Administrators weighing an early move to Linux 7.4 may want to hold off: the first round of benchmarking of the in-development kernel shows a mixed picture against the current 7.3 stable release, with modest gains in some workloads offset by small regressions in others.
According to Phoronix, which published the initial Linux 7.4 comparison, testing was run on a System76 Thelio Major workstation equipped with an AMD Ryzen Threadripper 9980X. The outlet found some performance improvements worth noting over Linux 7.3 — but also a number of small regressions, leaving the new kernel looking broadly flat rather than compelling on the hardware tested.
For administrators running this class of machine, the practical takeaway is a "wait and see" one. The results do not present a strong case for upgrading a production fleet now, and there is little downside in letting 7.4 complete its stabilization cycle through its first point releases before adopting it. Administrators who have standardised on 7.3 have a known quantity; 7.4 is still in development, and its performance profile is not yet settled.
There are caveats attached to any conclusion drawn at this stage. This is a preliminary benchmark pass on a single hardware platform — one workstation, one CPU — and results from an in-development kernel can shift as the merge window settles and fixes land. Regressions identified now may be resolved before 7.4 reaches general availability, and gains may erode just as easily. The findings should be read as an early indicator, not a verdict, and they may not generalise to other AMD platforms, let alone Intel systems or server-class deployments. Administrators sceptical of either conclusion can check the work themselves: Phoronix publishes its benchmark suites openly, so anyone with Threadripper hardware of their own can reproduce the comparison rather than waiting for the next round of testing.
Where a mixed, near-flat result does still matter is at scale. For build farms, continuous-integration clusters, rendering and simulation workloads, even small throughput differences compound across hundreds of machines over long run times. Teams operating in that regime will want to re-run the comparison on their own hardware — and to track subsequent Phoronix rounds as 7.4 matures — before drawing conclusions about capacity planning or node refreshes.
The broader context is a familiar one for anyone who follows kernel development: a new kernel cycle is not automatically a faster one. Feature work, scheduler changes and driver updates can trade performance in one area for stability or functionality in another, and the first benchmarks of an in-development release rarely capture the final shape of the kernel.
Bottom line for administrators: stay on Linux 7.3 for now, revisit the question once 7.4 has stabilised through its early point releases, and treat any upgrade decision as one to validate against your own workload rather than a single benchmark run.
正考慮盡早升級至 Linux 7.4 的系統管理員或須按兵不動：這款開發中 kernel 的首輪基準測試結果，與現行 7.3 穩定版本相比好壞參半——部分工作負載有溫和增長，卻被其他方面的小幅回歸所抵銷。
據率先發表 Linux 7.4 對比測試的 Phoronix 報導，測試在配備 AMD Ryzen Threadripper 9980X 的 System76 Thelio Major 工作站上進行。該媒體發現相比 Linux 7.3 有若干值得留意的性能提升，但同時也出現多處小幅回歸，令這款新 kernel 在所測硬件上整體表現持平，並未令人眼前一亮。
對於運行此類機器的管理員而言，實質結論是「觀望為上」。測試結果並不足以支持現時為整個生產環境集群升級，而且讓 7.4 先通過首幾個 point release 完成穩定化週期再作採用，並無太大損失。已將環境標準化至 7.3 的管理員手上是已知的變量；7.4 仍在開發之中，其性能表現尚未塵埃落定。
在此階段得出的任何結論都附帶保留條件。這是針對單一硬件平台——一部工作站、一款 CPU——的初步基準測試，開發中 kernel 的結果可能隨着 merge window 趨於穩定及各項修正落實而出現變化。現時辨識出的回歸問題，可能在 7.4 正式發佈前已獲解決；同樣，現有的增益也可能輕易流失。相關發現應視為早期指標，而非定論，而且未必適用於其他 AMD 平台，更遑論 Intel 系統或 server-class 部署環境。對任何一方結論存疑的管理員可自行驗證：Phoronix 公開發表其 benchmark suite，擁有 Threadripper 硬件者均可自行重現該對比測試，毋須等待下一輪測試結果。
此類好壞參半、近乎持平的結果，真正影響顯著之處在於大規模部署。對於 build farm、持續整合（continuous integration）集群、渲染及模擬等工作負載，即使吞吐量只有微小差異，在數百部機器及長時間運行之下亦會不斷累積。在此環境運作的團隊，宜先在自家硬件上重跑對比測試，並持續追蹤 Phoronix 隨 7.4 成熟而推出的後續測試輪次，然後才就容量規劃或節點更新作出結論。
對於關注 kernel 開發的人士來說，更大的背景並不陌生：新一個 kernel 周期並不自動等於更快的 kernel。功能開發、排程器改動及驅動程式更新，往往以某方面的性能換取另一方面的穩定性或功能，而開發中版本的首批基準測試，極少能反映 kernel 的最終面貌。
系統管理員的最終結論：暫時留駐 Linux 7.3，待 7.4 通過早期 point release 穩定下來後再重新檢視此問題，並切記任何升級決定都應以自身工作負載驗證，而非僅憑一次 benchmark 結果作判斷。