2523.59

The 2523.59 release supports the following operating systems:

  • Red Hat Enterprise Linux (RHEL): 8.10, 9.4, 9.6, 9.7, 10.0, and 10.1

  • Ubuntu 22.04 and 24.04

  • SUSE Linux Enterprise (SLES): 15 SP4, 15 SP5, 15 SP6, and 15 SP7

Note

When using Intel® Extension for TensorFlow with this release, ensure that you are using Intel® oneAPI Base Toolkit version 2025.2.144 or earlier, as later versions are incompatible.

To install packages for this release, follow the Installing LTS 2523.x guide for your specific operating system.

Improvements

General

  • Incorporated the latest security updates to address recent vulnerabilities, enhance protection, and ensure greater system reliability.

  • Introduced support for new operating systems: Red Hat Enterprise Linux 9.7 and 10.1.

  • Updated the signing key for KMD prebuilds to enhance security and reliability. To continue receiving updates securely while using this or future releases, download and install a new Distinguished Encoding Rules (DER) certificate. Previous releases remain signed with the key that was current at the time of their release. Therefore, if you are using an older release from 2024 or 2025, the old key is still required and remains valid. In that case, use the 2024 or 2025 key, depending on the release you are using.

Intel® Graphics Driver Backports for Linux* OS (i915)

  • Introduced a fast GPU recovery mechanism for Open Accelerator Module (OAM) that restores GPU functionality after fatal errors without requiring a full system reboot or GPU reset, significantly reducing downtime. To enable this feature:

    1. Remove pci=noaer from the kernel command line.

    2. Add pcie_ports=native and i915.enable_fatal_error_recovery=1 to the kernel command line.

    3. After rebooting, verify the configuration by running cat /proc/cmdline to confirm the kernel parameters and cat /sys/kernel/debug/dri/*/i915_params/enable_fatal_error_recovery to ensure it returns 1.

  • Added backport support for kernel version 6.15 and 6.17.

  • Fixed VM_BIND wakeup to use full memory barrier preventing missed wakeups caused by local read reordering.

  • Fixed error state compression to always provide output buffer when flushing stream.

  • Fixed scatterlist marker restoration to prevent reading past end of list.

  • Added the local KOBJ_ATTR_RO() definition to enable compilation on DKMS targets.

  • Converted MFDI eye margin sysfs to use DEV_ATTR for better DKMS kernel integration.

  • Fixed ifdef mismatch for remap_sg() to ensure struct members are defined when used.

  • Fixed PMU events decoupling on device release to prevent use after free.

  • Fixed vma_set_flags() to hold required semaphore.

  • Added sysfs entries to track MDFI eye margin errors (mdfi_eye_margin_error and mdfi_eye_margin_status).

  • Fixed IAF error scanning to skip uninitialized subdevices.

  • Added monitoring for IAF bridge and viral error status registers with logging to dmesg and uevent.

  • Fixed VM_BIND object lifetime to prevent use after free during asynchronous operations.

  • Added verification that GuC successfully copied hwconfig data.

  • Added hwconfig storage initialization to prevent exposing stale kernel data to userspace.

  • Added backport support for kernel version 6.16.

  • Fixed BLT operations on Alchemist GPUs to use full 64KiB pages.

  • Added timer to clean dirty cache after device idles.

  • Fixed potential deadlock by removing object flushing from GT parking.

  • Added px-cache shrinking on GT parking to prevent local memory exhaustion and fragmentation.

Intel® Graphics Compiler for OpenCL™

  • Changed the implementation of fabs to correctly return 0.0 for fabs(-0.0).

  • Reduced Software Scoreboard (SWSB) compilation time for subroutines.

  • Fixed Software Scoreboard (SWSB) for function calls to properly handle dependencies.

  • Prevented crash in GASRetValuePropagator when DISubroutineType information is missing in metadata.

  • Added devel package with generic symlinks and headers.

  • Added built-in and intrinsic to set SR0 value for clearing FP exception state.

  • Masked OOB delta for intel sub group shuffle to align behavior with Khronos equivalent.

Intel® Graphics Compute Runtime for oneAPI Level Zero and OpenCL™ Driver

  • Fixed heap pointer allocation to retry with alternative freed chunks vector on failure.

  • Started using stateless built-ins for copyRegion.

  • Fixed an issue with copying SIP binary in the debugging mode 2.

  • Allowed IP version format in device_options and improved error messages.

  • Fixed the Level Zero tag update on memory copy with external host pointer to ensure temporary allocations are cleaned before the next copy operation.

  • Correctly reported Intel Data Center GPU Max Series ondemand paging capabilities.