Nvidia RTX 4060/4070/4080曝光：全系5nm 工艺质变 RTX 4090性能表现/发售时间或9月发布：性能猛兽价格更贵了 售价1899美元
RTX 4060/4070/4080曝光：全系5nm 工艺质变 RTX 4090性能表现/发售时间齐曝光：性能猛兽价格更贵了RTX 4060/4070/4080 exposure: the whole series of 5nm process quality change RTX 4090 performance / release time exposure: performance beasts are more expensive The lineup of RTX 30 series graphics cards is basically complete, I believe the next generation of RTX graphics cards should have been mentioned in NVIDIA on the agenda.
The leaker Greymon55 shared that the GPU chip of the RTX 40 series graphics card has been basically confirmed, and the corresponding relationship is as follows: AD102 core: RTX 4080-90 series, 5nmAD103 core: RTX 4070 series, 5nmAD104 core: RTX 4070 series; 5nmAD106 core: RTX 4060 Series, 5nm where AD is the abbreviation of Ada Lovelace, female scientist, corresponding to the RTX 40 series graphics card GPU architecture code.
It can be seen that mainstream to high-end RTX 40 series graphics cards all use 5nm technology, which is a qualitative upgrade compared to Samsung's 8nm.
However, the same RTX 4070 has two cores, AD103 and AD104. One may refer to the mobile terminal and desktop terminal, and one may refer to the RTX 4070 and RTX 4070 Ti.
In terms of performance, the same whistleblower said that the AD102 core’s FP32 performance at 1.8GHz is about 66.5T, and the single-precision floating point at 2.5GHz can touch 90T. You must know that the unreleased RTX 3090 Ti’s official performance is only 40T.
The RTX 3090 Ti was unfavorable in its debut, and the ticket was bounced, because NVIDIA was secretive, and the outside world did not know when this card king would debut. It seems that the Ampere architecture has indeed come to an end.
It is believed that within NV, the power to actively prepare for the RTX 40 series is more powerful than solving the troubles of the RTX 3090 Ti.
A few days ago, foreign hardware blogger Graphically Challenged gave a price forecast for the RTX 40 series graphics card. Compared with his previous statement that the RTX 4090 will sell for $2,999 and the RTX 4050 will sell for $279, he has made corresponding adjustments because the graphics card The market is volatile and availability is improving.
Its latest video points out that the RTX 4090 is expected to be officially priced at $1,899. Of course, even this is $400 more expensive than the RTX 3090. According to the NV "exchange rate" of about 1:8, the retail price of the National Bank is likely to exceed 15,000.
In terms of schedule, GC concluded that it will be released in September and listed in October.
As for the specifications, previous reports said that the RTX 40 series GPU is Ada Lovelace, 5nm process, RTX 4090 corresponds to the AD102 core, there will be up to 12 GPC (graphics processing cluster), 72 TPC (texture processing cluster), 144 SM (streaming multiprocessor), and each SM continues to have 128 stream processors (CUDA cores), then there are as many as 18432 stream processors as a whole, which is an increase of 10752 cores compared to the current Ampere architecture GA102 core. 71%, doubling the overall performance.
The frequency of AD102 can reach 2.5GHz, and the floating-point performance can reach 90TFLOPS. Compared with RTX 3090, the increase is 152%, and compared with RTX 3090 Ti, the increase is 125%.
RTX 30系显卡的阵容已经基本完整，相信下一代RTX显卡在NVIDIA内部应该已经提上日程。
爆料达人Greymon55分享称，RTX 40系显卡的GPU芯片已经基本确认，对应关系如下：
AD102核心：RTX 4080-90系列，5nm
AD103核心：RTX 4070系列，5nm
AD104核心：RTX 4070系列；5nm
AD106核心：RTX 4060系列，5nm
其中AD是Ada Lovelace缩写，女科学家，对应RTX 40系显卡GPU架构代号。
可以看到，主流到高端的RTX 40系显卡均采用5nm工艺，相较于三星8nm，可谓质变升级。
不过，同样是RTX 4070，存在AD103和AD104两套核心，一种可能说的是移动端和桌面端，一种可能说的是RTX 4070和RTX 4070 Ti。
性能方面，此前同一个爆料人的说法是，AD102核心在1.8GHz下FP32性能约66.5T，2.5GHz下单精度浮点可摸到90T，要知道尚未发布的RTX 3090 Ti官宣成绩不过40T。
RTX 3090 Ti出师不利，惨遭跳票，因为NVIDIA讳莫如深，外界也不清楚这块卡皇到底何时登场。看起来，安培架构的确已经走到尽头。
相信在NV内部，积极准备RTX 40系的动力比解决RTX 3090 Ti的麻烦动力更足。
日前，国外硬件博主Graphically Challenged给出了一份RTX 40系显卡的价格预测，对比此前他说RTX 4090要卖2999美元，RTX 4050要卖279美元的说法，做出了相应调整，原因是显卡市场多变，供货正在改善。
其最新视频指出，RTX 4090预计官方定价1899美元，当然即便如此也比RTX 3090贵了400美元，按照1：8左右的NV“汇率”，国行零售价大概率突破1.5万。
时间安排方面，GC给出的结论是，9月发布，10月上市。
至于规格，之前的报道称RTX 40系GPU为Ada Lovelace，5nm工艺，RTX 4090对应AD102大核心，将有多达12个GPC(图形处理集群)、72个TPC(纹理处理集群)、144个SM(流式多处理器)，而每个SM继续128个流处理器(CUDA核心)，那么整体下来就有多达18432个流处理器，相比现在的安培架构GA102核心的10752个核心增加了71％，整体性能提升一倍。
AD102频率能到2.5GHz，浮点性能可达90TFLOPS，相较于RTX 3090的增幅达到152%、较RTX 3090 Ti的增幅达到125%。
