features written by AI but patches for AI-discovered security vulnerabilities [9]. His objection is to how they are routed: "This is all entirely pointless churn, and we're making it clear that AI-detected bugs are pretty much by definition not secret, and treating them on some private list is a waste of time for everybody involved - and only makes that duplication worse because the reporters can't even see each other's reports" [10]. He is not banning the tools, telling contributors to "use them in a way that is productive and makes for a better experience" [11]. The operational reading is his own: the tail end of a cycle now routinely carries more fixes than the maintainer wants [8].
Elsewhere, exFAT moves to the kernel's iomap, which ZDNET reports improves read and write speed on SD cards and USB sticks by as much as 87% [12]. Experimental folios up to 2 MB arrive, and iomap conversion work continues for ext4 and XFS [13][14]. There is initial AMDGPU support for HDMI 2.1 Fixed Rate Link, enabling uncompressed 4K at 120Hz and 8K [15], Intel's USB4STREAM for raw host-to-host transfer over USB4 and Thunderbolt links that bypasses the networking stack [16], plus Wi-Fi Aware and early Wi-Fi 8 groundwork [17].
Two things to watch. Whether the security list handling for AI-detected bugs actually changes, since Torvalds has now said publicly that the private-list route makes duplication worse [10]. And whether the scheduler gain survives contact with real tenancy: most users will not see 7.2 until fall distributions such as Ubuntu 26.10, and the benefits are concentrated in machines with complex CPU-cache layouts and storage-heavy workloads [18][19].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Linus Torvalds announced the release of Linux kernel 7.2 over the weekend.
The marquee addition in Linux 7.2 is Cache Aware Scheduling, a scheduling change that has been in the works for several years.
In his note on Linux 7.2's seventh release candidate, Torvalds wrote: "I can't say that I'm exactly thrilled about the size of this all. But it is what it is: the new normal with a lot of fixes, many of them due to review by various AI tools."
Previously the scheduler balanced load across CPU cores without checking whether threads on separate cores shared memory, so threads on different cores would re-fetch the same data, causing cache bouncing in which the system wasted time and resources pulling and re-pulling the same data.
Torvalds on the patch series: "The goal of the patch series is to aggregate tasks sharing data to the same LLC cache domain, thereby reducing cache bouncing and cache misses, and improving data access efficiency."
The cache-aware scheduling code includes guardrails to avoid overloading a cache domain, and avoids further thread movement when a process already consumes more than 25% of CPU time or more than one-third of a domain's load.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Direct maintainer quotes, single outlet, no independent measurement
The factual spine is strong in kind: multiple verbatim Torvalds quotes from the -rc7 note and release announcement, plus specific implementation detail such as the 25% CPU-time and one-third domain-load guardrails. It is weak in breadth: one publisher, no upstream commit or mailing-list citation reproduced, and the only quantitative performance figure is unattributed.
Mainline release only; downstream availability still months out
Adoption evidence is limited to the upstream event itself: 7.2 is tagged and source is available. The same source says non-bleeding-edge users will not get these features until fall distributions such as Ubuntu 26.10, and recommends waiting for distro packaging and validation on production systems. No deployment, usage, or enterprise-uptake data is supplied.
Framing runs modestly ahead of measurement
The 'AI-enriched' framing and the up-to-87% throughput number are more emphatic than the underlying evidence: the AI story is fix and report churn rather than AI-authored kernel features, the scheduler benefit is asserted without any benchmark, and one graphics scheduling change was reverted. The gap is moderate rather than severe because the maintainer quotes are verbatim and the article itself qualifies hardware dependence and distro lag.
Vendor-neutral upstream primary voice, ad-supported single relay
The sourced claims originate with the kernel maintainer's own release and -rc notes, which carry no commercial sales incentive and include self-critical material about patch volume and a rollback. The distortion pressure sits on the relay side: a single ad-supported technology outlet publishing an 'everything new' feature roundup has traffic incentive to lead with an AI angle and a large percentage figure it does not attribute. No sponsored benchmark, vendor briefing, or funding relationship is disclosed in the supplied material.
Quotes are reliable, corroboration and measurement are missing
Confidence is moderate-low. The maintainer-quoted claims about patch volume and the scheduler goal are safe to rely on, and the release event itself is unambiguous. Everything performance-related is single-source and unmeasured, and there is no second publisher against which to check the feature list, so quantitative conclusions should be held loosely.
build
The chokepoint moved: ABF film, not lithography, now caps China's accelerator output1 distinct publisher
security
Intel's 72 CVEs land in firmware, drivers and the AI tooling stack; AMD adds a dozen1 distinct publisher
product
Callosum raises $100m for mixed-silicon scheduling, and the 2x accuracy claim is still its own2 distinct publishers
build
The kernel's Assisted-by trailer makes AI help a signed line, not a rumour1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026