Stiggy930

其實點解唔直接用14nm FinFET，Samsung整得唔錯啊，到時溫度肯定低好多
sunnykoomos 發表於 2015-6-4 01:08

Risk production as named has risks. Not many company especially company like AMD is willing to take the risk.

And this prolly taped out way before Samsung's/GF's risk production.

Stiggy930

http://semiaccurate.com/2015/06/03/amd-shows-wont-say-fiji/

HD img:


http://wccftech.com/amd-fiji-xt- ... nterposer-pcb-size/

The actual chip is not that big. The inter-poser under the GPU and HBM however is big in term of surface area.

戀人未滿

我估有65mm * 65mm

manu_yip

所有係咭上最熱嘅chip都集中埋一齊，唔熱就假!!!
bigpo 發表於 2015-6-3 23:17

    果四粒HBM無行到GDDR5唔高頻, 應該無咁熱掛, 不過佢stack上去, 真係好難講

manu_yip

其實點解唔直接用14nm FinFET，Samsung整得唔錯啊，到時溫度肯定低好多
sunnykoomos 發表於 2015-6-4 01:08

    唔好淨係講14nm FinFET, Samsung本身都好似無乜做High power core嘅經驗?

    再者, 佢yield好似唔係高到服務到apple, 又服務到AMD, nV

manu_yip

Risk production as named has risks. Not many company especially company like AMD is willing to tak ...
Stiggy930 發表於 2015-6-4 10:00

    咁我估nV應該唔想再用28nm, 等TSMC 16nm FinFET+或者Samsung/GF 14nm提高產能
    AMD會跟住die shrink?

akiro

希望佢生生性性 比到個表現出黎

stefac03

nv大粒=落足料
amd大粒=低良率
e派 發表於 2015-6-4 00:18

AMD可以學NV推足4~5代>500 mm^2既GPU的話
Fanboy自然會收聲，正如當年GT200個堆。

jt9862

R9 Fury X