Four-RTX-5080 benchmark withdrawn: a complete table is not a substitute for run logs
club-3090 #1966 was posted and withdrawn within this reporting window. The author reports missing run logs, hand-entered errors and failed swap checks on Flash-Next rate arms. AraoAI is not reproducing the withdrawn rates.
Source: club-3090 社区报告撤回声明
A community benchmark describing QEngine 1.3.5, four RTX 5080 cards and multiple card counts was posted and withdrawn within this reporting window. club-3090 #1966 now replaces the rates with an explanation of evidence gaps. The body labels the withdrawal November 8; GitHub reports the last update at 20:47 UTC on November 7, which is early November 8 in Beijing and inside this window.
-
The issue is missing evidence, not an unattractive speed. The author says two million-context results had no per-boot logs, the lower intermediate long-context rung was never run, and the version A/B had no retained logs. A mean and two quality rows were also entered incorrectly. These are the author's public audit statements, not a claim that AraoAI independently inspected every underlying log.
-
Completing a request does not establish measurement validity. The author also says every Flash-Next rate arm failed the project's own swap check during measurement. Hardware names and engine versions cannot replace checks of the machine's state. Once the source withdraws its report, selecting appealing individual rates would recycle invalidated evidence.
-
Withdrawal does not prove the model or GPU failed. What is established is that this report does not currently support a selection decision. It does not show that four RTX 5080 cards cannot run the models or that all QEngine measurements are invalid. The author intends to remeasure and check each claim against logs. Until new evidence is available, the withdrawn report is excluded from AraoAI's benchmark drafts and hardware comparisons.
Check corrections and withdrawals before engine versions, configuration and rates. A complete table establishes formatting, not that every row was actually measured. This article reports the withdrawal without republishing its numbers or attributing motives. AraoAI has not independently reproduced the hardware runs.
Source: https://github.com/noonghunna/club-3090/issues/1966.