{"id":78867,"date":"2015-07-22T12:00:00","date_gmt":"2015-07-22T10:00:00","guid":{"rendered":"http:\/\/mygaming.co.za\/news\/?p=78867"},"modified":"2015-07-22T12:07:35","modified_gmt":"2015-07-22T10:07:35","slug":"nvidias-gp100-pascal-gpus-to-feature-4-hi-hbm2-and-a-4096-bit-memory-bus","status":"publish","type":"post","link":"https:\/\/mygaming.co.za\/news\/hardware\/78867-nvidias-gp100-pascal-gpus-to-feature-4-hi-hbm2-and-a-4096-bit-memory-bus","title":{"rendered":"NVIDIA\u2019s GP100 Pascal GPUs to feature 4-Hi HBM2 and a 4096-bit memory bus"},"content":{"rendered":"

As impressive as AMD\u2019s Radeon R9 Fury X2 will undoubtedly be<\/a>, especially given how well multiple Radeon R9 Fury X\u2019s scale<\/a>, all eyes have turned to NVIDIA in anticipation of their latest and greatest, next generation GP100 GPU, Pascal.<\/p>\n

Much like the Radeon R9 Fury X<\/a> and the rest of the Fiji range, NVIDIA\u2019s Pascal GPUs are set to make use of massive 4096-bit memory busses.<\/p>\n

Moreover, NVIDIA will also make use of HBM memory<\/a> in their latest flagship series, only they\u2019ll be adopting the second generation of HBM, HBM2.<\/p>\n

\"Hynix<\/a>

Hynix HBM & HBM2 Specifications. 256GB\/s is, well, massive.<\/p><\/div>\n

HBM2 promises significant performance as well as capacity enhancements over HBM and will allow NVIDIA to stack DRAM dies 4-Hi on their consumer-level GP100 GPUs. Their industrial-grade, professional Quadro and TESLA GPUs will receive 8-Hi stacks.<\/p>\n

Keep in mind that with 4-Hi HBM2 running at 1GHz on a 4096-bit memory bus, Pascal is set to have an absurd throughput \u2013 we\u2019re talking the same levels of bandwidth Jean-Luc Picard had to make use of on his Sovereign-class USS Enterprise starship.<\/p>\n

It\u2019s why NVIDIA\u2019s roadmap has Pascal, in comparison to older technology, in a league of its own.<\/p>\n

\"NVIDIA<\/a>\"NVIDIA<\/a>\"NVIDIA<\/a><\/p>\n

Throw in Pascal\u2019s TSMC 16nm FinFET process<\/a>, allowing for significantly denser and more power efficient GPUs than was possible on Maxwell 2\u2019s<\/a> 28nm process, and you have a heck of a lot of potential under the GP100\u2019s hood.<\/p>\n

TSMC\u2019s 16FF+ (FinFET Plus) technology can provide above 65 percent higher speed, around 2 times the density, or 70 percent less power than its 28HPM technology. Comparing with 20SoC technology, 16FF+ provides extra 40% higher speed and 60% power saving. By leveraging the experience of 20SoC technology, TSMC 16FF+ shares the same metal backend process in order to quickly improve yield and demonstrate process maturity for time-to-market value.\u201d<\/p><\/blockquote>\n

And just because that\u2019s not nearly enough for NVIDIA, Pascal will sport their newly developed NVLink technology, allowing for 5 to 12x faster transfer speeds than traditional PCIE 3.0 between the GPU and the rest of the PC.<\/p>\n

NVIDIA\u00ae NVLink\u2122 is a high-bandwidth, energy-efficient interconnect that enables ultra-fast communication between the CPU and GPU, and between GPUs. The technology allows data sharing at rates 5 to 12 times faster than the traditional PCIe Gen3 interconnect, resulting in dramatic speed-ups in application performance and creating a new breed of high-density, flexible servers for accelerated computing.\u201d<\/p><\/blockquote>\n

It should offer significant improvements for home PCs too.<\/p>\n