bit-tech.net

Leaked slide teases Intel Skylake features

Leaked slide teases Intel Skylake features

Intel's Skylake platform, the successor to the process-shrunk Haswell parts known as Broadwell, will include support for DDR4, PCI Express 4.0 and AVX 3.2 according to a leaked slide.

A leaked slide purporting to show Intel's roadmap for its Xeon family of server and high-performance computing processors, has offered a glimpse of the company's plans for the Skylake architecture.

Skylake is the name given to Broadwell's successor, itself the next processor generation after Intel's latest Haswell chips. Where Broadwell will merely be a simple process node shrink from 22nm to 14nm, retaining the same fundamental design as Haswell, Skylake is to be an evolution in architecture - and promises to introduce significant new features.

According to a slide leaked by a Polish university, removed by the site but not before German site PC Games Hardware was able to snag a copy, Intel's Skylake chips will include AVX 3.2, DDR4 and PCI Express 4.0 support among their improved feature set. The support for DDR4 will be of interest to performance enthusiasts, offering a new point-to-point design that allows for boosted clock frequencies and data transfer rates while running at a lower voltage than equivalent DDR3 memory.

While DDR4 is rumoured to be arriving on the desktop before Skylake, as part of Intel's Haswell-E enthusiast-grade processors, PCI Express 4.0 will be a first for the platform. Although a preliminary specification was released by industry consortium PCI-SIG back in 2011, the final specification isn't expected until 2014 or even 2015. Early drafts claimed that the standard will include double the transfer rate - 16 gigatransfers a second (GT/s) compared to 8GT/s - of PCIe 3.0, while also introducing new optimisations designed to reduce power draw during both idle and active states.

'Experts in the PCIe Electrical Workgroup carefully analysed a number of target bit rates for the next generation of PCIe architecture, taking into consideration several key factors including our ability to continue using low-cost materials,' claimed PCI-SIG chair Al Yanes at the time. 'We have concluded that 16GT/s is a feasible technical solution that satisfies our member companies’ requirements.'

The final feature announced on the slide is the introduction of AVX 3.2, the latest generation of Intel's Advanced Vector Extensions. Designed to build on the company's work on single instruction multiple data (SIMD) and streaming SIMD extensions (SSE), which allow the chip to perform a given instruction on multiple data simultaneously in order to improve performance, AVX 3.2 will replace the AVX 2 feature of Haswell and Broadwell. While Intel has yet to release details of AVX 3.2, it will likely result in similar improvements to AVX 2 - formerly known as the Haswell New Instructions. As with previous versions of AVX, however, it's not free performance: to take advantage of the new features, programmers will have to code and compile with AVX in mind.

The same slide also reveals a shift for Intel's Xeon Phi co-processor platform. The current Knights Corner PCIe add-in board, which contains 50 Pentium-style cores based on a 22nm process, will be succeeded by a 14nm version dubbed Knights Landing. The GDDR5 memory is to be replaced by lower-latency DDR4, but the big surprise is that the chip will also be made available in a socket format for the first time - allowing manufacturers to install Xeon Phi co-processors in high-performance computing systems alongside, rather than in the place of, graphics processing unit (GPU) based accelerator boards.

Intel, for its part, has refused to comment on the veracity of the slide or of any details surrounding Skylake. The platform itself is not expected to launch before 2016 at the earliest.

8 Comments

Discuss in the forums Reply
Corky42 5th July 2013, 10:54 Quote
Rumor has it Skylake will come with SATA Express to :)
SchizoFrog 5th July 2013, 11:36 Quote
As much as the architecture changes will be on Skylake (is it just me that wants to call it Skyfall?) I think it will take a year or so for DDR4 to mature and for GPUs to make use of the extra bandwidth so I think the real benefits won't actually hit the end user until Skylake's successor or later.
LEESY1 5th July 2013, 14:24 Quote
Hey guys. I thought Broadwell had been cancelled?. Last month there were reports Intel were having problems with the shrink to 14nm?, hence the delay on the shrink process.. Therefore they decided only to refresh Haswell next year (unless they are now calling this refresh Broadwell). Many tech sites mentioned this. There's also discussions on this on places such as Anandtech. Therefore Skylake was going to be the first 14nm Cpu from Intel.

http://www.dailytech.com/Report+Intel+Delays+14+nm+Broadwell+Schedules+Haswell+Refresh+for+2014/article31770.htm

Can anyone clear this up?

Lee
Corky42 5th July 2013, 15:31 Quote
Quote:
Originally Posted by LEESY1
Can anyone clear this up?

I think the only people who can clear up, what are after all nothing but rumors would be Intel and as they never comment on rumors we will just have to wait and see.
Stanley Tweedle 5th July 2013, 16:49 Quote
I need to find a way to max out my PCIe 3 so I can justify 4.

Oh wait... I don't have a PCIe 3 cpu. Oh well.
Blackshark 5th July 2013, 20:43 Quote
allowing manufacturers to install Xeon Phi co-processors in high-performance computing systems alongside, rather than in the place of, graphics processing unit (GPU) based accelerator boards

Excellent news
Gareth Halfacree 5th July 2013, 20:55 Quote
Quote:
Originally Posted by Blackshark
allowing manufacturers to install Xeon Phi co-processors in high-performance computing systems alongside, rather than in the place of, graphics processing unit (GPU) based accelerator boards

Excellent news
Yeah, nobody else seems to have picked up on that bit. If the slide is accurate, it's a game-changer for Intel - and it's hard to see it as anything other than a response to the threat that ARM poses. You look at the world's top supercomputers: in the vast majority, the Intel or AMD CPUs are only there to feed the GPUs with data - they account for 10 per cent or less of the overall compute power. As a result, manufacturers - including Nvidia - are looking at replacing the power-hungry x86 serial processors with nice low-power ARM chips.

Obviously, Intel can't allow that - but nobody's going to take them seriously if they suggest supercomputers based on Atom, and they can't be seen recommending GPU-based accelerator boards with Xeon chips. The solution, it would seem, is to socketise Xeon Phi and sell it as a 50-core CPU - not co-processor - for the HPC market. Then if customers stick Teslas in the spare PCIe slots, well, it's hardly Intel's fault, is it?

Imagine: a supercomputer that not only has 4,000 GPUs but 4,000 Xeon Phi chips - each of which offers 50 Pentium-class x86 cores. Massively parallel or wot?
Corky42 6th July 2013, 07:04 Quote
Quote:
Originally Posted by Gareth Halfacree
Yeah, nobody else seems to have picked up on that bit. If the slide is accurate, it's a game-changer for Intel - and it's hard to see it as anything other than a response to the threat that ARM poses.

Not only a response to ARM but anything that helps sell Xeon Phi chips hurts Nvidia's GPU business and its future server processor business as well as all these guys....
http://en.wikipedia.org/wiki/ARM_architecture#ARM_licensees
Quote:
Many semiconductor or IC design firms hold ARM licences: Analog Devices, AppliedMicro, Atmel, Broadcom, Cirrus Logic, Energy Micro, Faraday Technology, Freescale, Fujitsu, Intel (through its settlement with Digital Equipment Corporation), IBM, Infineon Technologies (Infineon XMC4000 32bit mcu family), Marvell Technology Group, MediaTek, Nintendo, Nvidia, NXP Semiconductors, OKI, Qualcomm, Samsung, Sharp, STMicroelectronics and Texas Instruments are some of the many companies who have licensed the ARM in one form or another.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums