IT之家 9 月 26 日消息,VideoCardz 今日发现,英特尔为旗下免费开源的游戏表现监测工具 PresentMon 发布了最新的 2.6.0 版本。
新版本重点优化了其自身资源占用,官方称高负载下 CPU 占用降低 78%,同时新增“Game Experience(游戏感受)”监测覆盖层,用于展示与画面流畅度和动画表现相关的指标。

PresentMon 2.6.0 兼容 Windows 10、Windows 11,以及 DirectX 9、DirectX 11、DirectX 12、OpenGL 和 Vulkan 应用,兼容 AMD 和 NVIDIA 显卡,兼容 ADL、NVAPI、NVML、IGCL 以及 WMI 等多种遥测接口。
新版最大的变化来自后台服务优化。英特尔调整了 ETW(IT之家注:Event Tracing for Windows,Windows 事件跟踪)数据刷新机制,将部分无需高精度计时的后台循环改为更低频率的 Sleep 等待方式,使高负载下 CPU 占用减少 78%,移动平台同样适用。
另外,当软件未主动监测或轮询遥测数据时,诊断日志刷新将被暂停,待机 CPU 占用率也从约 0.003% 降至接近 0%。
面向普通玩家,PresentMon 新增 Game Experience 覆盖层,集中展示英特尔认为与动态画面观感、动画平滑度相关的表现指标。

2.6.0 界面也开展了改进。此前无法使用的监测指标会直接消失,现在则会保留在列表中并以灰色表明,同时提示不可用原因,例如当前硬件不兼容、API 未给出或 PresentMon 当前版本尚未兼容。
在多显卡环境下,使用者还能同时表明多个显卡的遥测数据,并针对每项指标分别挑选不同设备的数据来源。

对于游戏实测人员和开发者,PresentMon 2.6.0 增加了多项分析能力。其中涵盖对 DirectX 12 PSO(Pipeline State Object)编译过程的监测,可记录编译次数、耗时以及编译繁忙占比,并写入帧数据和 CSV 输出。
新版还兼容更新的 Windows DDI Flip 事件,可记录每次 Flip 的时间戳、已完成和丢弃的 Flip 事件,同时为 CSV 输出增加 Source ID、Layer ID 等表明元数据,更方便分析多屏、多平面表明情形。
GPU 遥测维度,英特尔改进了借助 AMD ADL 接口统计专用显存占用的准确性,并调整遥测架构,仅轮询客户端实际请求的数据,从而进一步降低平台开销。
新版还修复了多项稳定性难题,涵盖自动目标识别失败、服务重启异常、多实例启动冲突,以及内部 API 兼容性等难题。同时,实验性的 Flash Injector 能力已被移除,以简化项目维护和构建流程。
本文来源:https://www.ithome.com/1/007/367.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
Thanks to IT Home netizen "A Different Feeling" for the tip!
According to IT Home on September 26, VideoCardz discovered today that Intel has released the latest 2.6.0 version of PresentMon, its free open-source game performance monitoring tool.
The new version focuses on optimizing its own resource usage. Intel said CPU usage under high load is reduced by 78%, and it also adds a new "Game Experience (game experience)" monitoring overlay for displaying metrics related to frame smoothness and animation performance.

PresentMon 2.6.0 is compatible with Windows 10 and Windows 11, as well as DirectX 9, DirectX 11, DirectX 12, OpenGL and Vulkan applications. It is compatible with AMD and NVIDIA graphics cards, and supports multiple telemetry interfaces including ADL, NVAPI, NVML, IGCL and WMI.
The biggest change in the new version is the optimization of background services. Intel adjusted the ETW (IT Home note: Event Tracing for Windows, Windows Event Tracing) data refresh mechanism, changing some background loops that do not require high-precision timing to lower-frequency Sleep waiting, reducing CPU usage by 78% under high load. The same applies to mobile platforms.
In addition, when the software is not actively monitoring or polling telemetry data, diagnostic log refreshes are paused, and idle CPU usage drops from about 0.003% to nearly 0%.
For ordinary players, PresentMon adds the Game Experience overlay, which aggregates performance metrics that Intel considers related to dynamic visual quality and animation smoothness.

The 2.6.0 interface has also been improved. Previously, monitoring metrics that could not be used would disappear directly from the list; now they remain in the list and are grayed out, along with a hint explaining why they are unavailable, such as incompatible current hardware, the API not providing data, or the current version of PresentMon not yet being compatible.
In multi-GPU environments, users can simultaneously display telemetry data from multiple GPUs and select the data source of different devices for each metric.

For game testers and developers, PresentMon 2.6.0 adds several analysis capabilities. Among them is monitoring of the DirectX 12 PSO (Pipeline State Object) compilation process, which can record the number of compilations, compilation duration, and busy percentage, and write these into frame data and CSV output.
The new version also supports the newer Windows DDI Flip events, which can record the timestamp of each Flip and the completed and dropped Flip events. It also adds metadata such as Source ID and Layer ID to CSV output, making it easier to analyze multi-screen, multi-plane presentation scenarios.
Regarding GPU telemetry dimensions, Intel improved the accuracy of tracking dedicated video memory utilization using the AMD ADL interface. The telemetry architecture was also adjusted to poll only the data that clients actually request, further reducing platform overhead.
The new version also fixes multiple stability issues, covering failures in automatic target identification, service restart anomalies, multi-instance launch conflicts, and internal API compatibility. Meanwhile, the experimental Flash Injector feature has been removed to simplify project maintenance and the build process.