Puff

http://www.linkedin.com/in/saketdoshi?_mSplash=1

Puff

引用:

原帖由 qcmadness 於 2014-2-17 21:30 發表

下代, 唔係再下代, 即係明年

但無論點篤，條數都係爛。當仲有代 20nm 排隊，16/14nm 條人龍排到 Fall 16 打後喎...
除非佢地打算 3Q 超短週期，或者呢級 APU 跳 20nm 直上 FinFET
查實 ARM 64-bit 都係差唔多情況，Seattle 賣得果 3-4 季度都已經 Fall 14 甚至 15 年啦，15 年得黎就全世界磨拳擦掌等上 FinFET。

真心亂估：最後會唔會得 GPU 同 Cat 用 20nm


Late edit:
http://www.linkedin.com/pub/rich-schultz/73/459/29b
14nm FinFET, DDR PHY design in progress... 會唔會開口中


[ 本帖最後由 Puff 於 2014-2-18 08:20 編輯 ]

Puff

引用:

原帖由 dom 於 2014-2-18 23:21 發表
AMD 咁整法　真係等 RIP
真係太慢

唔駛 Intel 出手

咁搞法悲觀估計 PC 16 年/17 年先叫數字上追平一兩個季度，甚至無。伺服器就一年半載
仲未計佢地完全 quit ge 可能，雖然唔似. substantial share in x86 CPU market 嘛

引用:

原帖由 dom 於 2014-2-18 23:21 發表
ARM 可以收佢工

佢自己咪搞緊 ARM，但如果佢地 ARM products 如果都係跟呢個工藝 schedule，有可能又係一 pat 野。因為依家咁睇推土機呢條 line 最快都要 fall 15 先有 20nm* (BTS/Xmas)...呢個時間 16nm BCOM ARMv8 都話預計量產添啦. 當然佢地依家可能轉走保守路線都唔定...，等煮熟晒先上菜。



* 台灣人話 28KV>28CZ>20 之後推土機先玩完，然後據 LNKIN 資料有最少一粒 28GF chip port to 20nm，雖然可能係 seattle.

[ 本帖最後由 Puff 於 2014-2-18 23:56 編輯 ]

Puff

引用:

原帖由 dom 於 2014-2-19 18:33 發表
Cancel faildozer la

Too epic fail to be fixed

canned la. only APU variants survive.

tw insider said 20 is last gen, i guess carrizo will be ported. not sure if 20 and 16/14 products will co-exist or not, say price charged by foundries beyond 28 is said to not scale down further anymore (lower $/xtor is previous driver of process migration IIUC). if 16/14 charges more than 20 per transistor... shall we see 20 fills the low-end & 16/14 fills mid to high-end? in this way we can dream for a rapid release of 16/14 products following 20 in one or two quarters.



[ 本帖最後由 Puff 於 2014-2-20 18:47 編輯 ]

Puff

引用:

原帖由 qcmadness 於 2014-2-21 00:56 發表

20nm Bulldozer variant應該好似Llano咁既stop-gap solution

只係覺得按 3-4Q *2 咁計排到 16Q4/17Q1 好肉酸
雖然再之前 28nm CPU (@14H2) 仲在生果時已經預左最早 early 16H2 見人
堆低功耗應該會早 D ... 掛。



[ 本帖最後由 Puff 於 2014-2-21 12:40 編輯 ]

Puff

引用:

原帖由 dom 於 2014-2-21 17:32 發表


Faildozer is causing them more harm than help.....
Especially now Intel is striking back hard with new Atom to compete with ARM
CAT is good but when Intel push hard on it won't last very long (Cat  ...

it was (ALWAYS!) girlfriend's fault I'd say, which causes AMD failed to hold the ground with Bobcat28 and earn more before the market is driving off the cliff. but everything related to cat and BD is settled now and pretty much predictable, so... eh, strategy beyond XV is a more interesting discussion - of course with the assumption of AMD's survival.



[ 本帖最後由 Puff 於 2014-2-21 17:53 編輯 ]

Puff

20nm A57 for tablet and dense server from AMD



[ 本帖最後由 Puff 於 2014-2-25 21:37 編輯 ]

Puff

引用:

原帖由 qcmadness 於 2014-2-25 22:57 發表

Jaguar is a competent architecture...

In this way we can sure we will have Cat on 20nm... due to the Great Design Recycling.
Still a big question mark for "high-perf APU" line.



[ 本帖最後由 Puff 於 2014-2-25 23:09 編輯 ]

Puff

引用:

原帖由 qcmadness 於 2014-2-25 23:08 發表

Should not be Cat on 20nm, maybe a new architecture

leopard la.
http://www.linkedin.com/pub/daniel-luke/5b/507/245

Puff

引用:

原帖由 qcmadness 於 2014-2-25 23:14 發表

This is for power consumption reduction

leopard should be a new revision, tho jaguar will likely be shrunk for PS4/XB1 (linkedin info). formosa insider also said 1xFF for custom arm. zen after cat and bd (1xFF implied). everything seems fit. but honestly speaking i just wish i can get a GreenArrow inside rmbp that got a HBM stack & allows me to play with hsa.


looking forward to kaveri notebook announcements. but MWC has started, hasn't it? too quite it seems.


[ 本帖最後由 Puff 於 2014-2-25 23:55 編輯 ]

Puff

big guy on 20