根据此前消息，RTX 40系显卡目前处于原型卡阶段，外形和参数都存在相当的调整空间。
今日（6月23日），爆料达人Kopite修正了一些对RTX 4090/4080/4070的规格说法。
其中RTX 4090从16128个CUDA上调为16384个CUDA，也就是从126组SM增加为128组SM单元，虽然依旧不是满血AD102核心（144组SM），但后者可以为TITAN或者RTX 4090 Ti或者RTX专业卡留下空间。
他继续强调RTX 4090的TGP是450W，匹配24GB GDDR6X显存（21Gbps，384bit）。
RTX 4080/4070显卡大曝光：意外缩水
RTX 4080虽然依旧集成10240个CUDA，可是改为18Gbps的GDDR6显存，总容量16GB。
至于RTX 4070，则是7168个CUDA，但显存严重“缩水”，从12GB 192bit GDDR6砍成10GB 160bit GDDR6。
关于发布时间，另一个消息源BC成，出于保护AIC伙伴清货RTX 30系库存考虑，NV仅会在10月份公布一款RTX 40系显卡，预计是高高在上的RTX 4090，RTX 4080/4070则纷纷延宕至2023年初。
According to previous news, the RTX 40 series graphics card is currently in the prototype card stage, and there is considerable room for adjustment in shape and parameters.
Today (June 23), the whistleblower Kopite revised some specifications for the RTX 4090/4080/4070.
Among them, RTX 4090 has been increased from 16128 CUDA to 16384 CUDA, that is, from 126 groups of SM to 128 groups of SM units, although it is still not full blood AD102 core (144 groups of SM), but the latter can be TITAN or RTX 4090 Ti or RTX professional cards leave room.
He went on to emphasize that the TGP of the RTX 4090 is 450W, matching 24GB of GDDR6X memory (21Gbps, 384bit).
RTX 4080/4070 graphics card exposure: unexpected shrinkage
Although the RTX 4080 still integrates 10240 CUDAs, it is changed to 18Gbps GDDR6 memory with a total capacity of 16GB.
As for the RTX 4070, it is 7168 CUDA, but the video memory is severely "shrinked", from 12GB 192bit GDDR6 to 10GB 160bit GDDR6.
Regarding the release time, another source BC Cheng, in order to protect the AIC partners to clear the stock of the RTX 30 series, NV will only announce one RTX 40 series graphics card in October, which is expected to be the top RTX 4090, RTX 4080/4070 They have been delayed until early 2023.
