跳到正文
DEV Community· inputlag·· 6 天前精選AI 評分80

Jensen Huang 離開 NVIDIA 之際,作者如何解讀 RTX 4070 Super 的 4X 測試結果

Jensen left NVIDIA. Here is how I read my 4070 Super 4X result.

AI 導讀

作者在 RTX 4070 Super 的 Forza Horizon 6 單人模式測試中,分別於關閉 FG、官方 2X 及 DLSSG SM86 4X 時錄得 100、180 和 320 FPS。320 FPS 是基準 100 FPS 的 3.2 倍,較官方 2X 的 180 FPS 高約 78%,並非實測四倍加速。作者強調這是單一電腦、單一路線的個人測試,沒有量度延遲;4X 是所選的生成模式。

推薦理由

文章把 4X 生成模式與實測 FPS 增幅分開,並指出單次測試及未量度延遲的限制,有助讀者避免把顯示幀率等同實際輸入更新率。

正文 · 原文

Cover image for Jensen left NVIDIA. Here is how I read my 4070 Super 4X result.

inputlag

100 → 180 → 320 FPS is the result that interests me most in the week Jensen Huang left NVIDIA.

Those were my readings in Forza Horizon 6 single-player on an RTX 4070 Super: FG off, official 2X, then DLSSG SM86 at 4X. One PC, one route, no latency measurement: a personal test, not a benchmark.

Separate the mode from the observed gain

The useful arithmetic is simple:

180 / 100 = 1.8×
320 / 100 = 3.2×
320 / 180 ≈ 1.78×

The final reading is about 78% above official 2X in this test. Calling it a fourfold measured speed-up would be wrong: 4X is the selected generation mode.

Generated frames also need their own label. The simulation and input pipeline don't acquire 320 fresh input samples per second merely because an FPS counter reaches 320.

What the leadership story changes

The November 12 announcement puts Colette Kress in the interim CEO role. For developers integrating graphics features, the practical things to watch are the support matrix, SDK behaviour, driver releases and test coverage.

PC Gamer worries that the management change could reduce gaming's influence. That's a forecast. My preference is to judge the next published compatibility decisions.

The mod README is particularly useful here: it distinguishes claimed support, diagnostic paths and hardware that the author has actually tested. Its RTX 40 route lacks on-hardware data from the author. My own successful run adds an anecdote; it doesn't fill that coverage gap for the whole series.

What I'd put in the next test

Record resolution, render preset, upscaler mode, driver and mod version. Repeat the same route with FG off, 2X and 4X. Capture frame pacing separately from displayed FPS, then measure input latency with a method that actually covers the input-to-display path.

That would make the 320 number much more useful than a prediction about what a new CEO might unlock.

References

Cover photo: Andrey Matveev / Unsplash (illustrative, not my PC).

來源:DEV Community · dev.to