NVIDIA official representatives finally talked about 970 GTX memory issue.
Fundamental graphic material became a diagram explaining the GPU GeForce GTX 970 graphics cards and behavior of the memory system. Let's start with the fact that GM204 in this configuration has 13 active computing units of the 16 available. Accordingly, the number of CUDA cores reduced to 1,664 pieces and blocks of texture sampling - up to 104 pieces. Moreover, contrary to previously published information, GeForce GTX 970 has only 56 ROP units, and the cache memory of the second level
reduced from 2048 KB to 1792 KB.
Comparison of GeForce GTX 980 and GeForce GTX 970 characteristics is given below:
It is important to understand that the memory bus is full - 256-bit, and even the theoretical memory bandwidth remains at the same level as the GeForce GTX 980. However, memory, GeForce GTX 970 is actually divided into two segments: fast volume of 3.5 GB and slow 512 MB.
Naturally, when accessing the second memory segment, performance may decline. Keep in mind that there is a conditional third memory segment - RAM, an appeal to which there is a bus PCI Express. In principle, the slow on-board memory still exceeds the RAM speed. On the other hand, this segment is not as fast as the first 3.5 GB, and this causes some complain among users.
In principle, the decision to select NVIDIA GeForce GTX 970 configuration was a compromise. It was possible to cut a niche with respect to memory GeForce GTX 980 and reduce the amount of memory, then "all would be equal," but the company chose to provide some extra performance, though not directly proportional, while retaining the attractive price of the product.
Colleagues from the site PC Perspective argue that outside of specialized tests for memory system in real games, the loss of GeForce GTX 970 performance from such a compromise arrangement does not exceed 4-6%.
Related Products :
|