官宣了：5090两千刀
版主： Jack12345
NC CHINESE AMERICANS FOR TRUMP 
#7 Re: 官宣了：5090两千刀
老黃透露 GeForce RTX 5070 Laptop GPU 性能大躍進，不單止遊戲性能可媲美RTX 4090 Laptop GPU，並且功耗只需一半。
草你特铐谱祖宗十八代加后十八代 你踏马的去死吧
#8 Re: 官宣了：5090两千刀
我私人助理說
The RTX 5070 is expected to be a significant performance upgrade over the RTX 4070. Here's a summary of its anticipated capabilities:
* High Performance:
* 1440p Gaming: It's designed to deliver consistently high frame rates (over 100 FPS) at 1440p resolution in demanding games like Cyberpunk 2077 and Hogwarts Legacy, even with ray tracing enabled.
* 4K Gaming: The RTX 5070 aims to match or even surpass the performance of the RTX 4080 Super in some 4K games, allowing for smooth gameplay at high settings.
* Ray Tracing Improvements:
* The RTX 5070 is built on the Blackwell architecture, which offers enhanced ray tracing capabilities compared to the RTX 4070. This translates to significantly better performance in games with ray tracing features.
* DLSS 3.5 Support:
* The RTX 5070 will support DLSS 3.5, the latest version of NVIDIA's frame generation technology. This can further boost frame rates and improve image quality.
Key Factors Contributing to RTX 5070's Performance:
* Architectural Improvements: The Blackwell architecture incorporates advancements in shader design, memory technology, and power efficiency.
* Increased CUDA Cores: The RTX 5070 is expected to have a higher number of CUDA cores than the RTX 4070, leading to increased processing power.
* GDDR7 Memory: The use of GDDR7 memory offers higher bandwidth, enabling faster data transfer between the GPU and system memory.
Overall:
The RTX 5070 is poised to be a powerful graphics card, especially for 1440p gaming. Its strong performance, advanced ray tracing capabilities, and support for DLSS 3.5 make it a compelling option for gamers and content creators alike.
Disclaimer: This information is based on current expectations and may be subject to change.
I hope this helps! Let me know if you have any other questions.
The RTX 5070 is expected to be a significant performance upgrade over the RTX 4070. Here's a summary of its anticipated capabilities:
* High Performance:
* 1440p Gaming: It's designed to deliver consistently high frame rates (over 100 FPS) at 1440p resolution in demanding games like Cyberpunk 2077 and Hogwarts Legacy, even with ray tracing enabled.
* 4K Gaming: The RTX 5070 aims to match or even surpass the performance of the RTX 4080 Super in some 4K games, allowing for smooth gameplay at high settings.
* Ray Tracing Improvements:
* The RTX 5070 is built on the Blackwell architecture, which offers enhanced ray tracing capabilities compared to the RTX 4070. This translates to significantly better performance in games with ray tracing features.
* DLSS 3.5 Support:
* The RTX 5070 will support DLSS 3.5, the latest version of NVIDIA's frame generation technology. This can further boost frame rates and improve image quality.
Key Factors Contributing to RTX 5070's Performance:
* Architectural Improvements: The Blackwell architecture incorporates advancements in shader design, memory technology, and power efficiency.
* Increased CUDA Cores: The RTX 5070 is expected to have a higher number of CUDA cores than the RTX 4070, leading to increased processing power.
* GDDR7 Memory: The use of GDDR7 memory offers higher bandwidth, enabling faster data transfer between the GPU and system memory.
Overall:
The RTX 5070 is poised to be a powerful graphics card, especially for 1440p gaming. Its strong performance, advanced ray tracing capabilities, and support for DLSS 3.5 make it a compelling option for gamers and content creators alike.
Disclaimer: This information is based on current expectations and may be subject to change.
I hope this helps! Let me know if you have any other questions.
光追騙你錢，千萬別上當
效果不咋樣，幀數跟不上
效果不咋樣，幀數跟不上
#9 Re: 官宣了：5090两千刀
要支持 DLSS4 的游戏才行pathdream 写了： 2025年 1月 6日 22:39 老黃透露 GeForce RTX 5070 Laptop GPU 性能大躍進，不單止遊戲性能可媲美RTX 4090 Laptop GPU，並且功耗只需一半。
新的AI假帧厉害了
#10 Re: 官宣了：5090两千刀
你天天都在看假帧
h264 265 vp9里面都有的P帧就是部分信息前向预测出来的
不仅如此还有B帧双向预测，当然在游戏里不能依赖后面的frame来预测
只要流畅合理就行了
#12 Re: 官宣了：5090两千刀
游戏相当于用后来的帧来预测了，DL算法train的时候已经包括完整数据了。所以不完全是extrapolation, 相当于interpolation. 游戏虽然是动态的，但场景有限，和视频类似，和real world不一样。Real world未来是不可预期的，譬如股市，ML用处就没那么大了。（ヅ） 写了： 2025年 1月 7日 11:59 你天天都在看假帧
h264 265 vp9里面都有的P帧就是部分信息前向预测出来的
不仅如此还有B帧双向预测，当然在游戏里不能依赖后面的frame来预测
只要流畅合理就行了
Frame Gen和DLSS是实打实的技术，不完全是忽悠人。不过4x frame gen对于2x frame gen就相当于DLSS performance和DLSS quality. 无论是空间还是时间，插的pixel越多效果就越打折，artifact越多。DLSS quality和2x frame gen属于可接受范围。DLSS performance和4x frame gen白给我也不太会用。
#13 Re: 官宣了：5090两千刀
還是大牛說得好，簡單明聊sephiroth 写了： 2025年 1月 7日 15:55 游戏相当于用后来的帧来预测了，DL算法train的时候已经包括完整数据了。所以不完全是extrapolation, 相当于interpolation. 游戏虽然是动态的，但场景有限，和视频类似，和real world不一样。Real world未来是不可预期的，譬如股市，ML用处就没那么大了。
Frame Gen和DLSS是实打实的技术，不完全是忽悠人。不过4x frame gen对于2x frame gen就相当于DLSS performance和DLSS quality. 无论是空间还是时间，插的pixel越多效果就越打折，artifact越多。DLSS quality和2x frame gen属于可接受范围。DLSS performance和4x frame gen白给我也不太会用。
光追騙你錢，千萬別上當
效果不咋樣，幀數跟不上
效果不咋樣，幀數跟不上
#18 Re: 官宣了：5090两千刀
非常怀疑游戏里面会用后帧做预测sephiroth 写了： 2025年 1月 7日 15:55 游戏相当于用后来的帧来预测了，DL算法train的时候已经包括完整数据了。所以不完全是extrapolation, 相当于interpolation. 游戏虽然是动态的，但场景有限，和视频类似，和real world不一样。Real world未来是不可预期的，譬如股市，ML用处就没那么大了。
Frame Gen和DLSS是实打实的技术，不完全是忽悠人。不过4x frame gen对于2x frame gen就相当于DLSS performance和DLSS quality. 无论是空间还是时间，插的pixel越多效果就越打折，artifact越多。DLSS quality和2x frame gen属于可接受范围。DLSS performance和4x frame gen白给我也不太会用。
跟视频不一样，视频压缩实际上一开始就拥有完整的视频信息，所以理论上用最后一帧的信息来压缩第一帧数据也不是毫无可能。游戏不应该会有向后的依赖性。
#19 Re: 官宣了：5090两千刀
不是直接用后帧插值预测。Machine Learning你可以看成个函数或black box. 函数的形式由ML的方法来提供，线性非线性都有。函数的input variable用现在的state, 譬如最近几帧和玩家input等。函数的参数是用整个场景calibrate出来的，包括未来可能到的地方，这个是DLSS training来发现的，老黄说他用Neuro Network. 所以整个算法间接用到以前没见到的信息。这和视频不一样，视频没有user interaction这部分input，但实际上idea是一样的，静态压缩和动态压缩的区别。（ヅ） 写了： 2025年 1月 7日 17:35 非常怀疑游戏里面会用后帧做预测
跟视频不一样，视频压缩实际上一开始就拥有完整的视频信息，所以理论上用最后一帧的信息来压缩第一帧数据也不是毫无可能。游戏不应该会有向后的依赖性。
#20 Re: 官宣了：5090两千刀
换个说法，我觉得在游戏里面不应该会用到t>0时候的任何信息来组装t=0的输出sephiroth 写了： 2025年 1月 7日 17:57 不是直接用后帧插值预测。Machine Learning你可以看成个函数或black box. 函数的形式由ML的方法来提供，线性非线性都有。函数的input variable用现在的state, 譬如最近几帧和玩家input等。函数的参数是用整个场景calibrate出来的，包括未来可能到的地方，这个是DLSS training来发现的，老黄说他用Neuro Network. 所以整个算法间接用到以前没见到的信息。这和视频不一样，视频没有user interaction这部分input，但实际上idea是一样的，静态压缩和动态压缩的区别。
普通不dlss的模式就是利用t<=0的信息来计算t=0的输出但没有预测部分。dlss利用了t<0即之前的画面输出或者其他参数比如你说的用户输入，来预测t=0的画面输出。