POPPUR爱换

标题: Sandy Bridge ES已经大量流入相关人士手中了 [打印本页]

作者: itany 时间: 2010-4-23 00:23
标题: Sandy Bridge ES已经大量流入相关人士手中了
本帖最后由 itany 于 2010-4-23 00:24 编辑

Sandy Bridge rumorsThe numbers
by Sascha Krohn
April 21, 2010
AN INFAMOUS ENTHUSIAST at XtremeSystems called "JCornell" has posted what seems to be the first information about Intel's next generation processor codenamed Sandy Bridge. Sandy Bridge is Intel's 32nm tock and is essentially a Clarkdale 32nm CPU with the 45nm GMCH shrunken to 32nm and merged onto the CPU die. Thats right, while AMD has been talking a lot about fusion for half a decade, Intel seems to be getting there first.

The CPUZ and Everest screenshots show an "A2" stepping, dual core, four thread, Sandy Bridge processor.

So whats new? First of all and most importantly, the socket. While the socket looks pretty much identical to an LGA1156 socket, it's not. Sandy Bridge will use the VRM 13 spec, which means it needs a PWM (power management) redesign and won't work in older LGA1156 mainboards, while current socket LGA1156 CPUs won't work in the new socket LGA1155 boards. Note how the CPU socket is actually labeled 115X. Maybe Intel will decide to remove a few pins shortly before the launch just as they did with the LGA1156 socket to confuse everybody once again.

The block is 100MHz instead of 133MHz and we heard confirmation that this time around Intel finally managed to get rid of the ck505 on the boards, something Intel wanted to do on socket LGA1366 already. This means all the clocks are created and managed internally, which has overclockers worried. According to rumors Sandy Bridge (115X) won't overclock as nicely as Bloomfield (1366), Gulftown (1366), Lynnfield (1156) or Clarkdale (1156) processors, which have a default base clock of 133MHz and reach 200MHz with aircooling easily and up to 240MHz if you're lucky.

The cache structure has apparently been updated and tweaked as well. Compared to Intel's latest 32nm Clarkdale processor Sandy Bridge has double the L1 Instruction cache width, 8-way versus 4-way, a smaller L3 cache of only 3MB versus 4MB, and a narrower L3 cache interface of only 12-way associativity versus 16-way.

It also supports AVX, an evolution of SSE which is going to replace that and doubles the vector registers' width from 128-bit to 256-bit.

This should boost the processor's floating-point performance, which is actually something games seem to like, but it is mostly aimed at the scientific field and is somewhat of a secret weapon for Intel against GPGPU. While Linux already supports AVX, Windows doesn't. Windows 7 SP1 is supposed to support it, but that's still at least a couple of months away and should come out at the end of the year. Which is another hint at the Christmas season.

And don't expect a notable performance boost. This is pretty much the same step Intel has taken in its Atom CPU lineup from Diamondville to Pine View, that is, shrinking the chipset and merging it into the CPU core, so the main focus is cost and power consumption. While AVX should improve floating-point performance, which games do like, we don't know whether the new registers will boost the performance of SSE code in current games as well, or if so by how much.

Where Sandy Bridge will be very interesting is in the mobile segment, which is where Intel is apparently planning to launch these chips first. This makes a lot of sense as Clarkdale in its mobile incarnation called Arrandale has so far failed to replace Intel's CULV lineup, with UM processors nowhere to be found and laptops based on it being delayed with no ETA. There have been hints from industry insiders that Intel couldnt get Arrandale power consumption low enough, not at reasonable yields at least, and has told its customers to wait for a couple of months. It might just skip the ultra low voltage Arrandales and go for Sandy Bridge instead.
Since most of the chipset has been shrunk to 32nm and integrated onto the CPU die, it should offer better dynamic power control and consume much less power overall and especially in idle compared to the 45nm GMCH on Clarkdale processors. While Arrandale processors come in three different TDP groups - M (35W), LM (25W), and UM (18W) - Sandy Bridge processors might knock that down a couple of watts and offer additional power savings that can't be expressed as TDP, such as idle power consumption and video playback power consumption, etc. Sandy Bridge has potential to enter Atom TDP territory while offering much better performance, something many people have been dreaming about ever since they bought a netbook.

In general there has been a trend of people getting netbooks, then realizing that they were too slow and looking for an alternative that just doesn't exist so far. CULV offers better performance but is still rather slow and consumes more power than Atom, so batteries don't last that long either. Sandy Bridge should fit perfectly here and should be fast enough. Equipped with the right battery, it might finally make the dream of a full working day long battery life come true.S|A

作者: reashion111 时间: 2010-4-23 00:24
全市鸟文看不懂

作者: itany 时间: 2010-4-23 00:24

全市鸟文看不懂
reashion111 发表于 2010-4-23 00:24

看不懂无所谓，看看图就行了，反正是半真半假的

作者: cfz45 时间: 2010-4-23 00:28
提示: 作者被禁止或删除内容自动屏蔽

作者: AlcatrazX 时间: 2010-4-23 09:14
提示: 作者被禁止或删除内容自动屏蔽

作者: lqhpop2007 时间: 2010-4-23 10:18
哇，那个115X主板也有了，牛。

作者: 吗都要 时间: 2010-4-23 10:43
完了完了，杯具要上演了

作者: cheng1088 时间: 2010-4-23 12:09

And don't expect a notable performance boost. This is pretty much the same step Intel has taken in its Atom CPU lineup from Diamondville to Pine View, that is, shrinking the chipset and merging it into the CPU core, so the main focus is cost and power consumption.

不用期待可观的性能提升。这和intel把Atom从Diamondville升级到Pine View一样，缩小芯片组整合进cpu，主要目的是成本和功耗

FUCK intel
I'm going to AMD

作者: amd```fans 时间: 2010-4-23 12:16

不用期待可观的性能提升。这和intel把Atom从Diamondville升级到Pine View一样，缩小芯片组整合进cpu，主要目的是成本和功耗

FUCK intel
I'm going to AMD
cheng1088 发表于 2010-4-23 12:09

however, amd is still being fucked by i7 920 which released in 2008

作者: itany 时间: 2010-4-23 12:24

不用期待可观的性能提升。这和intel把Atom从Diamondville升级到Pine View一样，缩小芯片组整合进cpu，主 ...
cheng1088 发表于 2010-4-23 12:09

对于Intel，可观的性能提升是指从P4到Core 2，从Penryn Xeon到Nehalem Xeon这种每核心80%以上的性能提升
对AMD来说，指的是从K7到K8，从K8到K10每核心20%以上的性能提升

作者: the_god_of_pig 时间: 2010-4-23 12:25
话说不知道AVX对整数性能有多大提高

作者: cheng1088 时间: 2010-4-23 12:28

The block is 100MHz instead of 133MHz and we heard confirmation that this time around Intel finally managed to get rid of the ck505 on the boards, something Intel wanted to do on socket LGA1366 already. This means all the clocks are created and managed internally, which has overclockers worried. According to rumors Sandy Bridge (115X) won't overclock as nicely as Bloomfield (1366), Gulftown (1366), Lynnfield (1156) or Clarkdale (1156) processors, which have a default base clock of 133MHz and reach 200MHz with aircooling easily and up to 240MHz if you're lucky.

block频率由133mhz降到100mhz，去掉主板上的CK505，所有频率在cpu内部产生和管理
还不让超频了

作者: 870717 时间: 2010-4-23 12:32
近日XtremeSystems论坛上的一位网友"JCornell"贴出了据称是Intel下一代32nm制程处理器产品Sandy Bridge的测试截图.简单地讲，Sandy Bridge就像是目前32nm制程Clarkdale集显处理器的改进版本，不过前者将Clarkdale封装内部集成的45nmGPU芯片原生集成到了32nm制程核心内部。以下是这位网友所张贴的Sandy Bridge产品外形图和测试截图，据截图中的CPUZ软件显示，这款Sandy Bridge处理器采用的是A2制程步进，采用双核四线程设计。

点击图片放大

插槽方面的变化:

尽管大致可视为Clarkdale处理器的改进版本，但两者之间仍然存在一些较为明显的差距。首先在处理器所采用的插槽方面便发生了很大的变化。尽管从网友提供的图片上看，Sandy Bridge处理器的插槽形式与Clarkdale使用的LGA1156并没有太大的区别，但实际上前者使用了新的LGA1155插槽，此外供电规范方面也改用VRM 13版本。这就意味着主板的供电模块必须为Sandy Bridge进行重新设计，而Sandy Bridge也无法在LGA1156的主板上使用。反过来，LGA1156插槽的处理器也将无法在LGA1155主板上使用。

时钟发生器设计以及可超性方面的变化:

有趣的是，Sandy Bridge处理器的内部总线频率从133MHz降到了100MHz。而且据传在Sandy Bridge处理器平台上，Intel会彻底放弃外置CK505时钟发生器的设计（当年推出LGA1366处理器时，Intel便曾想废除CK505部件），而把所有的时钟控制单元全部集成到处理器核心内部。这种设计对超频者而言可不是一个好消息。据外界的传言，采用LGA1155插槽的Sandy Bridge处理器在可超性方面将不如LGA1366 Bloomfield/Gulftown，以及LGA1156 Lynnfield/Clarkdale。后者默认内部总线频率为133MHz，可使用空冷轻松超到200MHz甚至240MHz。

缓存结构方面的变化:

另外，Sandy Bridge内部的缓存结构也相比过去有了一些变化和改进。比较Clarkdale，Sandy Bridge的一级指令缓存的联合路数增加了一倍，从过去的四路联合提升为八路联合；三级缓存则从4MB降为3MB，而且缓存的联合路数也从过去的16路联合降为12路联合。

指令集/性能方面的变化:

指令集方面，Sandy Bridge处理器将可支持SSE扩展指令集的升级版本AVX扩展指令集，这种指令集可以直接操作256bit向量寄存器，比过去128bit的工作效率提升了一倍。这种升级将对处理器的浮点处理性能有一定的提升作用，尽管游戏玩家对处理器的浮点性能一般都比较敏感，不过Intel这次加入AVX指令集的主要目的却是增强处理器的科学运算能力，以对抗日益猖獗的GPGPU。目前较新的Linux内核已经可以支持AVX指令集，不过Windows操作系统则还不具备这种支持能力，其中Windows7将在SP1更新补丁中追加对这项功能的支持（Windows 7 SP1补丁可能会在今年年底或圣诞期间发布）。

这里大家不要期望Sandy Bridge处理器在实际性能方面比过去有多大的提升。从Clarkdale到Sandy Bridge的性能提升幅度很可能会与Diamondville Atom处理器到Pine View Atom的性能提升那样十分有限（后者基本上只不过是把前者原来外置的芯片组整合到处理器核心中，并对处理器核心进行了制程的缩减）。因此预计Sandy Bridge推出的主要目的是进一步降低处理的制造成本和功耗。况且新加入的AVX指令集虽然看上去很美，但是目前还没有证据表明新的256bit寄存器在执行使用SSE指令的游戏时会不会有更高的执行效能。

主要针对笔记本市场:

在我们看来，对Sandy Bridge最感兴趣的应该是笔记本类移动产品用户，而预计Intel也会选择首先推出笔记本型Sandy Bridge处理器。联想到最近Intel推出的Clarkdale笔记型变种Arrandle处理器由于功耗难以控制到理想水平而表现不佳，无法完全取代CULV处理器的情况，那么这种看法的成立度便又提升了一层（据传Intel已经向其客户表示会跳过推出超低压版Arrandale处理器的计划，而将直接推出Sandy Bridge取而代之）。

由于Sandy Bridge平台中，原来外置的芯片组功能多数被整合到处理器核心内部，而且制程也升级为32nm，因此预计Sandy Bridge处理器在动态功耗控制和整体功耗控制方面应该比前代产品更为出色（Clarkdale中芯片组功能是以45nm制程的形式集成在处理器封装内部的另外一块芯片上的）。目前 Arrandale处理器主要分为三类TDP功耗不同的产品，M类（35W）,LM类（25W）以及UM类（18W），而 Sandy Bridge的功耗水平则有望达到与Atom处理器同级的水平。

作者: cheng1088 时间: 2010-4-23 12:34
五电容5200之类的以后估计没了
从频率上就可以彻底阉掉

作者: itany 时间: 2010-4-23 12:54

五电容5200之类的以后估计没了
从频率上就可以彻底阉掉
cheng1088 发表于 2010-4-23 12:34

某些人在nehalem出来之前就宣称“不能超频”云云，结果呢……
原帖上没有人说超频会削弱，反而有人说会更加Ocer Friendly

作者: cheng1088 时间: 2010-4-23 13:51
主要是发发牢骚，楼主见谅！

作者: wowman 时间: 2010-4-23 14:27
看来等推土机是明智的，逆超线程+AVX，耶

作者: Asuka 时间: 2010-4-23 14:50

看来等推土机是明智的，逆超线程+AVX，耶
wowman 发表于 2010-4-23 14:27

同学，醒醒，中午了

作者: zaknafein 时间: 2010-4-23 14:56
AVX对性能提升不明显么...

作者: knightmaster 时间: 2010-4-23 15:24

对于Intel，可观的性能提升是指从P4到Core 2，从Penryn Xeon到Nehalem Xeon这种每核心80%以上的性能提升 ...
itany 发表于 2010-4-23 12:24

你这是说P3只有K7一半的性能么....

作者: itany 时间: 2010-4-23 15:29

你这是说P3只有K7一半的性能么....
knightmaster 发表于 2010-4-23 15:24

和P3有什么关系？

作者: SOS-001 时间: 2010-4-23 15:29
没有中文的么？看不懂

作者: asdc22 时间: 2010-4-23 15:40
115X

作者: GF100 时间: 2010-4-23 16:58
速度真快！！！！！！！！

作者: gtnod32 时间: 2010-4-23 17:03
路过顶顶。。。。

作者: Coola 时间: 2010-4-23 20:01
实际出来性能如何还真难说，多了AVX，却缩了L3

另外Socket H2好像没有对应的6个或以上个核心的CPU？在Socket H2发布到Socket B2发布之间的约半年看来有点真空期的感觉

作者: zzaz 时间: 2010-4-23 20:05
请问有没有渲染测试?

作者: temp444 时间: 2010-4-23 23:39
低端CPU，会不会连AVX都不包括呢？

低端的总是少这个功能、少那个功能。全功能的CPU，至少在千元以上。

作者: kunnsr 时间: 2010-4-24 10:32
就这样用吧这东西估计出来专攻笔记本的台式继续1156 1366

作者: zidane1980 时间: 2010-4-24 10:57
这个很正常吧
出货前半年做好测试工作

作者: noytong 时间: 2010-4-24 12:39
学习一下~~~~

欢迎光临 POPPUR爱换 (https://we.poppur.com/)