婆罗门
精华
|
战斗力 鹅
|
回帖 0
注册时间 2012-6-5
|
本帖最后由 塔奇克马 于 2021-12-26 18:11 编辑
看回帖yuzu好像能行了,试了下帧数
配置老样子9900F 4.6G 2600 DDR4 3060TI
x2+1440P配置
古拉两个地方大概30+的样子,这要60帧数有点难度,有12代的测下吧,全程60不知道行不行.
----------------------------------
- 1.0.7047 - 2021-09-19
- Added:
- Adds a method to PhysicalMemory that attempts to write all cached resources directly, so that memory tracking can be avoided. The goal of this is both to avoid flushing buffer data, and to avoid raising the sequence number when data is written, which causes buffer and texture handles to be re-checked.
- Improves performance on Xenoblade 2 and DE, which were flushing buffer data on the GPU thread when trying to write compute data.
- May improve performance in other games that write SSBOs from compute, and update data in the same/nearby pages often.
- 1.0.7044 - 2021-09-19
- Added:
- Changes the RangeList to cache the Address and EndAddress within the list itself rather than accessing them from the object's properties.
- Greatly improves performance in Super Mario Odyssey (~1.25x), Xenoblade (required some WIP code for now) and most other GPU limited games.
- Improvement will generally depend on how many buffers the game binds and how many draws it does.
复制代码
继续提升性能
不过没看出来,古拉帧数还是不稳.FIFO下降
-------------------------------------------------
更新一周目的存档,只是能用的存档,并非完美档.
--------------------
- 1.0.7004 - 2021-08-20
- Changed:
- Removes overlapping textures from the texture cache, if the overlapping texture is not view compatible with the existing one, and its data has been modified.
- Greatly reduces memory usage in Xenoblade Chronicles: Definitive Edition and Xenoblade Chronicles 2, in areas with high usage (like Gormott).
- Fixes crashes and performance issues caused by the system running out of memory.
复制代码 这个就不测了,内存小的可以试试.
yuzu那边不知道啥情况.没怎么关注
-------------------------------
- 1.0.6970 - 2021-07-16
- Changed:
- Improves performance of buffer and texture flushes by using a different approach.
- Greatly improves performance of The Legend of Zelda: Skyward Sword HD.
- Pokémon Sword/Shield is also improved, however it is still slower than the LDN 2.3 build.
- Xenoblade Chronicles 2 is sligthly improved.
- Any other game that does a lot of flushes per frame should be faster aswell.
复制代码
古拉城镇提高3-4帧的样子,FIFO占用80+帧数就往下掉.最低帧数23左右.
----------------------------------------------------------------
发现yuzu可加载DLC了,不过不能直接设置分辨率倍率,推测可以用mod实现.
有报告无效,不过懒得试....
-----------------------------------------
1.0.6901 - 2021-06-02
- Changed
- Now uses quads on OpenGL hosts when supported.
- Improves OpenGL performance in Xenoblade Chronicles DE/2, Fast RMX, and potentially other games.
复制代码 黄金国动画那里大概提升2帧的样子。
古拉还是卡,估计古拉几年内不会流畅的.
1.0.6894 - 2021-05-24
- Changed
- POWER Update: implemented a new host mapped memory manager for major performance improvements across the board. Fastest option is now set by default.
- Increases FPS by significant margins in most games/scenarios (except where emulated GPU-limited).
- Reduces PPTC compilation times by roughly half.
- Reduces boot time of most games by 30-50%.
- Reduces size of JIT cache.
- Resolves "JIT Cache Exhausted" crashes.
- Resolves "AcquireSemaphore" crashes on all Ryzen CPUs known to exhibit the issue.
- Resolves slow character movement speed in the monastery in Fire Emblem: Three Houses.
- Noticeably improves/shortens "spool-up" period when first playing a game (note: does not negate the need to compile shaders).
复制代码
帧数掉下去,这次(也不知道是不是这次,也不知道是不是我换U的缘故,不过一般来说模拟器需求是CPU核心数+1也就是说这个大概5核心就够了.)又上来了一点点,这个场景提升5帧左右.
但是古拉城镇内依然不到20.还是拉跨.
-----------------------------
1.0.6770 - 2021-03-02:
- Changed
- GPU:
- Fixed various issues with view compatibility, synchronization and creation.
- Greatly improved texture view creation, keeping it within a strict storage-view hierarchy. (no view-views)
- Fixes a memory leak in Hyrule Warriors: Age of Calamity. Its VRAM usage is now rather low. :)
- Reworked texture memory tracking to use a structure called a "Texture Group".
- There is one texture group for each storage texture, and views of that storage cover a subset of its "Texture Group Handles".
- A handle can represent one or more sub-image, which can be a mip level, array layer, cubemap face or 3d texture slice.
- This allows memory tracking to be shared between views, which is good as they also share internal data.
- This fixes issues where cpu texture invalidation would be missed or repeat sync when using multiple views of the same texture.
- Affected games:
- Fixed swapping textures in Xenoblade Chronicles: Definitive Edition, Xenoblade 2, MH:GU, potentially other games.
复制代码 看上去修复了贴图错误,好像是贴图不会变蓝(法向贴图把当作纹理贴图显示出来)了.
但是帧数好像也降低了
开30次菜单没死掉,好像稳定性有提高,
-------------------------------------
更新帧数大幅提升
- 1.0.6122 - 2020-12-17
- Changed
- Interrupt GPU command processing when a frame becomes ready, presenting it immediately.
- Greatly improves frame pacing in games that do not wait for a present before starting to draw the next frame (and run below full speed)
- Examples: Xenoblade Chronicles: Definitive Edition, Xenoblade 2
- Vsync event and surface flinger consumption now happens at exact timings (eg. 16.6667ms) rather than rounding to the nearest millisecond (16/17ms)
- Greatly improves games that rely on the vsync event for timing, such as Link's Awakening, or unity games that aren't tied to frame count.
- Should result in less framerate fluctuation.
- Improves frame consistency in windowed mode, though it can still get out of sync and double frames. Exclusive fullscreen mode should be perfect.
- Overhauled Vsync off mode. Rather than simulating a vsync every 1ms, it will now simulate a vsync every time a frame is produced. This results in higher max framerates.
- Alongside, the artificial 60fps limit has been removed. If you disable your vsync using driver config, or have a high refresh rate screen, you can now go over 60fps.
- With a 60hz display and driver vsync enabled, you can disable guest vsync to pass through the host vsync signal. This may result in smoother windowed mode gameplay.
- FIFO% now takes a running total of processing time and total time, then calculates the % immediately before reporting it. This is more accurate.
- Before, it would take % values for each "frame", then average all of them. This was inaccurate when our "frame" boundary failed, or frames had uneven cost.
复制代码
帧数大幅提升,可以看看古拉这个帧数下应该没问题. CPU占用也涨了
当然图形有些错误应该没修正,稳定性我也不知道改善多少了(当然这个稳定性不影响存档的.)
顺便看了下黄金之国帧数也有提升
最后cg帧数
嗯...这个倒是没啥改变...
-------------------------------------------
24楼画质配置文件.不过我自己没看出区别来,也不知道改哪里,不知道有没有人测试下
- 0100f3400332c000\自定义文件\romfs\monolib\shader\lib_nx.ini
复制代码 类似于这样放置到mod文件夹中
观察log是否载入成功
----------------
ryujinx这个版本之后xb2的黑屏死机现象大幅改善..
原来版本loading,切菜单很大概率会死掉(2/3?),现在只是小概率会死掉(1/10?),
现在已有问题
-贴图错误,错位,偶尔发生.
-异刃立绘错位(好像是修复死机的新bug).
-帧数还是需要提高
相比yuzu的区别
-可以载入dlc.不会因为帧数影响分辨率.
在9600kf 4.8G的机器上有缓存的情况下大概除古拉以外20-30帧,更多核心和超线程应该影响不大
内存需要16G,系统虚拟缓存需要50G以及N卡.
现在是可玩的程度
大概模拟进度80%.
完全没有问题,大概还需要一两年?
开了x2(1440P?)分辨率,感觉好像有点像那么回事....
|
本帖子中包含更多资源
您需要 登录 才可以下载或查看,没有账号?立即注册
×
|