↩ Accueil

Vue lecture

Adjusting One Line Of Linux Code Yields 5x Wakeup Latency Reduction For Modern Xeon CPUs

A new patch posted to the Linux kernel mailing list aims to address the high wake-up latency experienced on modern Intel Xeon server platforms. With Sapphire Rapids and newer, "excessive" wakeup latencies with the Linux menu governor and NOHZ_FULL configuration can negatively impair Xeon CPUs for latency-sensitive workloads but a 16 line patch aims to better improve the situation. That is, changing one line of actual code and the rest being code comments...
  •  

New Linux Patch Improved NVMe Performance +15% With CPU Cluster-Aware Handling

Intel Linux engineers have been working on enhancing the NVMe storage performance with today's high core count processors. Due to situations where multiple CPUs could end up sharing the same NVMe IRQ(s), performance penalties can arise if the IRQ affinity and the CPU's cluster do not align. There is a pending patch to address this situation. A 15% performance improvement was reported with the pending patch...
  •  

Linux 6.19 ATA Fixes Address Power Management Regression For The Past Year

It's typically rare these days for the ATA subsystem updates in the Linux kernel to contain anything really noteworthy. But today some important fixes were merged for the ATA code to deal with a reported power management regression affecting the past number of Linux kernel releases over the last year. ATAPI devices with dummy ports weren't hitting their low-power state and in turn preventing the CPU from reaching low-power C-states but thankfully that is now resolved with this code...
  •  

AMD Making It Easier To Install vLLM For ROCm

Deploying vLLM for LLM inference and serving on NVIDIA hardware can be as easy as pip3 install vllm. Beautifully simple just as many of the AI/LLM Python libraries can deploy straight-away and typically "just work" on NVIDIA. Running vLLM atop AMD Radeon/Instinct hardware though has traditionally meant either compiling vLLM from source yourself or AMD's recommended approach of using Docker containers that contain pre-built versions of vLLM. Finally there is now a blessed Python wheel for making it easier to install vLLM without Docker and leveraging ROCm...
  •  

LLVM Adopts "Human In The Loop" Policy For AI/Tool-Assisted Contributions

Following recent discussions over AI contributions to the LLVM open-source compiler project, they have come to an agreement on allowing AI/tool-assisted contributions but that there must be a human involved that is first looking over the code before opening any pull request and similar. Strictly AI-driven contributions without any human vetting will not be permitted...
  •  
❌