也许有很多同学怀念10年前价格不到1500元的“甜点”显卡，比如GTX 460。再往前回溯5年，ATi Radeon X1650 PRO更是只要699元就能买到，而它的确是一张拥有当年顶级显卡6成性能的“甜点”。15年前的“甜点”，拥有ATI X1950 PRO六成的性能
很难想象RTX 3060 Ti如果只卖699元会怎样，事实上这也是不现实的！当年卖699元的“甜点”，不论是做工还是功能方面都无法和当今的千元卡相比。时代在变，想要“甜点”回到当初千元以内的价位或许只能是存在于回忆中遥不可及的梦想了。
不过这一次安培构架的发布多少给了玩家一些安慰，RTX 3080只需一半的价格就提供了比上代RTX 2080 Ti多出30%的性能。可是即便如此，5499元也不是大多数玩家能够承受的售价。
没有让大家等太久，在RTX 3080上市仅仅2个月之后，RTX 3060 Ti也终于来到了大家面前。这也是安培构架第一款售价在3000元以下的甜点产品。
果然比RTX 2080s要强！RTX 3060 Ti首发评测：第一款买得起的安培显卡
RTX 3060 Ti是在RTX 3070的GA104-300核心上阉割一刀后的产物，核心代号为GA104-300，晶体管规模达到了174亿，芯片面积392平方毫米，拥有5组GPC，內建有4864个CUDA核心，同时有152个纹理单元、80个ROP单元、152个Tensor核心、38个RT核心。
从上面的参数规格表可以看出，RTX 3060 Ti的规格与RTX 3070比起来，刚好就是拿掉了1组GPC，从原来的6组变为了5组，整体上的参数规格少了1/6。
唯一没有被阉割的就是显存，依旧是4个带宽为64bit的双通道显存控制器组成了总量为256bit的显存控制单元，大小为8GB，带宽448GB/s。
Perhaps many students miss the "dessert" graphics cards that cost less than 1500 yuan 10 years ago, such as the GTX 460. Going back 5 years, ATi Radeon X1650 PRO can be purchased for only 699 yuan, and it is indeed a "dessert" with 60% of the performance of the top graphics card of the year. The "dessert" 15 years ago, with 60% of the performance of the ATI X1950 PRO, it is hard to imagine what would happen to the RTX 3060 Ti if it only sold for 699 yuan. In fact, this is also unrealistic! The "desserts" that sold for 699 yuan back then can't be compared with today's thousand-yuan cards in terms of workmanship and functionality. Times are changing, and wanting "desserts" to return to the original price of less than one thousand yuan may only be an unattainable dream that exists in memory.
However, the release of the Ampere architecture this time gave players some comfort. The RTX 3080 provides 30% more performance than the previous generation RTX 2080 Ti at half the price. But even so, 5499 yuan is not a price that most players can afford.
I didn't make everyone wait too long. Only 2 months after the RTX 3080 was launched, the RTX 3060 Ti finally came to everyone. This is also the first dessert product of Ampere Architecture that sells for less than 3,000 yuan.
Sure enough, better than RTX 2080s! RTX 3060 Ti first evaluation: The first affordable ampere graphics card RTX 3060 Ti is a product castrated on the GA104-300 core of RTX 3070. The core code is GA104-300, the transistor scale has reached 17.4 billion, and the chip area 392 square millimeters, with 5 groups of GPC, built-in 4864 CUDA cores, at the same time there are 152 texture units, 80 ROP units, 152 Tensor cores, and 38 RT cores.
It can be seen from the above parameter specification table that the specifications of RTX 3060 Ti are compared with RTX 3070, just one group of GPC has been removed from the original 6 groups to 5 groups, and the overall parameter specifications are 1/6 less. .
The only thing that has not been castrated is the video memory. It is still 4 dual-channel video memory controllers with a bandwidth of 64bit to form a total 256bit video memory control unit with a size of 8GB and a bandwidth of 448GB/s.
A note on the performance of the Ampere GPU streaming processor: Usually the indicator to measure the GPU streaming processor is the computing power of FP32. The Turing architecture uses an equal number design of FP32 and INT32. However, due to the increasing demand for FP32 in today's games The more so, a more flexible design is adopted on the Ampere architecture. Half of the stream processors that can only calculate INT32 can execute FP32 or INT32 more flexibly. This design greatly enhances the computing power of FP32 while avoiding the waste of INT32 computing power.
Sure enough, better than RTX 2080s! RTX 3060 Ti first evaluation: The first affordable Ampere graphics card. Although there are not so many stream processors in terms of absolute number of stream processors, the innovation of the architecture has significantly improved the processing power of FP32 while avoiding INT32 idle performance waste. The more sensitive applications are to floating-point operations, the greater the performance gains under the new architecture.
In general, the number of stream processors of the RTX 30 divided by 1.5 is approximately equal to the performance of the RTX 20. Based on this conversion, the 4864 stream processors of the RTX 3060 Ti are roughly equivalent to the Turing GPU of 3242 stream processors, so there will be no big problem in terms of performance stronger than the RTX 2080 SUPER, and of course it will not be much stronger. The latter has 3072 stream processors.
关于安培GPU流出器效能的说明：
通常衡量GPU流处理器的指标为FP32的运算能力，Turing架构上采用的为FP32和INT32的对等数量设计，而由于当今的游戏中对于FP32的需求越来越多，因此在Ampere架构上采用了更灵活的设计，原本只能计算INT32的流处理器中有一半可以更灵活的执行FP32或INT32。这样的设计让FP32的运算能力大幅加强，同时避免了INT32的算力浪费。
果然比RTX 2080s要强！RTX 3060 Ti首发评测：第一款买得起的安培显卡
从绝对的流处理器数量上来看虽然并没有实际上看到的那么多，但是架构的创新显著提升了FP32的处理能力，同时避免了INT32闲置的性能浪费。对浮点运算越敏感的应用，在新的架构下获得的性能收益就越大。
大体上，RTX 30的流处理器数量除以1.5，就约等于RTX 20的性能。以此换算，RTX 3060 Ti的4864个流处理器大致等效于3242个流处理器的Turing GPU，因此在性能上强于RTX 2080 SUPER不会有太大问题，当然也不会强太多，后者拥有3072个流处理器。
详细评测 https://news.mydrivers.com/1/727/727222.htm
