The newest report suggest that Nvidia's upcoming GM200 GPU might not be optimized for double precision compute performance, which should make it similar to the GM204 GPU.

According to a new report from 3DCenter.com, Nvidia's GM200 GPU will not be great when it comes to double precision compute performance but rather be optimized for single precision compute performance, which is quite unusual for enthusiast class GPU and makes the GM200 GPU similar to the GM204 GPU, one behind the Geforce GTX 980 and the GTX 970 graphics cards.

While this is just a rumor report and has yet to be confirmed, means that Nvidia decided to target gamers with its latest GM200 GPU, rather than scientific research, professional applications and servers, where double precision floating point (FP64) performance actually matters. While FP64 does matter in those professional applications of the GPU, it raises the cost, the size as well as the power consumption of the chip.

Nvidia's latest enthusiast class GPU, the GK110 GPU, which can be found on the Quadro K6000 and GTX Titan graphics cards featured decent FP64 peak performance, with ratio between single precision and double precision floating point compute performance of 3:1.

If the next GM200 based Titan, or whatever Nvidia decides to call it, graphics card ends up with low double precision compute performance, this will have a significant impact on the professional graphics card market, where AMD is now doing quite well thanks to high double precision floating point compute performance of Hawaii GPU based FirePro W9150 of 2.8 TFLOPs.

Nvidia has possibly crunched the numbers which showed that most Titan buyers are actually gamers and might have a different graphics card ready for professional graphics card market. In any case, recent rumors suggested that we might see the new GM200-based GTX Titan graphics card sometime in mid-March.



Source: Wccftech.com.