Home / Component / Graphics / Nvidia ‘Pascal’ GPUs to offer up to 10X higher speed than ‘Maxwell’

Nvidia ‘Pascal’ GPUs to offer up to 10X higher speed than ‘Maxwell’

At its GPU Technology Conference Nvidia Corp. revealed some details about its next-generation code-named “Pascal” architecture of graphics processing units. As expected, the new GPUs will feature numerous innovations that will provide significant benefits in performance in various types of applications.

Nvidia noted three key technologies that the “Pascal” GPUs will have: mixed precision computing, support for stacked multi-layer 3D memory and NVLink GPU bus. As reported previously, the “Pascal” GPUs will be made using 16nm FinFET process technology, therefore, expect increased count of stream processors, and higher clock-rates in addition to architectural enhancements. According to Nvidia, in peak cases, when all improvements come into play, “Pascal” GPUs can be 10 times faster than “Maxwell” GPUs. In typical cases expect next-gen GPUs to be around two times faster than contemporary graphics chips.
nvidia_artworklatesteyeweb_mid
Mixed-precision computing enables “Pascal” architecture-based GPUs to compute at 16-bit floating point accuracy at twice the rate of 32-bit floating point accuracy, which is useful for a number of GPGPU tasks. In certain cases GPUs could do graphics work with 16-bit FP precision without degradation of quality, but not in all of them.

Stacked multi-layer 3D memory will increase memory bandwidth available to new GPUs. Nvidia does not reveal what kind of memory it plans to use, but depending on the Pascal’s availability timeframe, it could be first-generation HBM (with up to 640GB/s bandwidth) or second-generation HBM with unprecedented capacities and north from 1TB/s bandwidth. Extreme memory bandwidth brings benefits to virtually all types of applications, including video games. Graphics cards featuring “Pascal” GPUs and second-gen HBM will carry up to 32GB of memory, according to Nvidia.

NVLink is an energy-efficient, high-bandwidth bus that will be used for communications between Nvidia GPUs and IBM Power processors in supercomputers as well as for multi-GPU communications. Initial NVLink data-rates will be “at least” 80GB/s, but eventually NVLink’s bandwidth will be increased to 192GB/s or even higher.

Nvidia plans to release its first “Pascal” GPUs in 2016.

Discuss on our Facebook page, HERE.

KitGuru Says: While “Pascal” GPUs will not be 10 times faster than “Maxwell” graphics processors in all types of applications, expect the upcoming GPUs to offer significantly higher performance than predecessors due to increased amount of execution units, higher clock-rates, massively increased memory bandwidth and architectural enhancements. The majority of innovations will be enabled by thinner manufacturing process.

Check Also

Intel’s discrete GPU development is already under way

Last year, the graphics industry had a major shakeup, with Raja Koduri switching to the blue team to take the reins on discrete GPU development for Intel. Since then, things have been somewhat quiet, but this week that changed with some discrete GPU prototype details making their way onto the web.