Puff

2013 roadmap changes again.


95/125W Vishera CPU 照舊，<25W Kabini APU 照舊。15-100W Kaveri APU 卻是消失了，換上了 feature set 同 Trinity 接近的 Richland。
Richland is w/ Piledriver cores. 是不是 GCN 不知道，但據我所知 HSA 用 Trinity APU 也是能跑的。
製程未知。但係，名義上既 2013 HSA APU 只會有 Low-power 的 Kabini 和伺服器的 28nm 1-2P APU 就是鉆板上的事實了。

猜測
1. Trinity 和 Kaveri 的效能差距不大，而且接連後三代既 life cycle 係某 D 因素既影響下，於是 Kaveri 成了被宰的目標。
2. AMD 放緩開發週期，只係製程更新時先會更新設計
3. 跳過 28nm，直接向 20nm w/ TSV 進發
4. 抽調人手搞 ARM Server CPU
5. 最壞情況是這一級的 APU 取消，桌面直接用 Server APU，Notebook/Ultrathin 用閹割既 Server APU 同 Low-power APU 撐大局

[ 本帖最後由 Puff 於 2012-11-2 19:40 編輯 ]

Puff

引用:

原帖由 qcmadness 於 2012-11-2 19:40 發表
2,3,5都唔可能

3 有可能的，最少 2014 年也該是 20nm 了。TSV 就話可能性不高。
2 只是 1 的進階版。但倒也是可能。這是 OEM adoption 問題，Llano 可見一斑。



[ 本帖最後由 Puff 於 2012-11-2 19:45 編輯 ]

Puff

引用:

原帖由 qcmadness 於 2012-11-2 19:44 發表

2要咁做既話, 要Haswell同Broadwell唔推出先得
3更加搞笑, 2014唔會有大量20nm fab cacacity

信不信由你。20nm 2013 年中開始 ramp up. 即使不多，也不會少。

Puff

引用:

原帖由 jackli 於 2012-11-2 19:46 發表
無辦法,
好多人都只系用電腦上下facebook, 少少game..
依家手機都做到..

從來都系市場決定生死, 不是核心數目, 不是時脈.
我要勁cpu -> intel,
要勁gpu -> Nv,
要方便淨上網 -> smart phone

咁amd 有咩用?
所謂 "A ...

trad PC market growth 是 kind of disinflation，不是 deflation.
至於 APU... 見仁見智。天時地利人和未到。

Puff

引用:

原帖由 qcmadness 於 2012-11-2 19:49 發表

賭$100, 2013年尾無50k wpm以上

2014Q2 EPC$100

Puff

引用:

原帖由 qcmadness 於 2012-11-2 19:50 發表

HK$100, 2013 Q4無50k wpm

賭左我輸硬。2013 會有先怪，今年年尾先 risk production，係都係堆大 GPU 上陣先。

Puff

引用:

原帖由 qcmadness 於 2012-11-2 19:52 發表

你知就好啦, 咁明年唔用28nm, 等死嗎?
32nm > 28nm最少die size少10-20%

我講直接跳過 28nm 上 20nm 係講跳過 Kaveri 呢個 design，juz like Brazos -> Kabini.
而且 32nm 同 28nm 係咪真係有呢種 die size reduction 都係成疑，一黎唔同 process 二黎實際既 characteristics 唔知。好似 GF 14XM/TSMC 16SoC 咁，話就話係 16/14nm 姐，事實上係 hybrid 呢。

Puff

引用:

原帖由 qcmadness 於 2012-11-2 19:56 發表

跳過Kaveri會幾大鑊下
A記學製程係有learning curve的

Kaveri 之後就係 20nm. 唔講鳥，反正唔係 delay 就取消架啦，就如同米奇係老鼠。
最多咪 delay 到 14Q1. 再遲就取消架啦。

Puff

new info:
Fuad suggests that Richland is a 28nm chip and it is expected to be launched in mid 2013, meaning that Richland could probably be a 28nm "shrink" of Trinity. If that's true, I'd rather think Kaveri is canned, and the sucessor in 2014 will be Excavator-based thing on 20nm instead of a delayed 28nm Kaveri.

P.S. Richland is used to be the codename of 2C Trinity.
P.S.2. 原本以為傳說中既 review and changes in project 只係會郁 2014 及以後既野，點知呢... 半年都未夠。



[ 本帖最後由 Puff 於 2012-11-3 01:33 編輯 ]

Puff

話時話，TN 桌面版遲過 NB 版 5 個月... 會唔會 Kaveri NB 係 Q3/Q4 Launch... 然後桌面版 14Q1 見？

Puff

Mommy tells that the brother is still doing his homework, but it takes a bit more time to finish than expected. His young brother is so smart that his homework was finished last night. Mommy is checking his homework now. Ouch, the brand new Richland theme park will probably have 32 attractions, I think? Perhaps Mommy will take them to the theme park in the beginning of 2013. It is too crowded now.

Mommy is naughty. muhahahahaha.

[ 本帖最後由 Puff 於 2012-11-12 22:49 編輯 ]

Puff

AMD to release not only ARM CPU but also ARM APU in 2014

Puff

引用:

原帖由 qcmadness 於 2012-11-15 18:46 發表

APU SoC?

I am not sure how deep the SoC integration is, but I don't think secondary I/O will be enabled in processors covering the general-purpose server market, which generally uses a unified sockets for multi-processor scaling. Every socket having secondary I/O capability will be kind of waste... though it could be affordable as a trade-off to power and latency.

So BGA socket for IHVs interested in SoC/fabric, and LGA socket + FCH for IHVs that want generational compatibility?



[ 本帖最後由 Puff 於 2012-11-18 17:53 編輯 ]

Puff

引用:

原帖由 dom 於 2012-11-15 19:57 發表


Well good luck to them

Judging by their situation, selling one design into multiple markets seems the smartest way to go on. Thus, I assume the ARM APU from AMD will be relatively small and designed for power efficiency, as compared to the throughput/performance-driven x86 thing. But will it fit the tablet market at the same time? I have no answer. So having enough customer to cover the R&D cost is my question to the new ARM business of AMD.  

複製內容到剪貼板

代碼:

AMD products in 2014 =
[GPU] 2-3 GPU ASICs (VI)
[x86] + 1 mainstream APU for PC + 1 low power APU for new PC form factors & embedded systems + 1 performance APU + 1 x86 CPU
[ARM] + 1 ARM CPU SoC + 1 ARM APU
[???] + perhaps a mobile APU designed for ULP devices, with integrated ISP( and baseband?). hybrid? x86? ARM?
[SCB] + Xbox processor + Playstation processor

[ 本帖最後由 Puff 於 2012-11-18 18:02 編輯 ]

Puff

引用:

原帖由 dom 於 2012-11-18 22:28 發表
If they don't F___ up the driver again they maybe ok

Driver... let's not talk too far. If the APU is 200 mm^2 kind of thing and designed for consuming tens of watts, will you fit it into a tablet? Anyway, perhaps they are designing these chips for sub-one watt to 10 watts, and filling the downside of their power-scaled product portfolio. Then we may see one chip (with IP-block-level power gating) for two to three markets - say an ARM APU for microservers, embedded systems and high-end mobile devices.



What I can say is... AMD has no clear picture to the public - to both its existing markets and the emerging markets it's heading into. AMD tells everyone that they will have something for someone in the future, but the picture is in the mist, and nothing is shown up. The roadmaps seem changing again and again, and there is still no clear guidance in performance improvements or advancement in feature set. I understand their situation, but hiding all the things from the public won't help your reputation. It'll just be hurt.



[ 本帖最後由 Puff 於 2012-11-18 23:42 編輯 ]