NVIDIA are introducing their next generation Pascal GPU in 2016 which can introduce many new and key technologies to the green team. The Pascal GPU are the successor to the present generation Maxwell GPU and from the appearance of it, it's about to be a beast of a chip. that includes the most recent HBM2 and 16nm FinFET based mostly styles, Pascal GPUs can leverage NVIDIA’s dominance in each the end user & corp. world.
NVIDIA Pascal GPU may Feature seventeen Billion Transistors, nearly double The Transistors of Fiji
In an exclusive report printed by Fudzilla, the website reveals that NVIDIA’s next generation Pascal GPU can feature seventeen billion transistors crammed within its core. Currently, the flagship GM200 core found on the GeForce GTX Titan X comes with 8.0 Billion transistors whereas the contender, the Radeon R9 Fury X incorporates a total of 8.9 Billion transistors within its Fiji GPU. The seventeen Billion transistors on the Pascal GPU ar double the transistors found on the GM200 Maxwell and therefore the Fiji XT GPU core that is virtually insane. Pascal is supposed to be NVIDIA’s next high performance, cipher centered graphics design which is able to be found on all market segments that may embrace GeForce, Quadro and even Tesla. In accordance to TSMC’s 16nm process node, NVIDIA’s Pascal GPU won't solely feature the most effective performance in graphics however additionally the foremost power economical design ever created by a GPU manufacturer.
It was discovered a number of days past that NVIDIA’s Pascal GP100 chip has already been taped out on TSMC’s 16nm FinFET method, last month. this suggests that we will see a launch of those chips as early as Q2 2016. providing the electronic transistor count is correct, we will expect a progressive performance increase from Pascal across the vary of graphics cards that may be introduced.
TSMC’s 16FF+ (FinFET Plus) technology will offer on top of sixty five percent higher speed, around two times the density, or 70 % less power than its 28HPM technology. comparison with 20SoC technology, 16FF+ provides additional 40% higher speed and 60% power saving. By investing the expertise of 20SoC technology, TSMC 16FF+ shares identical metal backend method so as to quickly improve yield and demonstrate process maturity for time-to-market price.
The seventeen Billion transistors are an insane quantity however what’s a lot of insane is that the amount of VRAM that's about to be featured on the new cards. With HBM2, NVIDIA gets the leverage to feature much more memory than what’s presently allotted on HBM1 cards (4 GB HBM on Fury X, Fury, Nano, Fury X2). With HBM2, NVIDIA gets access to a lot of denser chips that may lead to cards with 16 GB and up to 32 GB of HBM memory across a vast 4096bit memory interface which is able to dominate succeeding high-resolution 4K and 8K game panels.Although they will ought to wait a bit bit longer due to AMD’s priority access to HBM2 with SK Hynix, the manufacturers of HBM. With 8Gb per DRAM die and 2 Gbps speed per pin, we have a tendency to get more or less 256 GB/s transmission per HBM2 stack. With four stacks in total, we are going to get 1 TB/s transmission on NVIDIA’s GP100 flagship Pascal that is double compared to the 512 GB/s on AMD’s Fiji cards and thrice that of the 980 Ti’s 334GB/s.
The Pascal GPU would additionally introduce NVLINK that is that the next generation Unified computer memory link with gen 2.0 Cache coherency options and 5 – 12 times the transmission rate of an everyday PCIe connection. this can solve several of the transmission flow problems that top performance GPUs presently face. one among the most recent things we have a tendency to learned regarding NVLINK is that it'll permit many GPUs to be connected in parallel, whether or not in SLI for gaming or for work usage. Jen-Hsun specifically mentioned that rather than four cards, users are ready to use eight GPUs in their PCs for gmae and work use.
With Pascal GPU, NVIDIA can come back to the HPC market with new Tesla product. Maxwell, though nice altogether regards was lacks necessary FP64 hardware and targeted solely on FP32 performance. This meant that the chip was about to be from HPC markets whereas NVIDIA offered their year previous kepler cards because of their solely Tesla based choices. Pascal won't solely improve FP64 performance however additionally feature mixed preciseness that permits NVIDIA cards to cipher at 16-bit at double the accuracy of FP32. this suggests that the cards can change 3 tiers of cipher at FP16, FP32 and FP64. NVIDIA’s way future volta GPU can additional leverage the cipher design because it is already planned to be a part of the SUMMIT and Sierra super computers that feature over 150 PetaFlops of compute performance and launch in 2017 that indicates the launch of Volta simply a year later Pascal for the HPC market.
0 Comments:
Post a Comment