HKSpot - Powered by Discuz! Board

標題: [業界消息] Engadget Interview with AMD's John Taylor [打印本頁]

作者: Puff 時間: 2013-1-11 22:11 標題: Engadget Interview with AMD's John Taylor

http://www.engadget.com/2013/01/10/amd-interview/

簡單 D 講
1. AMD 未來既設計將會更多係 SoC-mind driven，而唔係 CPU-centric
2. CPU 很重要，但有 D 野 GPU 同 fixed function 更好用 - GPU computing + 3rd party IP 啦即係
3. CPU Core 依然係大細孖，AMD 唔會飛起大孖既開發。換句話講，HP x86 CPU 同理。
4. 4C Temash peak power 可以去到 8W (同 Exynos 5 Dual 有得揮)，但 generic app power 可能係講緊 1W-2W 呢個 range.
5. 2CU for Temash

-

按我自己 speculated schedule 黎講，Excavator 係 BD 出爐以後先開始 evaluation & design 既核心。你可以計下 XV 係 get the thousands cuts cured based on SR 或者單純係 SR revision 既成數有幾大... 假設佢係 on 20nm。

[ 本帖最後由 Puff 於 2013-1-11 22:22 編輯 ]

作者: qcmadness 時間: 2013-1-11 22:15

SoC一定要啦
大細core又係一定要啦

作者: Puff 時間: 2013-1-11 22:21

引用:

原帖由 qcmadness 於 2013-1-11 22:15 發表
大細core又係一定要啦

至弊有 D 人認為細 core 可以一腳踢由上到下，只要有 many cores 就搞掂，點知上到 server 瓜到無得再瓜...。同 larrabee 鬥就可能得既，同 power 同 xeon 就難囉...。

作者: qcmadness 時間: 2013-1-11 22:22

引用:

原帖由 Puff 於 2013-1-11 22:21 發表

至弊有 D 人認為細 core 可以一腳踢由上到下，只要有 many cores 就搞掂，點知上到 server 瓜到無得再瓜...。同 larrabee 鬥就可能得既，同 power 同 xeon 就難囉...。

...

真係咁容易, ARM就唔駛Cortex A15啦

作者: Puff 時間: 2013-1-11 22:26

引用:

原帖由 qcmadness 於 2013-1-11 22:22 發表

真係咁容易, ARM就唔駛Cortex A15啦

A15 係一奇葩，聯同 anandtech 果篇 mythbuster，一齊証明當一切事物 scale 上黎既時候，最終既結果都係殊途同歸。
期待 2+ Ghz 既 Krait & A15... 同 9W Kabini 有得鬥下功耗。

作者: qcmadness 時間: 2013-1-11 22:26

引用:

原帖由 Puff 於 2013-1-11 22:26 發表

A15 係一奇葩，聯同 anandtech 果篇 mythbuster，一齊証明當一切事物 scale 上黎既時候，最終既結果都係殊途同歸。
期待 2+ Ghz 既 Krait & A15... 同 9W Kabini 有得鬥下功耗。

...

Atom結果都係無乜著數

作者: cheungmanhoi 時間: 2013-1-11 23:02

. CPU Core 依然係大細孖<--唔明

作者: qcmadness 時間: 2013-1-11 23:03

引用:

原帖由 cheungmanhoi 於 2013-1-11 23:02 發表
. CPU Core 依然係大細孖

一個low-power core arch.
一個high-performance core arch.

作者: Puff 時間: 2013-1-11 23:33

引用:

原帖由 qcmadness 於 2013-1-11 23:03 發表

一個high-performance core arch.

話唔定一陣 modular 到最後連 CMT 都無埋

作者: qcmadness 時間: 2013-1-11 23:35

引用:

原帖由 Puff 於 2013-1-11 23:33 發表

話唔定一陣 modular 到最後連 CMT 都無埋

CMT幾有用

作者: Puff 時間: 2013-1-12 01:06

Kaveri will first hit desktops in Q4. lol

作者: dom 時間: 2013-1-12 02:01

引用:

原帖由 qcmadness 於 2013/1/11 23:03 發表

一個low-power core arch. (ARM , Intel "Atom" , AMD "Bobcat" Family )
一個high-performance core arch. ( x86 / Power / etc

作者: dom 時間: 2013-1-12 02:06

http://www.pcper.com/reviews/Sho ... ri-side-Sea-Islands
WHAT THE ?!!

Socket FM2 and Socket FM2+ ?!!
...........................
and originally Trinity have PCI-e 3.0 support and got cut ?

DAMN IT AMD WTF are you doing

作者: XT 時間: 2013-1-12 14:56

引用:

原帖由 Puff 於 2013-1-11 23:33 發表

話唔定一陣 modular 到最後連 CMT 都無埋

結終太多cores有效能損失 ...

作者: Henry 時間: 2013-1-12 17:34

引用:

原帖由 Puff 於 2013-1-11 22:26 發表

A15 係一奇葩，聯同 anandtech 果篇 mythbuster，一齊証明當一切事物 scale 上黎既時候，最終既結果都係殊途同歸。
期待 2+ Ghz 既 Krait & A15... 同 9W Kabini 有得鬥下功耗。

...

2.XGHz A15 v.s. Haswell 7W......

作者: Puff 時間: 2013-1-12 17:38

引用:

原帖由 XT 於 2013-1-12 14:56 發表

結終太多cores有效能損失 ...

cores 唔會嫌多既就，當然，依家黎講桌面筆記本呢 D，4 粒強 core 係最 optimal，換句話講就係 ST performance.
所以係 shared resources 既 penalty 加上 thousands cuts 既乘數效應，造就左今日 dom 姐口中既 Failzoer。
Steamroller 係修補 Faildozer 既第一步，unshare 左個佔地不少既 decoder... 之後點發展，無人知啦。

唉，唔諗啦，反正我估遲早都會有 9+ cycles 既夾心 cache 啦。估咋。

[ 本帖最後由 Puff 於 2013-1-12 17:41 編輯 ]

作者: qcmadness 時間: 2013-1-12 18:17

引用:

原帖由 Henry 於 2013-1-12 17:34 發表

2.XGHz A15 v.s. Haswell 7W......

There is no Haswell 7w
http://discuz.servebbs.com/viewthread.php?tid=62878

作者: dom 時間: 2013-1-12 18:53

引用:

原帖由 Puff 於 2013/1/12 17:38 發表

cores 唔會嫌多既就，當然，依家黎講桌面筆記本呢 D，4 粒強 core 係最 optimal，換句話講就係 ST performance.
所以係 shared resources 既 penalty 加上 thousands cuts 既乘數效應，造就左今日 dom 姐口中既 Failzoer。
St ...

If they have enough time to fix it.....

Intel is chasing up way faster on iGPU than AMD push APU
when now some user may forgive weaker CPU power APU w/better iGPU

In the future ,APU not just face Intel Haswell, also ARM based competition....

作者: dom 時間: 2013-1-12 18:56

Mind you Intel / AMD Market share is changing, losing market share due to FAILDOZER and layoff workforce will further impact the ability for AMD to R&D and develop new product (GPU/CPU)

作者: qcmadness 時間: 2013-1-12 19:11

引用:

原帖由 dom 於 2013-1-12 18:56 發表
Mind you Intel / AMD Market share is changing, losing market share due to FAILDOZER and layoff workforce will further impact the ability for AMD to R&D and develop new product (GPU/CPU)

Due to Llano

作者: Puff 時間: 2013-1-12 21:05

引用:

原帖由 dom 於 2013-1-12 18:53 發表
If they have enough time to fix it.....

Intel is chasing up way faster on iGPU than AMD push APU
when now some user may forgive weaker CPU power APU w/better iGPU

In the future ,APU not j ...

Intel is chasing up with their manufacturing advantages, and there is no way of AMD to defeat them. For ARM, it is safe to assume ARM won't scale to notebooks and desktops in the near future - think the disastrous Windows 8 and RT. For the server space, it is generally the same case except the growing microserver market. But AMD is now expanding their own solution stack to compete with the ARM vendors, and they seems digging into the mobile space aggressively. Considering the public has a good response to Temash, it won't be too bad for AMD in the next 2 years in my opinion.

[ 本帖最後由 Puff 於 2013-1-13 01:31 編輯 ]

作者: qcmadness 時間: 2013-1-12 23:52

引用:

原帖由 dom 於 2013-1-12 18:53 發表
If they have enough time to fix it.....

Intel is chasing up way faster on iGPU than AMD push APU
when now some user may forgive weaker CPU power APU w/better iGPU

In the future ,APU not j ...

don't think so

作者: Puff 時間: 2013-1-13 02:29

引用:

原帖由 qcmadness 於 2013-1-12 23:52 發表

don't think so

Given that AMD has been ceiled by the memory wall, it shall be the situation except for the driver. I don't either expect any miracle or dirty solution on Kaveri.

[ 本帖最後由 Puff 於 2013-1-13 02:34 編輯 ]

歡迎光臨 HKSpot (https://bbs.hk-spot.com/)