作者: vsp 時間: 2015-10-6 13:04 標題: AMD Zen CPU Microarchitecture Details Leaked In Patch
本帖最後由 vsp 於 2015-10-6 13:05 編輯
AMD ZEN唔再用推土機2個core共用一個FP,Memory,FP latency大減,效能應會好番好多
加上抄埋intel hyper threading(SMT)設計,估計效能增長多5-10%
斷估唔會好似intel,低階U,i5無得用SMT,黎分market segment.
估計AMD ZEN四核心效能(有SMT)大約等於intel i5 90%或上水平
旗艦ZEN 8核心,效能應好過I7水平
以上全部斷估
詳情自己睇
AMD Zen CPU Microarchitecture Details Leaked In Patch – Doubles Down On IPC And Floating Point Throughput
http://wccftech.com/amd-zen-cpu-core-microarchitecture-detailed/
作者: yjfoeg 時間: 2015-10-6 13:35
IPC高,但會唔會3GHz封頂先
作者: vsp 時間: 2015-10-6 13:47
AMD今次以四核心為一個單位設計ZEN CPU,估計多工方面應該幾好
單核心效能有進步,但唔好期望太過高,BD 一個module8個integer pipeline,變為一個ZEN core 4個intger pipeline,少一半,要靠SMT減少效能損失
作者: Ksec 時間: 2015-10-6 14:34
其實其他 CPU 出之後要"學" 佢地既 Design 就唔難。 所以你話 Zen Performance 有 80-90% Skylake 我都唔會覺得奇。 但係 AMD 既 弱點係 GF。
就好似最近 Apple A9 Samsung 14 node Vs TSMC 16nm Node, D 人終於醒覺 small node 唔係一切。 ( 其實係有原因, 不過起碼D 人唔會再咁盲目相信 small Node = 好 )
作者: Stiggy930 時間: 2015-10-6 15:44
本帖最後由 Stiggy930 於 2015-10-6 17:18 編輯
Well. It depends on how wide (or efficient) that 4 INT pipes are [EDIT]and whether the front end can feed them up[/EDIT]. Furthermore, in Bulldozer family, those 8 INT pipes are split into 2 group of 4 pipes to serve 2 front ends. Can't do a direct comparison like what you did.
作者: cise4832 時間: 2015-10-6 16:30
本帖最後由 cise4832 於 2015-10-6 17:58 編輯
其實其他 CPU 出之後要"學" 佢地既 Design 就唔難。 所以你話 Zen Performance 有 80-90% Skylake 我都唔 ...
Ksec 發表於 2015-10-6 14:34
公道d講句
個benchmark result細過 +- 5%
分分鐘兩部同樣tsmc vs tsmc都有咁既variance
作者: qcmadness 時間: 2015-10-6 19:41
本帖最後由 qcmadness 於 2015-10-6 19:47 編輯
邊有8個pipeline?
4個ALU + 4個L/S咋

作者: ricky1992 時間: 2015-10-7 02:16
出左先諗都未遲
作者: yjfoeg 時間: 2015-10-7 02:31
呢樣真
最快都仲有一年有多, CPU既野到時要換先算囉
作者: Kundera 時間: 2015-10-7 09:08
其實其他 CPU 出之後要"學" 佢地既 Design 就唔難。 所以你話 Zen Performance 有 80-90% Skylake 我都唔 ...
Ksec 發表於 2015-10-6 14:34
there are lots of design goal variants under the same process
MPU vs SoC, high perforamance vs low power, pitch, dielectric material, layers...
even 14LPE vs LPP could yields 10% performance difference
作者: dom 時間: 2015-10-7 11:05
AMD d 野, 出左實物測試先講啦
Powerpoint 講幾好都無用
Faildozer 未出夠話 CMT 有幾好.......
作者: Ksec 時間: 2015-10-7 11:20
there are lots of design goal variants under the same process
MPU vs SoC, high perforamance vs low ...
Kundera 發表於 2015-10-7 09:08
Exactly. Even if AMD did get everything right with ZEN. Or not, even if AMD Copied everything exactly the same from Intel they still wont be able to complete against Intel in Performance / Power Usage.
There is only a slim hope GF buying Foundry from IBM would help..... but i am not holding my breath.
作者: Stiggy930 時間: 2015-10-7 13:37
Exactly. Even if AMD did get everything right with ZEN. Or not, even if AMD Copied everything exac ...
Ksec 發表於 2015-10-7 11:20
According to latest rumors from WCCFTech, Zen may be like Apple A9 will be made by both TSMC and GF. Apple seem to be very happy with TSMC consider TSMC made most of the A9?
作者: vsp 時間: 2015-10-7 17:09
記錯,8 pipelines係AMD Streamroller
AMD Streamroller增強唔少

作者: yjfoeg 時間: 2015-10-7 17:28
其實Bulldozer -> Piledriver -> Steamroller -> Excavator 每代既進步都唔細
不過比K10.5 -> Bulldozer既退步覆蓋晒咁解...
如果當初冇Bulldozer, K10,5慢慢演化落去, 到今日支持所有新指令集同用28nm的話, 其實AMD唔會落後咁多
試諗下一粒CPU, 八核, IPC比K10高20%, 用28nm, 行4.5GHz
點會輸比Intel先得架
作者: 只是新手 時間: 2015-10-7 17:32
回覆 11# dom
比我地打住個飛機先
我睇好佢
作者: ^.^ 時間: 2015-10-7 18:18
預2017年年中出啦。
預佢去到2016年尾又話要DELAY,例牌。
作者: qcmadness 時間: 2015-10-7 20:33
Still wrong, Steamroller一樣無改呢樣野
作者: sunnykoomos 時間: 2015-10-7 21:55
intel等呢一日等到頸都長
作者: toylet 時間: 2015-10-7 22:02
提示: 作者被禁止或刪除 內容自動屏蔽
作者: yjfoeg 時間: 2015-10-8 16:31
等得黎,呢一刻生存緊嘅人類,應該有一億左右死左
作者: cise4832 時間: 2015-10-17 23:33
Zen related information in earning transcript:
Second, we must continue delivering strong new products. This includes successfully executing key design milestones for our breakout Zen CPU core. Zen remains on track for availability in 2016 and, we believe, will return AMD to the mainstream server and high-end client markets in a significant way in 2017 and beyond.
So, let me talk overall about Zen and our roadmap, and then specifically to your question.
So, as I said in the prepared remarks, Zen is on schedule for availability in 2016 and a first full year of revenue ramp in 2017. As you know, these microprocessor projects are multiyear projects, so the architecture or the execution team very much in place. I think we are pleased with the progress and we will continue to work hard to meet our objectives in that area.
Yes, so, Harlan, let me couch it this way. So as we stated in the financial analyst day, we had a target of 40% IPC performance of Zen over our previous generation. We believe we are on track for that.
Relative to process technology, we have taped out multiple products to multiple fabs in FinFET and we believe that they're also on track in terms of overall ramp. So we continue to focus on both of those aspects, both the architecture and the process technology, but so far, so good.
Yes, so the way I would think about the enterprise, embedded, and semi-custom segment in terms of longer-term revenue growth, in 2016, as we said, we will have additional semi-custom revenue ramping in the second half of 2016, so I think that would be one driver. And then as we go into the medium term with the -- with Zen, that would be more of a 2017 revenue driver.
http://finance.yahoo.com/news/ed ... ence-075552233.html
作者: vsp 時間: 2015-10-18 12:37
intel放水,出年一齊玩AMD ZEN(如果冇delay)
Intel’s 10nm Cannonlake Delayed to 2H 2017
http://wccftech.com/intels-10nm- ... ktop-chips-1h-2017/
作者: bensiu 時間: 2015-10-21 15:10
仲有一年AMD點捱…
作者: dom 時間: 2015-10-21 20:58
intel放水,出年一齊玩AMD ZEN(如果冇delay)
Intel’s 10nm Cannonlake Delayed to 2H 2017
...
vsp 發表於 18-10-2015 12:37 PM
Intel 都唔想少一個 x86 partner 去頂下 ARM ..

