Lynnfield to boost Intel quad core share

                   

Quads currently account for only 10% of total sales

Intel’s flagship Core 2 Quad 65nm, Core 2 Quad 45nm Yorkfield and Core i7 Bloolfield 45nm quad-core CPUs are currently at some nine percent of total Intel’s OEM guidance for consumer market. Continue reading

Advertisements

MIT Graphene Multiplier May Push CPUs to 1,000 GHz

05235601207870056

MIT researchers have devised a graphene signal multiplier that produces a clean signal with low power

The computer industry is involved in a never-ending quest for the highest possible performance from computer components and other electrical devices. Over the years, there have been many advances that have boosted the computational power of computer systems from increasing the number of transistors to adding multiple processing cores.

Researchers across the world are hard at work on building microprocessors and other electrical components using a material discovered in 2004 called graphene. A group of researchers at MIT announced on March 19 that new findings made could lead to much faster microprocessors in the future. The new findings could lead to cell phones and other communications equipment that can transmit data much faster than devices available today. Continue reading

24 Samsung SSDs Linked Together for 2GB/Sec

What would you do if Samsung gave you only 24-hours of hands-on time with a stack of solid-state drive (SSD) engineering samples to do some viral marketing with? For you this is surely just an academic question, but for Paul Curry of The Viral Factory in London, it was a very real challenge. And he took the challenge to the limits of where only the truly geekiest would go: he custom-built an 8-core, dual-RAID, Windows Vista system, utilizing 24 256MB MLC SSDs, for a total of 6TB of storage.

Curry’s system used an Intel Skulltrail D5400XS motherboard, with two Intel 3.2GHz QX9775 Quad-Core processors, 4GB of 800MHz FB-DIMM DDR2 SDRAM, two ATI Radeon HD 4870 X2 graphics cards, an Adaptec 5 Series RAID card, an Areca 1680ix-24 RAID card, and two Corsair HX1000W power supply units. And, of course, 24 Samsung SSDs. If this sounds like it was a difficult system to build, let’s just say he ran into a few problems along the way…
First of all, getting everything inside the case was such a tight fit, that Curry had to saw off part of the Zalman coolers to squeeze them in. He also managed to fry a motherboard and a 1,000-watt power supply. He replaced the motherboard and soldered the power leads from two 1,000 power supplies together (“connected their power_on line and gave them common ground“) to make sure the rig had enough juice–the motherboard and CPU were powered by one of the power supplies, while the system’s drives and add-in cards were powered by the other. The total output under load before the drives were added was about 1,400-watts, and around 1,500-watts after all 24 SSDs were installed. Curry also had to remove one of the two Radeon HD 4870 X2 graphics cards as he found it was drawing too much power from the PCI-e bus and preventing the Areca RAID card from initializing.

It also took Curry several iterations of setting up the RAID controllers until he had a configuration that didn’t saturate the controllers. He finally settled on a configuration that had 10 of the SSDs connected to the Areca RAID controller in a RAID 0 array, 8 of SSDs connected to the Adaptec 5 Series controller in a separate RAID 0 array, and the remaining 6 SSDs connected directly to the motherboard’s on-board SATA ports as stand-alone drives. The optical drives were disconnected during testing to maximize available throughput for the on-board SATA ports. All testing was done at stock speeds, although Curry did experiment with some overclocking “for the fun of it“–the system remained stable with the CPUs running up 3.6GHz; and he even got it up to 4GHz, but at that speed the system was “wobbly.”

The Samsung SSDs’ specifications state that they support up to a write speed of 200MB/Sec and a read speed of 220MB/Sec. So how did the system perform? Here are some of the test results highlights:

  • 2121.29MB/Sec sequential reading using IOMeter
  • 2000.195MB/Sec sequential writing using IOMeter
  • Loaded all Microsoft Office Apps in 0.5 seconds
  • Opened 53 apps in 18.09 seconds
  • Ripped a 700MB DVD transfer in 0.8 seconds

And all of this was done with “zero sector failures” on any of the drives. With data transfer rates in excess of 2GB/Sec, this puts the throughput of the rig on par with the theoretical limits of Fibre Channel. Curry was even able to get throughput above 1GB/Sec with only 9 drives. As one of the primary uses for SSDs will be in data centers, this exercise shows the potential transfer rates that might be achieved in such an environment.

by Daniel A. Begun

分析:Nvidia对处理器相关技术蓄谋已久

Xbitlabs报导:2000年起,在Nvidia尚未推出他们的统一渲染器架构之前,有关它们正在准备进入CPU市场的传言就开始流传。尽管他们直到约9年后的最近才承认确有其事,但我们却认为Nvidia已经对处理器,至少是对处理器的相关技术蓄谋已久

在10年前,GPU与CPU之间的区别可谓天差地别,而如今随着GPU和CPU技术的发展,现在GPU的可编程性被大大加强,而CPU的核心数目也在向多核,内置内存控制器方向发展,甚至可能在不久的将来集成显示核心,因此,就像消费电子产品与个人电脑产品的区别越发模糊那样,CPU与GPU也越来越具备同样的特征。

而Nvidia对整合的理念有深刻理解,并且和其它芯片厂商一样,已经为进入微处理器市场准备了多年。

首先,Nvidia早早便加入了HyperTransport联盟,当时HyperTransport还被称为光速数据传输技术(LDT),而那时AMD也刚刚在K8处理器上采用这种接口技术。这就解决了Nvidia微处理器计划中的总线接口问题。

其次,Nvidia还加入了SOI组织。该组织研发的技术可以用于制造各种芯片与电子元件,是制造微处理的基本技术。

第三,09年1月份Nvidia让William Dally替代David Kirk任首席科学家,前者是并行计算领域的知名专家,也是整合显卡处理器领域的专家。另有消息称Nvidia数年来已经雇用了多位Stexar的微处理器专家为其服务。由此看来,任命Dally先生只是Nvidia庞大计划中的一小部分而已。

既然Nvidia已经拥有了技术人才,总线接口技术和SOI制程技术,那么很显然Nvidia已经在为片上级系统设计整合显卡的微处理器。剩下的问题是他们计划何时推出他们足以挑战AMD和Intel的产品。我们认为,在AMD与Intel的相关产品完全占领整合市场之前,Nvidia自然就会推出自己的产品并加以宣传。这一幕迟早都会发生。