本模组对 Create 工厂仪表的渲染进行了低垂果实(即低成本、高收益)优化。
为何之前缓慢
工厂仪表是一个方块实体,而原版模组的渲染器每帧都会为仪表路径进行顶点变换。考虑到路径在大部分帧之间几乎没有变化,这种开销非常高。现有的解决方案如 createbetterfps 和 Flerovium 并未改变这种立即模式模型;本模组利用 Flywheel 提供了实例化支持。
限制条件
物品可能拥有自定义渲染器。这些物品将回退到较慢的渲染路径。
本模组仅适用于客户端。它不会优化服务器端的刻(tick)。
本模组是针对 Create 工厂仪表渲染的一种临时性、低垂果实式的优化。它并不能提供全面提升你帧率的完整解决方案。
本模组主要优化的是 CPU 帧时间,而非 GPU 帧时间。如果使用了光影,提升幅度可能会很小。
评估结果
在受控场景中进行评估:超平坦世界,拥有一面 10x30 的墙壁,每个方块上有 4 个仪表。在 SparkC 分析器中观察到显著改进。结果如图所示。

在真实世界场景中评估:参考一个游玩时长 80 小时以上的生存世界,站在同一位置不动。关闭垂直同步,帧数无上限,Sodium 设置为 CPU 预渲染 1 帧,Photon 光影使用 Ultra 预设。预热完成后,PresentMon 显示的数据(10秒间隔)如下:
| 指标 |
未启用 |
已启用 |
变化 |
平均帧间隔 MsBetweenPresents |
11.291 ms |
10.618 ms |
减少 0.673 ms,约 5.96% |
| 对应吞吐量 |
88.57 FPS |
94.18 FPS |
增加约 6.3% |
| p50 帧间隔 |
10.644 ms |
10.252 ms |
减少约 3.7% |
MsGPUTime |
11.290 ms |
10.617 ms |
减少 0.673 ms |
MsGPUBusy |
10.802 ms |
10.498 ms |
减少 0.304 ms,约 2.8% |
MsGPUWait |
0.489 ms |
0.120 ms |
减少 0.369 ms,约 75% |
MsGPULatency |
12.594 ms |
14.526 ms |
增加 1.932 ms |
MsInPresentAPI |
0.124 ms |
0.109 ms |
两者均非常小 |
PresentMon 报告的 CPUUtilization |
29.10% |
20.16% |
减少约 31% |
This mod does a low-hanging-fruit optimization for Create factory gauge rendering.
Why It was Slow
Factory Gauge is a block entity, and the base mod renderer does vertex transform for gauge paths every frame. this has very high overhead considering the path is mostly unchanged across frames. Existing approaches such as createbetterfps, Flerovium does not change this immediate mode model; This mod utilizes flywheel to provide instancing support.
Limitations
Item may have custom renderer. those item will fallback to slow path.
This mod is client only. it doesn't optimize server ticks.
This mod is ad-hoc, low-hanging fruit optimization for create factory gauge rendering. it doesn't provide a full solution to making your fps better.
This mod optimizes CPU frame time more than GPU frame time. if shader used, improvement may be small.
Evaluation
Evaluated in controlled scenario: superflat world, with a 10x30 wall, each block has 4 gauges. significant improvement in sparkc profiler. result in figure.

Evaluated in realworld scenario: referencial 80h+ survival world, same spot standing still. VSync off, max frame no limit, sodium: CPU render ahead 1 frame, photon shader ultra preset.after warm up, PresentMon shows (10s interval)
| Metric |
Without |
With |
Change |
Avg frame interval MsBetweenPresents |
11.291 ms |
10.618 ms |
Decreased by 0.673 ms, ~5.96% |
| Corresponding throughput |
88.57 FPS |
94.18 FPS |
Increased by ~6.3% |
| p50 frame interval |
10.644 ms |
10.252 ms |
Decreased by ~3.7% |
MsGPUTime |
11.290 ms |
10.617 ms |
Decreased by 0.673 ms |
MsGPUBusy |
10.802 ms |
10.498 ms |
Decreased by 0.304 ms, ~2.8% |
MsGPUWait |
0.489 ms |
0.120 ms |
Decreased by 0.369 ms, ~75% |
MsGPULatency |
12.594 ms |
14.526 ms |
Increased by 1.932 ms |
MsInPresentAPI |
0.124 ms |
0.109 ms |
Both very small |
CPUUtilization reported by PresentMon |
29.10% |
20.16% |
Decreased by ~31% |
评论 0
登录后参与评分
还没有评论,来抢沙发吧