真牛的国产CPU商品开卖了,40纳米,16核1.6GHz

125 views
Skip to first unread message

wormwang

unread,
Apr 3, 2014, 9:13:30 PM4/3/14
to sh...@googlegroups.com
40纳米,16核1.6GHz 64bit指令集
http://www.jztec.cn/Product/gchpt/3.html

none_nobody

unread,
Apr 4, 2014, 12:07:30 AM4/4/14
to sh...@googlegroups.com
看上去不错啊,源自于DEC的Alpha 21164,4x128bits DDR3-1600

孑影

unread,
Apr 4, 2014, 12:09:50 AM4/4/14
to sh...@googlegroups.com
喵,,那个防火墙,好像很厉害的来着,,

从硬件到,软件全部,重写了一遍 ?

这么吊啊,,

#风起看云涌,叶落品人生#


--
-- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
---
您收到此邮件是因为您订阅了Google网上论坛中的“Shanghai Linux User Group”论坛。

要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+un...@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout

Chaos Eternal

unread,
Apr 4, 2014, 12:11:01 AM4/4/14
to sh...@googlegroups.com
这个无字表格是怎么回事


--
-- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
---

孑影

unread,
Apr 4, 2014, 12:11:55 AM4/4/14
to sh...@googlegroups.com
http://www.jztec.cn/Product/gchpt/6.html

这个,,我能信不 ?

#风起看云涌,叶落品人生#

none_nobody

unread,
Apr 4, 2014, 12:18:34 AM4/4/14
to sh...@googlegroups.com
不管你信不信,反正我是信了。

现在做CPU没什么难了吧,占领市场要不这个做出来难多了。

On Friday, April 4, 2014 12:11:55 PM UTC+8, 孑影 wrote:
http://www.jztec.cn/Product/gchpt/6.html

这个,,我能信不 ?


none_nobody

unread,
Apr 4, 2014, 12:19:21 AM4/4/14
to sh...@googlegroups.com
啧啧,看的真仔细。

On Friday, April 4, 2014 12:11:01 PM UTC+8, Soahc Lanrete wrote:
这个无字表格是怎么回事

Wales Wang

unread,
Apr 4, 2014, 12:20:38 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
可以。40nm工艺还是1年半以前的。
神威蓝光超级计算集群早就发表。

Wales Wang

Wales Wang

unread,
Apr 4, 2014, 12:22:19 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
只做到1.6GHz,Intel/AMD可以做到3GHz以上。
3年前研发能力还是有一点差距的。

Wales Wang

Wales Wang

unread,
Apr 4, 2014, 12:24:46 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
正在量产的是28nm,正在小规模生产的有14nm。


Wales Wang

bladesilent

unread,
Apr 4, 2014, 12:21:06 AM4/4/14
to sh...@googlegroups.com
软件系统有:
神威睿智VMM1.0
神威睿智编译器2.0,
神威睿智JDK1.7

名字起得略霸气

liyaoshi

unread,
Apr 4, 2014, 12:37:25 AM4/4/14
to sh...@googlegroups.com
求内购

Wales Wang

unread,
Apr 4, 2014, 12:36:45 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
我猜下一代全部是ARM64了。
AMD刚发布样品,8核2GHz,ARM A57架构。Broadcom和Cavium都在做,其他新公司也有几个。今年底,明年都会出来。

还不算针对移动终端的ARM64,三星,高通,MTK,Apple都是。

Intel x86一下变成非主流,就像Windows在平板电脑时代也是非主流。

Wales Wang

liyaoshi

unread,
Apr 4, 2014, 12:42:28 AM4/4/14
to sh...@googlegroups.com
arm64的浮点也不咋样唉

none_nobody

unread,
Apr 4, 2014, 12:51:07 AM4/4/14
to sh...@googlegroups.com
团购价格会低么? 28nm走起。

On Friday, April 4, 2014 12:37:25 PM UTC+8, liyaoshi wrote:
求内购


Wales Wang

unread,
Apr 4, 2014, 12:53:33 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
ARM64 有很多实现,如Apple只用ARM v8指令集,其他都是自己做。
最差水平是直接买硬IP,即已经做完版图设计了,直接生产即可。但是,硬IP当然最贵。


Wales Wang

Wales Wang

unread,
Apr 4, 2014, 12:56:02 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
不知道。我等ARM64。
做的人,那么多,竞争激烈,应该比Intel便宜。芯片成本基本是1平方毫米/1美元。
Intel太暴利了,售价有10倍毛利!!

Wales Wang

david pu

unread,
Apr 4, 2014, 1:03:19 AM4/4/14
to sh...@googlegroups.com
x86和Windows成非主流还是要很长的时间啊。
我遇到的客户一般是问:你ARM64能跑Windows吗?不是有虚拟化吗,能跑X86的软件吗?
即使他们不想用X86,但是摆脱不了上面现有的软件,特别是很多行业软件。。
() ASCII Ribbon Campaign
/\ Keep it simple!

Wales Wang

unread,
Apr 4, 2014, 1:08:53 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
服务器软件可以不用x86。

Xeon的暴利占Intel利润一半,也够他痛苦了。
平板电脑和智能手机也没Intel的事。

Intel x86 Windows爱用的就去用吧。

其他都是ARM64加Linux了。

Wales Wang
> 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

liyaoshi

unread,
Apr 4, 2014, 1:51:30 AM4/4/14
to sh...@googlegroups.com
我们的客户是希望一个cortex A57 soc,跑一个android一个qnx

你听了有没有蛋疼的感脚?

您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

liyaoshi

unread,
Apr 4, 2014, 1:53:16 AM4/4/14
to sh...@googlegroups.com
ARM的xen 就是greenhills的那帮人在维护

我真的看不出来有谁会这么跑,

手持,桌面,移动 这种应用,关虚拟技术毛事

Yiling Cao

unread,
Apr 4, 2014, 2:07:39 AM4/4/14
to sh...@googlegroups.com
鬼信

Wales Wang

unread,
Apr 4, 2014, 2:11:16 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
不会呀,RThread提出OS的bigLITTLE,平时运营RThread省电,一旦收到需要全Linux的应用需求,RThread启动VMM里的Linux满足全部应用功能。

Wales Wang

Wales Wang

unread,
Apr 4, 2014, 2:17:58 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
哈哈。要么赌赌钱。
人家是做得太大了,Tsmc等地方不给生产。
老美各处都派人的,发现太大的,不给排生产。

Wales Wang

Wales Wang

unread,
Apr 4, 2014, 2:28:02 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
还有做宇航级芯片的,那个高能辐射就是个氦原子核以近光速打上来。
芯片要做特殊设计,抗的住。中国航天有点牛,是芯片业有点实力的。

Wales Wang

none_nobody

unread,
Apr 4, 2014, 2:35:57 AM4/4/14
to sh...@googlegroups.com
M$ 平台 VS 编译出来,比 GCC 编译出来慢基本一倍。

ICC  在Intel 更快一点。




On Friday, April 4, 2014 1:08:53 PM UTC+8, wormwang wrote:
服务器软件可以不用x86。

姚 飞

unread,
Apr 4, 2014, 2:40:43 AM4/4/14
to sh...@googlegroups.com
偷来的锣不能敲,这帮狗东西连个文档都不敢公开嘛。


在 2014年4月4日,上午9:13,wormwang <worm...@yahoo.com> 写道:

> 40纳米,16核1.6GHz 64bit指令集
> http://www.jztec.cn/Product/gchpt/3.html
>
> --
> -- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
> --- 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

david pu

unread,
Apr 4, 2014, 2:41:50 AM4/4/14
to sh...@googlegroups.com
车载上好像有很多这种需求哦。

Perchouli

unread,
Apr 4, 2014, 2:44:41 AM4/4/14
to sh...@googlegroups.com
落后了,对国产CPU的认识只停留龙芯系列。
这强调安全性… 有点疑问的是,操作系统/普通硬件留后门可以理解,CPU写入后门而且实现监控的可能性多高?是什么原理实现的?…


--
-- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
---
您收到此邮件是因为您订阅了Google网上论坛中的“Shanghai Linux User Group”论坛。
要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+un...@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout



--

Very Respectfully.Your obedient servant
Perchouli

Wales Wang

unread,
Apr 4, 2014, 2:45:08 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
呵呵
等等ARM64吧,Alpha,Sparc给练练手的。

Wales Wang

Wales Wang

unread,
Apr 4, 2014, 2:50:16 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
对L3做个检查电路,发现特定数据,关闭节能控制电路,你CPU要么过热保护停了,甚至可能热坏了。
这还没证据,他也可能是正常热坏的。

Wales Wang

Wales Wang

unread,
Apr 4, 2014, 2:59:21 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
搞芯片电路还麻烦。
Snowden泄露了,美国人做了硬盘和网卡Firmware里的木马。

在Firmware做手脚就容易多了。而且各种芯片结构不同,到底有没后门,用户无法查。而且,若出厂后,NSA/CIA派人去换Firmware,原研厂商还撇清关系。

Wales Wang

Qian Hong

unread,
Apr 4, 2014, 3:12:23 AM4/4/14
to sh...@googlegroups.com
2014-04-04 14:44 GMT+08:00 Perchouli <jp.ch...@gmail.com>:
> 落后了,对国产CPU的认识只停留龙芯系列。
> 这强调安全性… 有点疑问的是,操作系统/普通硬件留后门可以理解,CPU写入后门而且实现监控的可能性多高?是什么原理实现的?…


我是外行,不是很懂,道听途说的:理论上可以在CPU中设计一些后门,导致用户(自以为)通过硬件随机数生成器生成的随机数其实是伪随机数,并且只有厂商知道其中的秘密,必要的时候可以根据这个漏洞破解该机器上计算出来的哈希/密码之类的东西。
当时我们在测一个TCP SYN
cookies的潜在安全漏洞,组里没人能完全搞明白,碰巧找到一位比利时来的密码学博士找他请教,这大哥读了半天代码讨论了半天,居然告诉我们这个bug理论上估计就不存在测试的方法,然后就开始扯密码学扯后门,最后我们光顾着扯淡忘记干活了。。

Google搜了一下,似乎有类似的介绍,没耐心读完,求总结:
http://arstechnica.com/security/2013/12/we-cannot-trust-intel-and-vias-chip-based-crypto-freebsd-developers-say/

--
Regards,
Qian Hong

-
http://www.winehq.org

Wales Wang

unread,
Apr 4, 2014, 3:17:24 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
举个例子,密码算法强度是128位,但是密码加密体系里一般用随机数做实际密钥。
若随机数强度只有40位,那加密算法的强度128位是虚的,因为密码总共才40位。

Wales Wang
> --
> -- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
> ---
> 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

liyaoshi

unread,
Apr 4, 2014, 3:56:14 AM4/4/14
to sh...@googlegroups.com
台积电不是台湾人管的么?

28nm的产线,台积电不是很久以前就帮苹果造那啥了么

台积电不是号称14nm的厂子多在造了吗


R

Wales Wang

unread,
Apr 4, 2014, 3:58:28 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
中国大陆是被特殊管住的。
Tsmc的14nm有点慢了,Intel,IBM都有了。


Wales Wang

liyaoshi

unread,
Apr 4, 2014, 3:59:20 AM4/4/14
to sh...@googlegroups.com
号称台积电的14nm技术可以秒掉 intel???

Wales Wang

unread,
Apr 4, 2014, 4:03:55 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
只能呵呵了,推迟3次了。

Wales Wang

liyaoshi

unread,
Apr 4, 2014, 4:05:57 AM4/4/14
to sh...@googlegroups.com
话说,我听下来的意思,EMC要出ARM版的机器了?

Wales Wang

unread,
Apr 4, 2014, 4:07:23 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
EMC的硬件估计还是Intel,毕竟Intel前CTO在EMC做高管。

Wales Wang

liyaoshi

unread,
Apr 4, 2014, 4:09:13 AM4/4/14
to sh...@googlegroups.com
八卦一下,

话说,这个君正,跟北京君正有啥关系么?

Wales Wang

unread,
Apr 4, 2014, 4:12:50 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
应该没关系

Wales Wang

liyaoshi

unread,
Apr 4, 2014, 4:22:54 AM4/4/14
to sh...@googlegroups.com
再回到这个神威片子上面, 你们谁用过?

看封装,100瓦TDP 扛得住么?

none_nobody

unread,
Apr 4, 2014, 6:06:41 AM4/4/14
to sh...@googlegroups.com
那年看过 2164 的alpha,  表示很期待在机器的架构中增加一个alpha.

Wales Wang

unread,
Apr 4, 2014, 6:15:46 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
过去10多年了,Alpha那时的特性在当今的高性能CPU里都有:如多ALU,内含MMU,多级流水线,乱序执行,执行多线程,SIMD浮点指令。

ARM,x86,Sparc,Power就看谁做的性价比好了。

Wales Wang

Yu Changyuan

unread,
Apr 4, 2014, 6:28:47 AM4/4/14
to shlug

维基上说alpha最后买给intel了。那么这个cpu不涉及专利什么的吗(就像龙芯后来也买了mips的授权)?

wormwang

unread,
Apr 4, 2014, 4:23:08 AM4/4/14
to sh...@googlegroups.com
我记得Alpha卖了多处。Intel买了一些专利。首席架构师去了AMD,就是 AMD64的架构师,AMD64做得很成功,他也就老了退休了。也算在退休前报了Intel的仇。

中国当时买了一些21164的东西。都过去超过15年了。
芯片内微体系架构近10年没啥新发明了。多ALU,内含MMU,多级流水线,乱序执行,执行多线程,SIMD浮点指令,64bit,每一个都 是在2000年前发明的。

Intel近年想搞的,芯片内光纤通讯失败了,stream浮点ALU生产要求高成品率低用途少也没做。芯片内放更大的L3,L4 cache也是成本问题放弃。
这2003-2013年的十年,就是每2年升级一下工艺,放更多地处理器核,SIMD指令增加多几个。

Intel/AMD/IBM/SPARC 翻炒几轮旧料,自然容易被新的芯片厂商,甚至如MTK超越。MTK某天做个超级便宜的ARM64 服务器版本,Intel/AMD全要吐血。
MTK 和芯片代工厂的关系甚至比AMD与TSMC的关系密切。

ARM 64 A57微架构已经包括了:多ALU,内含MMU,多级流水线,乱序执行,SIMD浮点指令,64bit,每IPS的低功耗。所以,移动版本ARM64与服 务器版本大多数方面是类似的,服务器CPU就强一点在执行多线程,流水线层数多一点,SIMD浮点指令多一点,只是最大发热功率一个不能大于 2W,另一个可以做到100W。




On 2014/4/4 18:28, Yu Changyuan wrote:

维基上说alpha最后买给intel了。那么这个cpu不涉及专利什么的吗(就像龙芯后来也买了mips的授 权)?

none_nobody

unread,
Apr 4, 2014, 7:21:28 AM4/4/14
to sh...@googlegroups.com
内容很多。

就想说一点,arm 会不会因为太关注于移动平台,而懒得向服务用发展呢?

这好像有种固步自封,天天吃肉,不觉得有什么危机存在的感觉。

wormwang

unread,
Apr 4, 2014, 5:19:05 AM4/4/14
to sh...@googlegroups.com
ARM公司的ARM v8即ARM64指令集是通用的,ARM自己直接做的A57/A53设计实现是面向移动终端的。

但是ARM v8其他实现,如AMD/Cavium/Boradcom 就是做服务器的,有些不是完全的A57设计实现,有自己的ARM v8增强实现。
谣传Apple Mac电脑要改ARM64,就是因为iPad/iPhone的A7 CPU的ARM 64实现全部是Apple自己做的,可以全面掌控。而且,A7 CPU的内部微架构和Intel i7几乎一样,除了功耗限制不同。

Intel估计只能给Apple高折扣优惠,求着Apple别换ARM。Apple的利润主要来自iPhone/iPad,他会给iPad配 键盘(已经发布了),做11-12寸的新型号。
我估计Mac电脑2-3年内不会换CPU,外围非开源应用软件又要再重编译优化一次,估计抱怨还是有的。

像Apple这样软件硬件通吃公司,21世纪太少了。




On 2014/4/4 19:21, none_nobody wrote:
内容很多。

就想说一点,arm 会不会因为太关注于移动平台,而懒得向服务用发展呢?

这好像有种固步自封,天天吃肉,不觉得有什么危机存在的感觉。


On Friday, April 4, 2014 4:23:08 PM UTC+8, wormwang wrote:

ARM 64 A57微架构已经包括了:多ALU,内含MMU,多级流水线,乱序执行,SIMD浮点指 令,64bit,每IPS的低功耗。所以,移动版本ARM64与服 务器版本大多数方面是类似的,服务 器CPU就强一点在执行多线程,流水线层数多一点,SIMD浮点指令多一点,只是最大发热功率一个不能大于 2W,另一个可以做到100W。


馒头簸箕

unread,
Apr 4, 2014, 8:21:13 AM4/4/14
to sh...@googlegroups.com
俺去过56所参观,当面问那个老太太,丫坚决否认跟Alpha的关系,嘿嘿。它们有一批人在上海做研发。

他们的超级计算机是水冷的,水冷系统设计还是不错的。

发自我的 iPad

Wales Wang

unread,
Apr 4, 2014, 8:41:18 AM4/4/14
to sh...@googlegroups.com, sh...@googlegroups.com
呵呵。证实,听说CPU研发部在上海张江。

Wales Wang

Ben Luo

unread,
Apr 4, 2014, 8:53:03 AM4/4/14
to shlug

国内出了好几个出名的假货。这个是真的吗?真不敢信了

Wales Wang

unread,
Apr 4, 2014, 9:13:11 AM4/4/14
to sh...@googlegroups.com, shlug
虽然不能说自古以来,但是毕竟从199x年以来一步一步就有产品的。

A1750 CPU,C30 DSP,C64/67 DSP, FT1000/1500,SWxxxx CPU…
这些都是公开的,在56所干的人也全部是NUDT毕业生。

Wales Wang

SniperJackson

unread,
Apr 5, 2014, 10:18:58 AM4/5/14
to sh...@googlegroups.com
On Friday, April 4, 2014 12:22:19 PM UTC+8, wormwang wrote:
只做到1.6GHz,Intel/AMD可以做到3GHz以上。
。。。。。。大哥,不同指令集的CPU不能直接比较主频的好不。。。。。。

 
3年前研发能力还是有一点差距的。

Wales Wang

在 2014-4-4,下午12:18,none_nobody <lyx...@gmail.com> 写道:

不管你信不信,反正我是信了。

现在做CPU没什么难了吧,占领市场要不这个做出来难多了。

On Friday, April 4, 2014 12:11:55 PM UTC+8, 孑影 wrote:
http://www.jztec.cn/Product/gchpt/6.html

这个,,我能信不 ?


--
-- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
---
您收到此邮件是因为您订阅了Google网上论坛中的“Shanghai Linux User Group”论坛。
要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+unsubscribe@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout

Wales Wang

unread,
Apr 5, 2014, 11:52:56 AM4/5/14
to sh...@googlegroups.com, sh...@googlegroups.com
呵呵
那是10年前这么说法。
现在谁都可以实现各个微体系架构,5级流水线,还是50级都可以做。

难做的是一定高主频时供电和时钟电路到位,同时还要适当时刻关闭它们要省电。没这3点,现代CPU/DSP根本不能工作,只能当小电炉用。

所以最近几年,简单化看,看数字集成电路设计水平,看它最好主频。针对性看,看GFlops/W。SWxxxx的GFlops/W相当好。

Wales Wang
要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+un...@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout

liyaoshi

unread,
Apr 5, 2014, 9:09:45 PM4/5/14
to sh...@googlegroups.com

论GFlops/w ,c66x 是我知道的最厉害的了吧

Wales Wang

unread,
Apr 5, 2014, 9:19:21 PM4/5/14
to sh...@googlegroups.com, sh...@googlegroups.com
CPU比CPU,DSP比DSP,还有GPU呢。

像AMD APU那样,更难对比了。

Wales Wang

liyaoshi

unread,
Apr 6, 2014, 12:47:20 AM4/6/14
to sh...@googlegroups.com
我想说的是,16核,含cpu +主板 ,1500以内,我或许考虑买一个,贵了,不买

毕竟隔壁人家 e3 1230 3.2G 8thread 才1k出头

none_nobody

unread,
Apr 6, 2014, 1:41:20 AM4/6/14
to sh...@googlegroups.com
其实主频、核数都不是最关键的因素。以某之陋见,CPU cache  大小才起决定的作用,只要编译软件能够充分利用cache就好了。片上 cache 这事与半导体生产工艺相关,没有太难的逻辑电路设计,没有指望大的片上cache。21164 1995年有64M L3-cache , 期望有片上超过256M L1, 那一定是出现幻觉了。

只能做出CPU设计,做不出设计CPU的设计软件,以及咩有支撑的生产工艺都是虚幻。

我是跟风想玩耍  alpha 的,只能是便宜越便宜越好。

Wales Wang

unread,
Apr 6, 2014, 1:50:17 AM4/6/14
to sh...@googlegroups.com, sh...@googlegroups.com
是的,当今芯片技术瓶颈叫:内存墙,访问内存比CPU速度慢100倍。
所以power7配40MB L3 cache,面积达300平方毫米。可是,这么大面积,生产成本高。真是矛盾…
Wales Wang

Ben Luo

unread,
Apr 6, 2014, 2:16:38 AM4/6/14
to shlug

目前服务器上真正量产使用的只剩下x64了吧。power有人用吗

Wales Wang

unread,
Apr 6, 2014, 2:38:47 AM4/6/14
to sh...@googlegroups.com, shlug
Sparc,power都还有。Itanium还活着。

Wales Wang

liyaoshi

unread,
Apr 6, 2014, 3:43:44 AM4/6/14
to sh...@googlegroups.com
你说的L3 cache ,不是片内的,1995年的技术,只可能是外接的,tsop封装512K 最多了,当年多是打在主板上的

到了Pentium-II开始,才封装在slot-1的板子上,64M有点玄,

cache本质就是SRAM 快慢关键看主频和带宽

现在cortex-M3 这种方案,很多继续外扩的sram,但是现在外接的片子,你看不到1G主频的

当年AMD的athlon就是用cpu一半的主频跑L2cache,所以能第一个达到1G主频 250nm还是350nm的技术




Wales Wang

unread,
Apr 6, 2014, 3:56:05 AM4/6/14
to sh...@googlegroups.com, sh...@googlegroups.com
L3 cache曾经有片外的。特别是power做过。

这两年都谈片内的。Intel Xeon都最大有25MB了,正在设计的有128MB的。
没办法内存太慢,L2是每核独享的,只能做大L3。

我预计大家1,2年里能看到300平方毫米以上含128MB cache,双万兆以太网的大芯片。

因为,设计做得好,容错做得好,大面积芯片还是商业可行的。




Wales Wang

none_nobody

unread,
Apr 6, 2014, 7:12:31 AM4/6/14
to sh...@googlegroups.com
wiki 看到一个 L2 3X3M + L3 16M , 然后还有 L3 24M  , 还有 L2 15X256kB,

难道 IA 们不好意思说 L1 了么?

你说的片上2年内 128M,就算这样也还是很吓人。I 都出来说摩尔定律玩不下去了。


On Sunday, April 6, 2014 3:56:05 PM UTC+8, wormwang wrote:
L3 cache曾经有片外的。特别是power做过。

这两年都谈片内的。Intel Xeon都最大有25MB了,正在设计的有128MB的。
没办法内存太慢,L2是每核独享的,只能做大L3。

我预计大家1,2年里能看到300平方毫米以上含128MB cache,双万兆以太网的大芯片。

因为,设计做得好,容错做得好,大面积芯片还是商业可行的。




Wales Wang

在 2014-4-6,下午3:43,liyaoshi <liya...@gmail.com> 写道:

你说的L3 cache ,不是片内的,1995年的技术,只可能是外接的,tsop封装512K 最多了,当年多是打在主板上的

到了Pentium-II开始,才封装在slot-1的板子上,64M有点玄,

cache本质就是SRAM 快慢关键看主频和带宽

现在cortex-M3 这种方案,很多继续外扩的sram,但是现在外接的片子,你看不到1G主频的

当年AMD的athlon就是用cpu一半的主频跑L2cache,所以能第一个达到1G主频 250nm还是350nm的技术


在 2014年4月6日 下午1:41,none_nobody <lyx...@gmail.com>写道:
其实主频、核数都不是最关键的因素。以某之陋见,CPU cache  大小才起决定的作用,只要编译软件能够充分利用cache就好了。片上 cache 这事与半导体生产工艺相关,没有太难的逻辑电路设计,没有指望大的片上cache。21164 1995年有64M L3-cache , 期望有片上超过256M L1, 那一定是出现幻觉了。

只能做出CPU设计,做不出设计CPU的设计软件,以及咩有支撑的生产工艺都是虚幻。

我是跟风想玩耍  alpha 的,只能是便宜越便宜越好。



On Sunday, April 6, 2014 12:47:20 PM UTC+8, liyaoshi wrote:
我想说的是,16核,含cpu +主板 ,1500以内,我或许考虑买一个,贵了,不买


--
-- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
---
您收到此邮件是因为您订阅了Google网上论坛中的“Shanghai Linux User Group”论坛。
要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+unsubscribe@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout

--
-- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
---
您收到此邮件是因为您订阅了Google网上论坛中的“Shanghai Linux User Group”论坛。
要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+unsubscribe@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout

Wales Wang

unread,
Apr 6, 2014, 7:35:57 AM4/6/14
to sh...@googlegroups.com, sh...@googlegroups.com
L1太小,影响不大。
摩尔定律瓶颈在10nm,Intel吹牛说瓶颈是7nm,CMOS栅极才5个原子厚,即使7nm工艺也就在4年以后。
Intel搞一些软件投资,也是怕芯片硅工艺撞墙了。
10nm之后,换GaN或者InP材料时,各家都是新人,说不定冒出来弯道超车的。

Wales Wang
要退订此论坛并停止接收此论坛的电子邮件,请发送电子邮件到shlug+un...@googlegroups.com
要查看更多选项,请访问https://groups.google.com/d/optout

liyaoshi

unread,
Apr 6, 2014, 10:59:22 AM4/6/14
to sh...@googlegroups.com
300平豪宅阿

这玩意玩出来,怎么切割和封装啊,散热是个大问题

散热片压得紧一点,很容易就碎掉了

none_nobody

unread,
Apr 6, 2014, 11:05:55 AM4/6/14
to sh...@googlegroups.com
没事儿,人类都已经分化成两个物种了,接硅光人和裸人了。

芯片这事或者就有了堆叠毛细管散热基座的混成结构,最终演化成一个类似脑组织的硅堆。

Wales Wang

unread,
Apr 6, 2014, 11:07:48 AM4/6/14
to sh...@googlegroups.com, sh...@googlegroups.com
power7近567平了。

Wales Wang

Wales Wang

unread,
Apr 6, 2014, 11:17:30 AM4/6/14
to sh...@googlegroups.com, sh...@googlegroups.com
Intel Xeon E7 88xx V2也有541平,4.3billion晶体管,37.5MB L3
卖6681USD 
155W TDP,已经是小电炉了。
工作电压0.7到1.1V,可以了解那么小地方要供应150A大电流

Wales Wang

在 2014-4-6,下午10:59,liyaoshi <liya...@gmail.com> 写道:

liyaoshi

unread,
Apr 6, 2014, 10:03:32 PM4/6/14
to sh...@googlegroups.com
再大点,以后可以当瓷砖了,

A--你家厕所什么牌子瓷砖?
B--intel的,超级防漏,你家呢?

none_nobody

unread,
Apr 6, 2014, 10:21:06 PM4/6/14
to sh...@googlegroups.com
要是这种瓷砖都用全新的,

土豪,我们做朋友吧。

On Monday, April 7, 2014 10:03:32 AM UTC+8, liyaoshi wrote:
再大点,以后可以当瓷砖了,


Chaos Eternal

unread,
Apr 7, 2014, 8:21:47 AM4/7/14
to sh...@googlegroups.com
kolmogrov 复杂性原理:证明一个串是随机串这个问题是不可计算的。

2014-04-04 15:12 GMT+08:00 Qian Hong <frac...@gmail.com>:
> 2014-04-04 14:44 GMT+08:00 Perchouli <jp.ch...@gmail.com>:
>> 落后了,对国产CPU的认识只停留龙芯系列。
>> 这强调安全性… 有点疑问的是,操作系统/普通硬件留后门可以理解,CPU写入后门而且实现监控的可能性多高?是什么原理实现的?…
>
>
> 我是外行,不是很懂,道听途说的:理论上可以在CPU中设计一些后门,导致用户(自以为)通过硬件随机数生成器生成的随机数其实是伪随机数,并且只有厂商知道其中的秘密,必要的时候可以根据这个漏洞破解该机器上计算出来的哈希/密码之类的东西。
> 当时我们在测一个TCP SYN
> cookies的潜在安全漏洞,组里没人能完全搞明白,碰巧找到一位比利时来的密码学博士找他请教,这大哥读了半天代码讨论了半天,居然告诉我们这个bug理论上估计就不存在测试的方法,然后就开始扯密码学扯后门,最后我们光顾着扯淡忘记干活了。。
>
> Google搜了一下,似乎有类似的介绍,没耐心读完,求总结:
> http://arstechnica.com/security/2013/12/we-cannot-trust-intel-and-vias-chip-based-crypto-freebsd-developers-say/
>
> --
> Regards,
> Qian Hong
>
> -
> http://www.winehq.org
>
> --
> -- You received this message because you are subscribed to the Google Groups Shanghai Linux User Group group. To post to this group, send email to sh...@googlegroups.com. To unsubscribe from this group, send email to shlug+un...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/shlug?hl=zh-CN
> ---
> 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

Chaos Eternal

unread,
Apr 7, 2014, 8:29:52 AM4/7/14
to sh...@googlegroups.com
没有明显证据表明这两个物种之间发生了充分的生殖隔离。

Chaos Eternal

unread,
Apr 7, 2014, 8:31:03 AM4/7/14
to sh...@googlegroups.com
你是说用烧瓷砖的工艺来做芯片么?氧化铝参杂工艺能解决?

Wales Wang

unread,
Apr 7, 2014, 8:34:25 AM4/7/14
to sh...@googlegroups.com, sh...@googlegroups.com
瓷砖可以做多种,这是单晶硅那种。

Wales Wang
> 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

none_nobody

unread,
Apr 7, 2014, 11:50:55 AM4/7/14
to sh...@googlegroups.com
您这一下子直接搬出一个K研究所,给了一个答案,这不科学。需要过程。

看到过一个用色子做的翻滚式机械的随机码生成器,这不需要证明吧。

On Monday, April 7, 2014 8:21:47 PM UTC+8, Soahc Lanrete wrote:
kolmogrov 复杂性原理:证明一个串是随机串这个问题是不可计算的。


none_nobody

unread,
Apr 7, 2014, 11:53:47 AM4/7/14
to sh...@googlegroups.com
你当那些微信约炮吊附近的人的都是在玩耍?

还有网络红娘成交的和婚介所配对的完全是两种人类。

数据藩篱还是明显的。

On Monday, April 7, 2014 8:29:52 PM UTC+8, Soahc Lanrete wrote:
没有明显证据表明这两个物种之间发生了充分的生殖隔离。

Qian Hong

unread,
Apr 7, 2014, 11:53:57 AM4/7/14
to sh...@googlegroups.com
2014-04-07 20:21 GMT+08:00 Chaos Eternal <chaose...@shlug.org>:
> kolmogrov 复杂性原理:证明一个串是随机串这个问题是不可计算的。
谢谢Chaos,学习了,好一个大坑,看不完的资料。

none_nobody

unread,
Apr 7, 2014, 12:20:26 PM4/7/14
to sh...@googlegroups.com, frac...@gmail.com
kolmogrov 到底是个研究所还是个个人,需要先证明一下。结果就真有人肉的证明过程。

kolmogrov 过于强大,跨多种领域,以至于他的资料多的看不过来。还是直接给 arxiv 号吧。

Qian Hong

unread,
Apr 7, 2014, 12:24:09 PM4/7/14
to none_nobody, sh...@googlegroups.com
2014-04-08 0:20 GMT+08:00 none_nobody <lyx...@gmail.com>:
> kolmogrov 过于强大,跨多种领域,以至于他的资料多的看不过来。还是直接给 arxiv 号吧。


我认为Chaos指的是这个题 http://en.wikipedia.org/wiki/Kolmogorov_complexity

Chaos Eternal

unread,
Apr 7, 2014, 2:59:42 PM4/7/14
to sh...@googlegroups.com
生殖隔离的意思是,这两个物种之间无法通过交配产生能交配生殖的下一代。
很明显,你说的这两种无尾裸猿之间还没有发生这种生殖隔离现象,最多只能算是两个亚种。

Chaos Eternal

unread,
Apr 7, 2014, 3:03:52 PM4/7/14
to sh...@googlegroups.com
你这个色子只能算是一个oracle而已。图灵说,就算加入了oracle的图灵机还是会有停机问题。

况且,你也不知道NSA那边是不是正好有一个跟你这个色子处于纠缠态的另一个色子呢。
> --
> -- You received this message because you are subscribed to the Google Groups
> Shanghai Linux User Group group. To post to this group, send email to
> sh...@googlegroups.com. To unsubscribe from this group, send email to
> shlug+un...@googlegroups.com. For more options, visit this group at
> https://groups.google.com/d/forum/shlug?hl=zh-CN
> ---
> 您收到此邮件是因为您订阅了Google网上论坛中的“Shanghai Linux User Group”论坛。

Chaos Eternal

unread,
Apr 7, 2014, 3:21:07 PM4/7/14
to sh...@googlegroups.com
一般来说,单质不叫陶瓷。。。

单晶更不能算陶瓷了。。。比如铝土矿烧结了可以叫陶瓷,但是在氢氧焰中生长出来的单晶一般叫刚玉

Wales Wang

unread,
Apr 7, 2014, 7:06:45 PM4/7/14
to sh...@googlegroups.com, sh...@googlegroups.com
那改叫硅晶砖吧

Wales Wang

liyaoshi

unread,
Apr 7, 2014, 9:16:10 PM4/7/14
to sh...@googlegroups.com
其实我一直想,要是能把整台机器,全多封装在一个砖头大小的块里面,只要插到墙上,统一规格,那该有多好

其实cpu和nand一样,属于易耗品,大不了,到时候就换


none_nobody

unread,
Apr 7, 2014, 10:08:06 PM4/7/14
to sh...@googlegroups.com
在一个富含氢氧化合物的星球上,是用水泥块做框架,然后木板做小格,然后封装一些具有自移动能力的碳水化合的模块进行的。这些碳水化合物模块大多具有趋光性。绝大多数也都具有可替代性。而且只消耗另一种框架里面的数值信息。

但框架还是不能统一,没法做到插拔切换,具体还要研究一下框架与星球的连接紧密程度及拔掉框架后对星球的损伤情况。

liyaoshi

unread,
Apr 7, 2014, 11:21:59 PM4/7/14
to sh...@googlegroups.com
你是猴子请来的喵星人么 ?



Yu Changyuan

unread,
Apr 7, 2014, 11:33:52 PM4/7/14
to shlug

这是在说nuc吗?

liyaoshi

unread,
Apr 8, 2014, 12:36:52 AM4/8/14
to sh...@googlegroups.com
我觉得ncu这种把网线和电源 直接做在外面的壳子上(做的跟金手指一样),就可以工业标准化了

馒头簸箕

unread,
Apr 8, 2014, 1:32:24 AM4/8/14
to sh...@googlegroups.com
类似CPU封装那种封装貌似是日本京瓷的专业。无锡56所说它们掌握了此技术,可以生产。

发自我的 iPad
> 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

馒头簸箕

unread,
Apr 8, 2014, 1:33:45 AM4/8/14
to sh...@googlegroups.com
品种,类似各种哈巴狗

发自我的 iPad
> 您收到此邮件是因为您订阅了 Google 网上论坛的“Shanghai Linux User Group”论坛。

Wales Wang

unread,
Apr 8, 2014, 1:35:55 AM4/8/14
to sh...@googlegroups.com, sh...@googlegroups.com
封装已经不是问题了。


Wales Wang

liyaoshi

unread,
Apr 8, 2014, 1:36:53 AM4/8/14
to sh...@googlegroups.com
巴掌大小的DIE怎么封装啊

Wales Wang

unread,
Apr 8, 2014, 1:43:57 AM4/8/14
to sh...@googlegroups.com, sh...@googlegroups.com
呵呵。就我所知,巴掌大的芯片有10年历史了。
所以,技术不是问题,就是贵。

Wales Wang

liyaoshi

unread,
Apr 8, 2014, 1:47:47 AM4/8/14
to sh...@googlegroups.com
巴掌大的DIE ?
求出处

Wales Wang

unread,
Apr 8, 2014, 1:50:23 AM4/8/14
to sh...@googlegroups.com, sh...@googlegroups.com
power系列CPU一直那么大。我还知道其他类似大小的。

Wales Wang

Wales Wang

unread,
Apr 8, 2014, 1:52:02 AM4/8/14
to sh...@googlegroups.com, sh...@googlegroups.com
更精确讲,过去到现在是小孩巴掌大,未来会出现成人巴掌大的。

Wales Wang

liyaoshi

unread,
Apr 8, 2014, 1:52:25 AM4/8/14
to sh...@googlegroups.com
我看到过的多是几片合在一起,打在基板上的才这么大

liyaoshi

unread,
Apr 8, 2014, 1:58:39 AM4/8/14
to sh...@googlegroups.com
你说的是这种么?


这种也是把小的硅芯片贴在基板上面的

It is loading more messages.
0 new messages