打印

[硬件] KAVERI 幾時先有消息

引用:
原帖由 cheungmanhoi 於 2013-10-30 10:15 發表

咁 pure cpu果d fx都係落fm2+?
定係唔會有 fx
no FX

because I suspect AMD will go 20nm in 2015

TOP

引用:
原帖由 qcmadness 於 2013-10-30 10:16 發表
because I suspect AMD will go 20nm in 2015
Delay several quarters and jump to finfets! </holy crap>

I am a bit skeptical if there is FX successor on 20nm. I suspect still a no, unless Excavator is really evaluated to be competitive with Haswell Xeon counterparts, and can be released in early 15. Otherwise 16 for any non-APU is a safer assumption, as we will arrive the interchange station at that time for a better train.


[ 本帖最後由 Puff 於 2013-10-30 14:49 編輯 ]

TOP

引用:
原帖由 cheungmanhoi 於 2013/10/30 10:15 發表

咁 pure cpu果d fx都係落fm2+?
定係唔會有 fx
No FX, AMD is giving up FX + AM3+

TOP

引用:
原帖由 Puff 於 2013/10/30 10:44 發表

Delay several quarters and jump to finfets!

I am a bit skeptical if there is FX successor on 20nm. I suspect still a no, unless Excavator is really evaluated to be competitive with Haswell Xeon cou ...
By that time AMD RIP already

TOP

引用:
原帖由 dom 於 2013-10-30 18:05 發表


No FX, AMD is giving up FX + AM3+
THAT MEANS NO MORE FX?
ONLY APU?


AMD

TOP

引用:
原帖由 cheungmanhoi 於 2013-11-1 16:41 發表

THAT MEANS NO MORE FX?
ONLY APU?


AMD
In short term, yes & they are pushing APU/GongGongAthlon+dGPU solutions.


In long term, no one knows.

TOP

引用:
原帖由 Puff 於 2013-11-1 20:06 發表

In short term, yes & they are pushing APU/GongGongAthlon+dGPU solutions.


In long term, no one knows.



a10 5800k
有無 i5 lv cpu(mobile)

TOP

引用:
原帖由 Puff 於 2013/11/1 20:06 發表

In short term, yes & they are pushing APU/GongGongAthlon+dGPU solutions.


In long term, no one knows.
Their APU solution only competitive with GPU performance , but INTEL already catching up and
the rubbish GF production issue (poor yield rate) is killing AMD
天然系長髮眼鏡娘 最高
Lucky Star 聯盟 - 美幸
Kancolle - 大淀, 翔鶴 (太太), 烏海 , 瑞鶴

TOP

引用:
原帖由 cheungmanhoi 於 2013-11-4 10:34 發表




a10 5800k
有無 i5 lv cpu(mobile)
u mean quad-core or dual-core?

TOP

引用:
原帖由 Puff 於 2013/11/5 22:59 發表

u mean quad-core or dual-core?
更係Quad Core 啦
天然系長髮眼鏡娘 最高
Lucky Star 聯盟 - 美幸
Kancolle - 大淀, 翔鶴 (太太), 烏海 , 瑞鶴

TOP

引用:
原帖由 Puff 於 2013-11-5 22:59 發表

u mean quad-core or dual-core?
勁唔勁得過
有無 i5 desktop版咁勁?
or 35w mobile咁勁

TOP

引用:
原帖由 cheungmanhoi 於 2013-11-6 10:49 發表

勁唔勁得過
有無 i5 desktop版咁勁?
or 35w mobile咁勁
average multi core performance
A10-5800K ~= i3-2100 < i3-3220 <<< i5 << i7

single thread performance
think it yourself

mainstream 2C4T i5 = lower clock i3 in desktop
ultra-expensive 4C8T i7 = lower clock i7 in desktop


for kaveri or any kinds of AMD products in the short term, lowering your expectation is always better. let it surprise you rather than disappoint you. I would say +30% is kind of unrealistic NOT because I am an AMD hater, but 4M4C result shows you that twice the decoders don't benefit significantly at all (but overprovision is needed as this is the price of IPC). Although I didn't account of any other potential architectural changes here, I'm pretty sure that 10% is a safe bottom line. 15% will be appreciated. above 20% will be a surprise.

rumour has it that SteamrollerB has made some changes to make it more "competitive". at design level it seems to be a derivative of a completed rev A (linkedin info). but even we know this (I guess it is probably changes aligned to crappy GF's process change), it doesn't help much as the problem is in the fundamental of this family of architecture, and from fixing it in the HLD level to the final product shipping takes you nearly a half decade.



[ 本帖最後由 Puff 於 2013-11-6 15:29 編輯 ]

TOP

引用:
原帖由 dom 於 2013-11-6 01:53 發表


更係Quad Core 啦
依家先知 i5 mobile 無 QC

TOP

引用:
原帖由 Puff 於 2013-11-6 15:15 發表

依家先知 i5 mobile 無 QC
即係咁.....
我用緊粒i5 3210m
35w cpu 2c4t


amd粒58k 用50w(我當gpu 用左50w)
都追唔上
我用粒3210 stack相做d heavy load都覺得慢(當然....i7都唔係快超多)


但fx u 用sb950真係sb lor

[ 本帖最後由 cheungmanhoi 於 2013-11-6 17:01 編輯 ]

TOP

CineBench R15, 1 (for 1C)/2 (for 2C) rendering thread, same bucket of background applications
@ 3.67 Ghz
2M2C: ~153 cb
1M2C: ~131 cb
1M1C: ~76 cb (scaled down from 78 cb @ 3.71-3.84 Ghz)

yield ~70% over 1M1C when having only one more integer core
yield ~16% over 1M2C when having more decoder and FPU resources & L2 caches
yield 100% when everything is doubled



side note:
Thuban K10 1C @ 3.67 Ghz: est. 97
Haswell 1C: 141?

[ 本帖最後由 Puff 於 2013-11-6 17:30 編輯 ]

TOP