Új hozzászólás Aktív témák

• #37409 Raymond félisten Alogonomus #37404

  Új Válasz 2022-02-24 14:36:12 #37409

  Új hozzászólás

  Összes hozzászólása itt Válaszok az összes hozzászólására itt Válaszok erre a hozzászólásra

  Privát üzenet küldése

  

  Raymond

  félisten

  válasz Alogonomus #37404 üzenetére

  Abban a tablazatban hatulrol (az 5nm-rol) indulnak egy bizonyos spec chip-el es a nagyobb node-oknal aszerint szamoljak mennyi jonne ki egy wafferbol hogy mekkora lenne az a chip az adott node-on az 5nm-hez kepest:

  Chip transistor density. Our model uses, as a baseline, a hypothetical 5 nm
  GPU with the specifications of Nvidia’s Tesla P100 GPU, which OpenAI used
  in 2018 to train the breakthrough AI algorithm OpenAI Five. 191 The P100
  GPU is fabricated at TSMC at the 16 nm node and contains 15.3 billion
  transistors in a chip (die) area of 610 mm2 , translating to a transistor density
  of 25 MTr/mm2 . 192 A 300 mm diameter silicon wafer produces 71.4 610
  mm2 GPUs on average. 193 Our hypothetical 5 nm GPU has a chip area of
  610 mm2 and given its greater transistor density than the P100 GPU, 90.7
  billion transistors. 194 Table 7 presents estimated TSMC transistor densities for
  nodes between 90 and 5 nm. For nodes in the 90 to 7 nm range, our model
  uses a hypothetical GPU with identical specifications, including transistor
  count, as the hypothetical 5 nm GPU, except with a transistor density
  associated with the hypothetical node. Therefore, GPUs with nodes larger
  than 5 nm will respectively have an area greater than 610 mm2 , resulting in
  differing numbers of GPUs fabricated per wafer as shown in Table 7.
  However, the model could equivalently be interpreted as accounting for one
  chip at the 5 nm node, but at any given larger node, multiple chips totaling
  the same transistor count as one 5 nm chip.

  Forras: az itt letoltheto PDF 41. oldala [link]

  Privat velemeny - keretik nem megkovezni...

Új hozzászólás Aktív témák