Research window: 2026-11-17 05:00 ~ 2026-11-18 05:00 (Beijing time, 24 hours) Sources: Google News aggregation (24 query sets in Chinese and English, via proxy), Hacker News, Phoronix, CNX Software, LinuxGizmos, RISC-V International official website and blog, openRuyi official website and main repo, RISC-V member company website sweep (391 companies), riscv100.top leaderboard sweep, plus commit and release searches on GitHub across organizations and key repositories including XiangShan/NEMU/difftest/XSCache, CORE-V CVW/CVA6/core-v-verif, DAMO XuanTie, ESWIN, Sophgo, Nuclei, Espressif, SpacemiT and spacemit-robotics, Sipeed, revyos, openRuyi-Project, and Radxa Linkr.


In This Issue

  • Today’s Highlight: StarFive JH-B100 enters mass production and delivery; RISC-V server BMC reaches customer systems (11-17)
  • I. Open-Source Community and Software Ecosystem
    • openRuyi: nineteen commits — OpenSBI bumped to 1.9, OpenSSL 3.6.5 introduced (11-17)
    • revyos/linux reworks the 7.3.y config baseline, landing 20 platform patches in one go (11-18)
    • XiangShan Kunminghu: unified refactor of CSR and bundle naming (11-17)
    • CORE-V: two commits — RVFI CSR logging merged, CVW dependency cleanup reverted (11-17)
    • riscv.org blog: five-question interview during SC26 (11-17)
  • II. Product and Ecosystem Updates
    • Radxa Linkr releases v1.4.2 through v1.6.0 in a batch; plugin center and H.265 encoding in place (11-17)
    • Purism official press release confirms Librem 5 v2 specs and delivery terms (11-17)
  • III. Domestic Team Deep Dive
    • SpacemiT xslim: quantization stack pushes five commits on the agent branch (11-17)
    • Sophgo sophon-tools: se-rag-core adds tiered model config and bounded fallback (11-17)
  • IV. Embodied Intelligence and Robotics Special Section
    • SpacemiT robotics stack integrates Hiking perception-based locomotion policy; depth maps enter the policy loop via ZMQ (11-17)
    • manipulation and manifest wrap-up: motion-completion docs added, simulated camera output added to ignore list (11-17)
  • V. Summary and Trend Observations
    • Three-day trend: management plane lands first, software stack catches up on versions, robotics stack catches up on perception (11-15 ~ 11-17)
  • Supplementary Observation: riscv100.top Upcoming adds two previews (11-18)
  • Appendix: Full source list and source verification table

Issue Summary: Today’s only substantive news item is a single one, but it carries enough weight — StarFive announced that its server BMC chip, the JH-B100, has entered mass production and is shipping to customers, with server systems built around it about to enter mass production. This is the first time RISC-V has secured a mass-production delivery position in the data center management plane. On the open-source side, it was a day of “version catch-up”: openRuyi pushed OpenSBI to 1.9, introduced OpenSSL 3.6.5, and brought xgrammar and tvm-ffi into the repository; revyos reworked the 7.3.y kernel config baseline, arranging switches for six platforms in one pass. On the domestic team side, SpacemiT pushed five consecutive commits on the xslim quantization branch and integrated a third-party perception-based locomotion policy into its robotics stack; Sophgo added multi-backend configuration and network fallback to the RAG component of its SE series.


Today’s Focus: StarFive JH-B100 Reaches Mass Production and Delivery, RISC-V Server BMC Enters Customer Systems (11-17)

On November 17, StarFive announced that its RISC-V-based server BMC (Baseboard Management Controller) chip, the JH-B100 (昉·惊鸿-B100), has achieved mass production and been delivered to customers. The official statement is that the chip has passed product introduction and review processes at multiple server manufacturers and internet companies, and customer server products carrying it are about to enter mass production. Shipment volumes, customer names, and production schedules were not disclosed.

Specifications per StarFive’s official product page: a quad-core in-house RISC-V core (19,200 DMIPS) with a coherent network-on-chip (NoC), compatible with DDR4-3200 and DDR5-4800, integrating PFR (Platform Firmware Resilience) and PROT functions, supporting Chinese cryptographic standards SM2/SM3/SM4, secure boot, and DICE; interfaces include dual PCIe 4.0 endpoints, full-featured eSPI Slave, PECI, I2C/I3C, USB 3.2 Gen1, SGPIO, CAN-FD, IPMB, MCTP, and NCSI, with two built-in LTPI controllers to support dual-node management; the accompanying SDK follows the Yocto standard and can connect directly to OpenBMC. Beyond CPU server motherboards, applications also include compute trays, switch trays, coolant distribution units, and power shelves in AI infrastructure.

The significance of this news lies in the “position” rather than “performance.” The BMC does not participate in computing, but every server needs one, and this position has long been held by Arm and 8051-class controllers; previously, RISC-V had only sampling and validation at this position, with no mass production delivery. The logic for choosing it as a breakthrough point is also clear: performance requirements are far lower than the main CPU, the software stack (OpenBMC, Yocto) is mature, and the validation cost for customer introduction is manageable. The ecosystem endorsements StarFive cites are Intel and AMI—the former completed validation of Intel CPU management functions on the CRB platform, and the latter integrated the JH-B100 SDK into its BMC firmware product OneTree; both were disclosed previously at Intel Connection and the Dishui Lake Forum, and this issue marks the first time they have reached the “mass production delivery” stage.

Viewed with restraint, several points still warrant observation. First, “about to enter mass production” is the customer-side framing; actual volume depends on system manufacturers’ production schedules, and BMC procurement decisions often follow the main CPU platform. Second, StarFive did not provide shipment volumes, making it impossible to judge whether this is small-batch delivery or volume shipment. Third, the long-term test for RISC-V BMC lies in firmware maintenance and security updates; OpenBMC’s community support and CVE response speed will need to run through one or two product cycles before they can be assessed. On the same day, the MIIT-administered media and several industry media outlets republished this news, narrating it alongside the “15th Five-Year Plan” item on “advancing R&D and industrialization of the fifth-generation reduced instruction set,” a clear indication that the policy context and commercial progress are being tied together.

Date: 2026-11-17 Source: Component Network: JH-B100 Officially Delivered, IT Home: StarFive RISC-V Server BMC Chip Achieves Mass Production and Delivery, StarFive Official Website JH-B100 Product Page

I. Open-Source Community and Software Ecosystem

openRuyi’s Nineteen Commits: OpenSBI Bumped to 1.9, OpenSSL 3.6.5 Introduced (11-17)

The openRuyi main repository landed 19 package commits over the course of November 17. On the firmware side, OpenSBI was updated to 1.9; the crypto stack gained openssl-3 at version 3.6.5, with CI builds switched to prefer OpenSSL; base libraries bumped krb5 to 1.22.3 and fixed dnf5 builds under json-c 0.19; on the desktop side, openruyi-desktop-setup-kde received missing dependency declarations. A batch at 15:03 focused on Python and the AI inference stack: huggingface-hub 1.32.0, sentence-transformers 6.3.0, xgrammar 0.2.9 (flagged as a security fix), apache-tvm-ffi 0.1.12, click 8.5.1, hf-xet 1.5.2, plus a new rust-tokio-with-wasm-proc 0.8.

Three items deserve separate attention. OpenSBI 1.9 is a firmware baseline follow-up that determines whether new platforms can boot in M-mode; introducing OpenSSL 3.6.5 and making CI prefer it means the distro is shedding its 1.1-series compatibility baggage, at the cost of rebuilding every package that depends on it; and the arrival of xgrammar and tvm-ffi in the repository shows that AI inference dependencies like “structured output” and “TVM interfaces” have been folded into the distro’s scope rather than left to users to pip install themselves. The official news page’s latest items remain the two from October 19 (ecosystem infrastructure, Ruyi Firmware) and the 2026.09 release; no new version appeared within the window.

Date: 2026-11-17 Source: openRuyi-Project/openRuyi, openRuyi official news page

revyos/linux Overhauls 7.3.y Config Baseline, Lands 20 Platform Patches at Once (11-18)

At 04:13 on November 18, revyos/linux pushed 20 config commits to the revyos/7.3.y branch in a single batch—all 20 share an identical commit timestamp, with author dates reaching back to December 2025, indicating a bulk landing after a branch overhaul rather than work done overnight. The content enables options platform by platform: ESWIN EIC7700, UltraRISC DP1000, SpacemiT K3 and K1, T-Head TH1520, Sophgo SG2042/SG2044, plus a Picoheart errata switch and a set of generic options; it also adds cross/native build CI, a linuxboot config and debian_defconfig, and changes the package naming rule to use the major version number.

Aligning config files across six platforms at once shows that RevyOS has shifted from “a distro for one board” to “one kernel config covering multiple RISC-V product lines.” For downstream vendors, the value of such changes is less config work: when a new board arrives, first check whether the defconfig has a switch for your SoC—far faster than writing a dts from scratch. The risk lies in the same place—one branch simultaneously carries errata and hacks for six platforms (commits include workarounds forcing the xe driver onto DG1/DG2), so any one vendor’s change can affect another’s regression; adding cross and native build CI provides a safeguard.

Date: 2026-11-18 Source: revyos/linux (revyos/7.3.y)

XiangShan Kunminghu: Unified Refactor of CSR and Bundle Naming (11-17)

At 10:27 on November 17, the XiangShan main repository committed a refactor on the kunminghu-v3 branch by author Lemover, standardizing the naming of CSRs and related bundles. This is a pure refactor with no functional change, an interface-consistency cleanup after feature freeze.

Worth clarifying is the distinction between pushes and commits: the pushed timestamps of the XSCache, NEMU, difftest and GEM5 repositories all fall within the window, but their default-branch commit logs contain no in-window records, so these are judged to be pushes to other branches or tags (XSCache has an active dev-oceanus-draft branch series) and are not listed as separate items. Kunminghu V3’s recent activity (first a memory-access queue defect, now a naming refactor) sits on the maintainability line, unrelated to performance tuning.

Date: 2026-11-17 Source: OpenXiangShan/XiangShan

Two CORE-V Items: RVFI CSR Logging Merged, CVW Reverts Dependency Cleanup (11-17)

At 06:20 on November 17, openhwfoundation/core-v-verif merged PR #2675 (rvfi-csr-logging from ankit-cybertron), adding CSR logging to the RVFI interface. The same day, CVW first merged PR #2099 (clearing uv.lock dependency warnings and enabling uv updates), then reverted it entirely with a follow-up commit.

RVFI is the observation port for RISC-V formal verification and differential testing: it exposes each instruction’s PC, register reads/writes and memory accesses to the verifier, but CSR state was previously absent, so privileged-mode behavior could only be compared indirectly. With CSR logging, per-instruction changes to registers like mstatus and mtvec can be checked directly, which is especially useful for verifying exception and interrupt paths. The CVW revert is a different kind of signal—dependency-warning cleanup looks harmless, but once lock files and update policies change, regression-environment reproducibility suffers, so a revert is preferable to proceeding with uncertainty.

Date: 2026-11-17 Source: core-v-verif PR #2675, openhwfoundation/cvw

riscv.org Blog: Five-Questions Interview During SC26 (11-17)

At 11:40 on November 17, RISC-V International published a new “V Questions with…” installment featuring Doug Norton, CMO of Inspire Semiconductor and chair of the HPC-AI Society, and Nick Brown, senior research fellow at EPCC (the UK national supercomputing center at the University of Edinburgh). The backdrop is SC26 taking place in Chicago that week; the article notes the two could not meet on the show floor, so the interview was conducted online, with an AI-generated avatar collage as the image.

From the page content retrieved, this is a personality interview rather than a product announcement, with themes centered on RISC-V’s HPC software stack and ecosystem. It belongs in the daily briefing not because “there’s something new” but because of the narrative framing: during supercomputing’s annual conference, the RISC-V side is still offering people and ecosystem rather than peak-FLOP figures. This contrasts with domestic vendors’ concrete moves this month in BMC, robotics and distros—both sides are avoiding a head-on compute comparison.

Date: 2026-11-17 Source: riscv.org: V Questions with Doug Norton and Nick Brown

II. Product and Ecosystem Updates

Radxa Linkr releases v1.4.2 through v1.6.0 in a batch, with plugin center and H.265 encoding in place (11-17)

Radxa’s browser-based remote management product line Linkr released four versions in a batch at around 15:2x on November 17: v1.4.2, v1.5.0, v1.5.2, and v1.6.0, with all four releases published less than five minutes apart; on the same day, the harness-linkr repository was also newly created with an Initial commit. Four versions appearing on the same day looks more like historical versions being backfilled all at once after the repository went public, and should not be read as four releases in a single day.

According to the release notes, v1.6.0 adds H.265 encoding, custom HTTPS certificates, bidirectional clipboard sync (requiring installation of the executable bundled with the Linkrd plugin on the controlled machine), a 720p 180fps EDID preset, and a watermark toggle; v1.5.0 introduces a plugin center (upload, install, start/stop, uninstall), two networking plugins for ZeroTier and EasyTier, screenshot OCR, experimental IPv6, and adds toggle control to the Agent API; v1.5.2 supports double-tapping the FN key to toggle the Wi-Fi hotspot, automatically selecting 5GHz or 2.4GHz based on the connected band, and fixes the inability to switch resolution when a Radxa Dragon Q8B is used as a video input source; v1.4.2 adds audio to H.264 WebSocket mode, disables Wi-Fi by default to reduce power consumption, and adds experimental touch input.

Taken together, Linkr’s direction is “orchestratable out-of-band management”: the plugin center handles feature extension, the Agent API handles external invocation, the networking plugins handle cross-subnet access, and OCR and clipboard handle human-in-the-loop operations. The release notes do not disclose the hardware form factor or the SoC used, and this report makes no judgment on its architecture.

Date: 2026-11-17 Source: radxa-linkr/linkr releases, radxa-linkr/harness-linkr

Purism official press release confirms Librem 5 v2 specs and delivery terms (11-17)

Purism published an official press release on November 17 confirming that the Librem 5 v2 uses an octa-core RISC-V processor, runs PureOS, and is priced at $1,599 for crowdfunding, and giving delivery terms: approximately 14 to 15 months after the crowdfunding campaign ends, with rewards and timelines not guaranteed because development is still ongoing. The press release also clarifies the engineering scope: graphics, peripherals, power management, and application compatibility all need to be completed. CTO Jonathon Hall’s statement was that “RISC-V is the future, and Librem 5 v2 is the product that brings this to mainstream users.”

This was already covered by this report when the crowdfunding campaign launched on November 14 (octa-core SpacemiT K1, 5.7-inch 720×1440, 8GB RAM, 128GB storage, 4500mAh replaceable battery, dual SIM, hardware kill switches); this issue only adds the two new points from the official post: first, the delivery formula is refined from “2028” to “14 to 15 months after the crowdfunding campaign ends”; second, the official post for the first time lists “ARM64 application compatibility” among the items that must be tackled. Follow-up coverage from the same batch of foreign media (OMG! Ubuntu, EIN News, Andro-News, Notebookcheck) is consistent with the press release and is not counted as separate items.

Date: 2026-11-17 Source: Purism: Purism Launches Librem 5 v2

III. Deep Dive into Domestic Teams

SpacemiT xslim: Quantization Stack Pushes Five Commits in a Row on agent Branch (11-17)

SpacemiT’s model compression toolchain xslim pushed five commits in a row on November 17 on the agent/feat/operator-layer-quantization branch (from 08:46 to 19:35), all centered on quantization strategy: selecting per-layer calibration methods based on user data, parsing and validating a unified precision plan, validating fusion provenance and effective precision boundaries, budgeted automatic mixed precision, and task-metric search with controlled local fp16. Another branch, agent/chore/add-agent-harness, added a repo-level agent harness and contribution guide on November 13.

The direction of these commits is “turning quantization from hand-written configuration into search”: given a precision budget and a task metric, let the tool decide which layers use which precision, rather than manually filling in a table for each layer. At the same time, provenance and fused validation appear repeatedly, indicating they have been burned by fused-operator precision mismatches on real models—a quantized convolution gets fused by the backend, and the precision actually in effect is not the same as what the config file says; without validation, you get a silent discrepancy where “the config says int8 but it’s running fp16.” The naming of the agent branch is also worth noting: the toolchain is starting to organize code and skills around “for use by automated agents,” rather than only for humans.

On the same day, spacemit-com/onnxruntime merged an upstream change (selecting model package variants by EP priority); this is upstream synchronization and does not involve RISC-V-specific changes. The xslim main branch had no commits during the window; all progress is on feature branches.

Date: 2026-11-17 Source: spacemit-com/xslim (agent/feat/operator-layer-quantization), spacemit-com/onnxruntime

Sophgo sophon-tools: se-rag-core Adds Layered Model Configuration and Bounded Fallback (11-17)

Sophgo’s sophon-tools submitted a feat(se-rag-core) commit on November 17 at 13:30: layered OpenAI configuration, plus CF-first with bounded fallback for the built-in gateway (commit IDs MYS-2924 / MYS-292x). The changes are concentrated in the RAG component’s model access layer—splitting the OpenAI-compatible configuration into multiple layers (default, environment, deployment level) and merging them by priority; the gateway portion preferentially uses the CF (Cloudflare) route and falls back a limited number of times after failure.

This is not technically sophisticated, but the direction is clear: the LLM application stack on the SE series (SG2042 / SG2044 servers) is moving toward operability. Layered configuration solves the problem of “the same codebase using different model endpoints in development, testing, and at customer sites”; CF-first with bounded fallback solves cross-border access stability, and the number of fallbacks must be bounded—otherwise a single failure will send requests to all backup routes. It should be noted that this is an application-layer tool, not a chip or driver; this report includes it because it reflects that what Sophgo is supplementing beyond server hardware is the “out-of-the-box usable LLM service” segment.

Date: 2026-11-17 Source: sophgo/sophon-tools

IV. Special Chapter on Embodied Intelligence and Robotics

SpacemiT Robotics Stack Integrates Hiking Perceptive Locomotion Policy, Depth Maps Enter Policy Loop via ZMQ (11-17)

At 16:04–16:05 on November 17, two repositories under spacemit-robotics each merged one commit: PR #33 “support Hiking perceptive locomotion” in humanoid_common and PR #20 “add Hiking perceptive control simulation” in humanoid_unitree_g1, both authored by chenzhaoqi. The humanoid_common change spans 16 files with 938 lines added, covering: a Hiking policy adaptation layer (depth encoding, observation history, joint mapping, model checksum), publishing MuJoCo camera depth via ZMQ using offscreen EGL rendering, validating depth metadata, timestamps, and history length before entering RL, support for online RL handover (action history initialization, readiness checks, preparation feedback), retaining the current policy on handover failure or timeout, and adding perceptive build dependencies with rebuild of consumers when RL headers change.

Judging from the commit scope, this interface corresponds to the publicly available Hiking perceptive locomotion policy (Hiking in the Wild, Tsinghua IIIS and Shanghai Qi Zhi Institute, Unitree G1 with a front-facing depth camera, depth at 60Hz, policy at 50Hz); the official source has not stated the exact origin of the adaptation target, so this is an inference based on the commit contents. Three engineering points worth noting: retaining the old policy on handover failure is a safety fallback, preventing the robot from losing control at the moment of policy switching; the model checksum prevents policy files from being replaced or version-mismatched; and rebuild-on-header-change exists because they hit the C++ struct layout mismatch pitfall—the commit notes state that staged installs preserve source file timestamps, so when RL headers change but cached objects are not recompiled, the linked struct layouts do not match.

Making third-party policies into a replaceable adaptation layer is a necessary step for a robotics stack moving toward a “policy ecosystem”: a chip vendor cannot train all policies itself; what it can do is standardize the three segments of perception input, policy inference, and action handover, so that externally trained policies can run as soon as they are plugged in. Platforms like K3 provide vector and multi-core capabilities at this position, along with headroom for running MuJoCo and depth rendering.

Date: 2026-11-17 Source: humanoid_common PR #33, humanoid_unitree_g1 PR #20

manipulation and manifest Wrap-Up: motion-completion Docs Added, Simulated Camera Output Added to Ignore List (11-17)

On the same day, the manipulation and manifest repositories under spacemit-robotics landed six commits in total. On the manipulation side: PR #4 merged motion-completion related documentation, and PR #5 and PR #6 added MuJoCo logs and simulated camera frame directories to the ignore list respectively; on the manifest side, PR #33 (refactor/finish-manipulation) was merged.

These are all housekeeping changes, but the order is worth noting: adding docs first and then cleaning up artifacts is usually the action of “closing out a piece of work, preparing for handover or external release.” For the manipulation direction, the motion-completion documentation shows that what they care about is not single-grasp success but the completion and continuity of action sequences; excluding simulated camera frames from the version repository is because such data runs to tens of GB per day, and a single mistaken commit would blow up the repository.

Date: 2026-11-17 Source: spacemit-robotics/manipulation PR #4, spacemit-robotics/manifest

V. Summary and Trend Observations

Today there is only one main thread: the management plane lands first. The mass-production delivery of StarFive JH-B100 marks the first time RISC-V has secured a “mass-production delivery” position in the data center rather than “sample validation.” The BMC does not participate in computing, but every server needs one, and volume scales with the whole machine; its performance requirements are far lower than the main CPU, the software stack is backed by OpenBMC and Yocto, customer onboarding validation costs are manageable, and the supply chain has long offered only two choices: Arm and 8051. Breaking through at this position is far more realistic than a head-on collision on the server main CPU.

The second judgment is that the software side today is “catching up on versions” rather than “shipping new things.” openRuyi pushes OpenSBI to 1.9, introduces OpenSSL 3.6.5, and brings xgrammar and tvm-ffi into the repository; revyos reorganizes the kernel configuration baseline, arranging the switches for six platforms in one pass. Neither will make headlines, but they determine whether customers can avoid tweaking configurations and directly use existing inference stacks after receiving a board. A large part of RISC-V’s software gap hides in this kind of daily alignment.

The third judgment is that progress on the robotics side lands in the “adaptation layer.” SpacemiT plugs the Hiking perceptive policy into humanoid_common, using ZMQ to transmit depth maps, checksums to validate models, and retaining the old policy on handover failure—these are engineering details, not paper metrics, but they determine whether third-party policies can run on K3. A chip vendor cannot do all policies; what it can do is standardize the interface, and today there was concrete action on this.

Three-Day Trend: Management Plane Lands First, Software Stack Catches Up on Versions, Robotics Stack Adds Perception (11-15 ~ 11-17)

Looking at November 15 through 17 together, three lines each advanced. On the whole-machine and chip side, Purism launched the Librem 5 v2 crowdfunding on November 14 and issued an official press release on November 17 refining the delivery timeline to “14 to 15 months after the crowdfunding ends”; StarFive pushed the RISC-V BMC to mass-production delivery on November 17. Both add weight to “RISC-V entering the whole-machine BOM”—one is a consumer phone, the other a data center server management chip.

On the open-source core and verification side, FURSOY on November 15 unveiled an out-of-order RV64 core capable of booting Linux 6.6.50; on November 16 XiangShan was closing out memory access queue defects while CVW was filling in vector memory access; on November 17 core-v-verif brought CSR state into the RVFI log while XiangShan did naming refactoring. What these four share is a shift from “it runs” to “it is verifiable and maintainable,” which is also the precondition for open-source cores to enter product view.

On the software and toolchain side, openRuyi spent two consecutive days on version tracking and configuration alignment, while SpacemiT worked on automated search in the quantization toolchain and policy adaptation in the robotics stack. Taken together, the window for architectural narrative is narrowing and the window for engineering delivery is opening: over the next six months, what determines RISC-V’s position will not be another new chip launch event, but whether this kind of delivery, configuration, and adaptation can be sustained.

Supplementary Observations

riscv100.top: Two New Upcoming Previews (11-18)

This round of inspection of riscv100.top was judged as “no major changes (minor updates only).” Two new previews were added to Upcoming: SpacemiT K3 X100 (Vectorization/IME) paired with XSCC’s AI optimization technology; and LineShine (CPU-Only TOP500 Leader). Under the cross-architecture and RISC-V channels, the six SPEC CPU2006/2017/2026 leaderboards show no new records and no score changes, and there were no additions or removals of committee members.

It should be noted that Upcoming is a preview slot, not a result. Both previews point to new data to come from the “K3 plus XSCC compiler” combination, and the Vectorization/IME evaluation is especially worth watching—public data under third-party leaderboard standards has long been lacking for the actual gains of vector and matrix extensions. If corresponding results appear in the next inspection, they will be handled as leaderboard entries.

Date: 2026-11-18 Source: riscv100.top

Appendix

Source Verification Table

Source Verification Result
Google News aggregation (24 query sets in Chinese and English, plus additional Chinese team terms and vendor English queries) Within the window, the only substantive new item was the StarFive JH-B100 mass production delivery, which has been included as today’s lead. Excluded items: Purism’s November 17 official press release on Librem 5 v2 and multiple follow-ups from foreign media are a continuation of the November 14 crowdfunding event; this issue only adds the delivery formula and ARM64 compatibility wording newly introduced in the official release; the November 17 reprints of Lingrui Zhixin’s Pre-A+ funding (RFID World, Pintu) are in fact the November 13 event, already noted as not to be re-included in the November 16 report; Alibaba’s November 17 re-publication of the T-Head spin-off and Hong Kong listing is the November 10 event, and its subject is PPU and the Yitian line, so it is not a RISC-V item; the November 17 Chinese reprint of FURSOY is the November 15 event, already included in the November 16 report; Ingenic’s core wording, the Beijing integrated circuit forum preview, and the WIM2026 guest confirmation are all company profiles or event previews and are not included
Hacker News / Phoronix / CNX Software / LinuxGizmos / RISC-V International Two items on HN within the window: Purism Librem 5 v2 (55 points, 43 comments) is an already-included event; the other is a paper on separation-logic verification of the xv6 kernel on RISC-V using AI agents (1 point, 0 comments), which lacks sufficient traction and is a preprint, so it is not listed separately; Phoronix and CNX Software had no RISC-V-specific new articles within the window; LinuxGizmos is the same as November 16; the riscv.org blog added a five-questions interview on November 17, which has been included, and its news page had no additions
GitHub organization and key repository checks (30+ organizations and repositories) Active within the window: XiangShan main repo 1 commit (kunminghu-v3 branch), openRuyi-Project main repo 19 commits, revyos/linux 20 commits (batch landing of branch restructuring), core-v-verif 1 commit, cvw 2 commits, spacemit-com 2 repos (xslim feature branch 5 commits, onnxruntime upstream sync 1 commit), spacemit-robotics 5 repos (humanoid_common 2 commits, humanoid_unitree_g1 2 commits, manipulation 5 commits, manifest 1 commit, realtime-link no commits within the window), sophgo 4 repos (sophon-tools 1 commit, plus pushes to trusted-firmware-a, build, and edk2-non-osi with no commits on the default branch within the window), sipeed 3 repos (NanoKVM 1 commit for UI style unification), radxa-linkr 2 repos (four linkr releases, harness-linkr initialization), espressif 29 repos (all esp-idf upstream syncs, with no direct RISC-V items). DAMO XuanTie (XUANTIE-RV), ESWIN (eswincomputing), and Nuclei (Nuclei-Software) had zero pushes within the window, and no new articles on the public side. The anonymous API did not hit rate limits this issue; organization activity is based mainly on each repository’s Atom commit feed, and pushes to non-default branches may be undercounted
openRuyi dual sources The latest on the official news page is two articles from October 19, 2026 (ecosystem infrastructure, Ruyi Firmware) and the 2026.09 release; there was no new version within the window; the 19 commits in the main repo have been included per the logs
riscv100.top inspection No major changes (only minor updates): two new preview entries under Upcoming, no new records or score changes across the six leaderboards, and no additions or removals of committee members; placed as supplementary observation toward the end of the main text per user request
RISC-V member website inspection (391 sites) 53 sites reported content changes, 73 failed to fetch, and the rest were unchanged. Spot checks of the changed items were all homepage marketing rotations (Aalto, Bologna, CMCF, CNCF, Ubitium, etc.), with no new products, funding, or architecture announcements, so they are not included
Exclusion checks The one Sipeed NanoKVM commit within the window is frontend style unification, not a RISC-V kernel or firmware change; Espressif’s 29 repo pushes are concentrated in esp-idf and the ESP32-C series Bluetooth libraries, part of the regular release cadence, with no new chips this issue; Radxa Linkr’s four versions released on the same day are judged to be backfilled records after publication and are not interpreted as “four versions in one day”

Complete Source List

Limitations: This issue covers a 24-hour window (November 17, 05:00 to November 18, 05:00). On the media side there was only one vendor press release; the main content comes from GitHub commit logs and official release notes. Technical details are based on commit titles, release notes, and repository documentation, and the code was not verified line by line. StarFive did not disclose JH-B100 shipment volumes, customer names, or the specific timing of customer system mass production; BMC specifications are taken from the official product page and the press release, and actual configurations are subject to the customer motherboard. “About to enter mass production” is the customer-side wording and may be delayed. The 20 revyos commits share the same commit time but span nearly a year in author dates, so they are treated as the conclusion of branch restructuring; whether they contain undisclosed upstream patches cannot be confirmed. The five xslim commits are on a feature branch, and whether they will enter the mainline is not yet determined. The judgment of the Hiking adaptation target is based on the correspondence between the commit content and the wording of public papers; the official source did not name it. Radxa Linkr’s four versions released on the same day are treated as backfilled records; another possibility is a concentrated release after a repository migration, which requires further observation. The 53 “content changes” in the member website inspection were determined by page text hash; after reading each one, all were marketing updates, but there is a possibility that individual announcements were misjudged. The local search backend hit quota limits during collection, and some queries were completed via backup channels; the dates and links of included items were all rechecked against the original sites.