GTX 970 Controversy: Explained!
Couple of months ago, NVidia launched their new series of GPU`s promising outstanding performance at resolutions of up to 4K. As the weeks passed the GTX 980 and 970 raised their crowns and affirmed that they are the best graphics card on the market at the date. All of this turned really bad for the 970, when some reviewers detected that there were some problems with the memory allocation on this SKU.
Basically, GTX 970 is a partial disable of the 980, allowing them to use the faulty GPUs that weren`t good enough for the 980. This isn`t a new concept by any stretch of the imagination, and it`s not the cause of the memory allocation problem. The real problem is that NVidia, in addition of three SMX units being disabled, they decided to reduce L2 Cache of the chip from 2MB to 1.75MB and also disable some ROPs.
Some of the reviewers ran some test and came with the result that the GPU couldn`t reach 512 MB of the 4GB of V-RAM that this monster has. This translated into lower performance in games that require more than 3.5GB of memory. That`s how it all got started.
NVidia responded to this as a miscommunication error between the designing department and the marketing one. They said that the memory is allocated in a different way from past Graphics Card, and the memory is divide in two sections: one 3.5GB sections and one 0.5GB.
Further research on the fan side discovered that the GPU has 56 ROPs, not the advertised 64 ROPs. The segmentation isn`t a bat thing if done properly, but the way that NVIDIA did it, was that the 3.5 GB segment has higher priority than the 0.5GB one on the crossbar(more on this later). This made the smaller size “partition” so slow that it made no difference of it being there or not.
The crossbar is the medium between the GPU and the V-RAM. Think of it as a conveyor belt. On one side, the packages are picked up and on the other side, the packages are loaded. The problem is when the offloading is waiting on the loading part to start. This slows the whole process.
Why NVidia chose to reduce the ROPs remains a mystery, and why they lied to us also remains unknown. Don`t think that the chip designer’s aren`t reading reviews… They do, and if it was a communication problem, when they saw that the specs are scrambled on reviews, than they would`ve announced the whole food-chain. In any case, this controversy isn`t doing them any good, and they should`ve been fair from the get-go.
If you enjoyed the article, than don`t forget to leave a like. If your feelings are more complicated than this, leave a comment telling me what it`s your opinion on this issue.