Tag Archive | "a4"

ARM Products and Platforms Primer and Resource List for Mobile Internet Devices in 2010. (Updated)

Tags: , , , , , , , , ,


This article is updated from Jan 2010. See below for history.

The ecosystem around ARM, its designs and licensee products can be difficult to understand so I’ve been researching the current status and have summarized the important products, brands and technologies in the article below. You’ll find details on all the important keywords and technologies, links to all the important CPUs and platforms, a reading list and some ‘tip’s’ for CES2010 which starts this week. All the information is based on my own knowledge and research so if you spot any errors, please be sure to let us all know in the comments. I know there are some CPU designers and ARM partners reading this site so again, if there’s anything that needs changing or adding, please help us all by adding a comment below.

armarch

Overview of processor architectures in current ARM products

For the purposes of this article I’m ignoring anything other than ARM v7 architecture CPUs. In my opinion and experience the previous (ARMv6, ARMv5) designs don’t provide the processing power needed for the quality web experience expected in  MIDs. Note: Tegra (current version 2009) is an ARM11 multi-core CPU  implementation based on ARMv6 architecture. I expect Nvidia to move to ARMv7 in their next Tegra product.

ARM v7 is a CPU architecture, not a CPU.

ARM V7 is a processor architecture. It’s a design that belongs to ARM and it can be used to build CPUs that can process around 2 Dhrystone MIPS/Mhz. (Wikipedia – Dhrystone) That’s about 2000 DMIPs/Ghz which is not far from the processing power achieved by simple, non-hyperthreading Intel Atom cores also used in MIDs. DMIPs processor performance is not the only measurement of device speed but it’s a good starting point and is relevant to web rendering and web applications.

Processing power

In general, the architecture is being used to make CPUs and platforms running at between 600-1000Mhz (about 1200-2000 DMIPs) although there is one known implementation that has been tested at over 2Ghz. Power consumption per core is said to be around 300mw per 2000DMIPs using the latest manufacturing processes although this figure can vary greatly depending on implementation.

Remember that the CPU core takes only a small percentage of the power drain in a working device where the total in-use power budget including screen lighting, radios, audio, gpu, storage and DC components can span from 2W to 10W. (Texas instruments thinks that a 1W MID is possible though)

In comparison with the Intel Atom CPU the ARM v7 architecture can be used to make CPUs that consume about 1/3 – 1/5th of the power of an Intel Atom CPU for the same DMIPs computing power.

ARMv7 can also be used to make multi-core CPUs where up to 4 cores can be used to provide over 8000 DMIPS of computing power. (Assuming the software is built to handle multi-processing hardware)

(Further note: All these figures based on research, marketing figures, experience, testing and technical documents that I have read during this research.)

ARM does not manufacture CPUs.

The architecture created by ARM is used to do two things:

  • The architecture is used by ARM themselves to make a complete processor implementation which may include other ARM property. The CPU implementation is then licensed out to third parties who can either mass produce the CPU or build and manufacture complete computing platforms including graphics, sound, power, memory, etc. One example is the Ti OMAP 3 and 4-series platforms.
  • The architecture gets licensed out to third parties who make their own processors and platforms based on it. One example is the Qualcomm Snapdragon platform.

ARM have a number of processors that they’ve built using ARMv7 architecture and all fall under the ‘Cortex’ brand. There are real-time and highly embedded versions but the ‘A’ versions are the ones that interesting for general mobile computing tasks. ARM have three versions of the CPU. The A8 (currently in products), A9 (high end, multi-core capable due in products in 2010) and A5 (small, low-cost, due in products in late 2010) versions

Licensees.

When final products are completed by licensees you will often see them marketed under different brands. This is where it gets very confusing so I’ve listed most of the main players below along with notes and links to their various products and brands. Note that some ARM licensees are not made public and therefore no information is available.

ARM Cortex A8

ARM has 9 licencees. 8 are public.

Cortex A8 Product brief (PDF)

Licensee notes.

  • Ti OMAP 3 platforms 34xx 35xx 36xx  using Cortex A8 CPU core up to 800Mhz. Used in Nokia N900, Archos 5 devices for example.
  • Freescale iMX5 family of CPUs based on Cortex A8. Use in the Sharp Netwalker for example. iMX515 is focused at mobile internet and includes graphics co-processing.
  • Samsung. S5PC100 application processor includes the Cortex A8 CPU core (E.g. Odroid)
  • Samsung / Intrinsity – Hummingbird A very specific implementation of the Cortex Core using a tightly defined manufacturing process.
  • Zii Labs ZMS08 Core of the ‘stem cell’ computing platform. (Q1 2010 volume shipment) 1Ghz implementation.
  • Matsushita (Panasonic) Details unknown.
  • PMC-Sierra (storage, switching, routing solutions) Details unknown.

Cortex A9 (Multi-core capable)

Cortex A9 Product information (PDF)

ARM has 9 licencees. 6 are public.

Licensee notes.

  • Ti OMAP Dual-core 44xx platform. Full production expected 2H 2010.
  • ST Electronics Cortex A8 licensee (set top boxes)
  • St Ericsson – U8500 Dual Cortex A9/HSPA Modem platform. (Note Nokia and Symbian will be using this and it includes a Mali GPU and 1080p recording capability)
  • Broadcom (ARM news. No products announced yet.)
  • Nvidia Tegra 2 Dual-core
  • NEC Electronics. No information available.
  • Update: 28 Jan 2010. It appears that Apple may have licensed Cortex A9 for the iPad A4 CPU. (Source) Update: It’s ARMv7. Apples own implementation.

Cortex A9 MPCore Hard Macro – 2Ghz implementation of Cortex A9 using specific TSMC silicon manufacturing process.

Cortex A5

Cortex A5 is a multi-core, low cost ARM V7 implementation previously known as Sparrow.

Link to ARM information

No known implementations at present.

Other ARM v7 implementations:

  • Qualcomm Snapdragon platform (QSD8x50) uses single core CPU (Scorpion) based on ARMv7 architecture.
  • Qualcomm QSD8672 dual core platform at up to 1.5Ghz.
  • Marvell Armada 500 / 510 platform (PDF product brief) Up to 1.2Ghz

Notes for CES 2010 (UPDATED)

Key platform announcements to watch out for at CES 2010. Expect demonstrator products planned for 2H 2010. (All Cortex A9)

  • Nvidia Tegra 2 – Announced.
  • Ti OMAP 44xx – No news.
  • St Ericsson U8500 – Demonstrated by Movial on a set top box.

Expected in MIDs and smartbooks shipping in 1H 2010 on the following platforms. (Single core ARmv7 and Cortex A8 implementations.)

  • Freescale iMX5
  • Ti OMAP 3
  • Qualcomm Snapdragon
  • Nvidia Tegra
  • Zii Labs ZMS08

Product rumors / expectations: CES 2010.

Further reading:

Update: ARM presentation at CES.

I had a chat with ARM at CES and they followed up by sending me a PDF. Here’s one slide from that PDF that is directly relevant here. It lists ‘candidate’ platforms for Mobile Computing. We listed all of these platforms above but it’s good to know what ARM sees at potential platforms.

Click to Enlarge.

Note to PR agencies for ARM ecosystem partners. Put one twitter/web address in the comments and we’ll add it to the further reading list.

If you find the information useful for your work, please consider a small donation to help us continue the work. UMPCPortal is independent and funded purely through donations and advertising revenue.  Many thanks

Updated: 14th Jan 2010

Updated: 18th Jan 2010 (added ARM slide)

Updated: 28 Jan 2010 (added possible Apple licensee (for A4 processor) information.

Updated: 12 May 2010 with more product information

How Fast is Moorestown for Browsing? Faster than an iPad? Does it Matter? Analysis and Simulated Test Video

Tags: , , , , , , , ,


ipad_viliv In press events on Tuesday, Intel launched Moorestown and gave journalists their full marketing package on the smartphone and tablet-focused platform.  The key highlight was ‘performance’ and one element I want to focus on is Web browsing.

In tests I’ve been doing with many devices over the last 6 months using the SunSpider javascript benchmark, the Atom CPU, running at 1.3Ghz, whips the A4 CPU into the ground with a >3X speed advantage. Moorestown, with its 1.5Ghz clock looks to improve that to a 4x speed advantage according to the slide below. That’s a huge win for web-based applications implemented in AJAX.

Of course, javascript is only a component of the total time needed to fetch and render the average web page so I decided to do a real world test.

moorestown-web

To simulate how Moorestown would perform under web-browsing conditions I took the Intel Atom-based Viliv X70, a 7” tablet with a 1024×600 7” touchscreen running Windows XP Home, installed the latest Google Chrome, locked the CPU to 800Mhz and did some random web browsing tests. Remember that the Menlow platform used in the X70 is very close to the architecture used on Moorestown. The CPU are GPU are architecturally almost exactly the same. I side-by-sided it with an iPad which is, according to my tests with the Archos 5, X10 and HTC Desire (all running high-end ARM V7 architecture cores) the fastest ARM-based browser solution out there.

The video below shows that the browsing speed with the Viliv at 800Mhz is almost neck and neck. An 800Mhz Atom on a multi-tasking OS matches a 1Ghz A4 on a single-tasting OS. Ignore the UI and product, this is just a test of web page loading speed.

In the second part of the video I boosted the Atom CPU up to 1.3Ghz, the maximum on this platform, to simulate what would happen when a Moorestown smartphone ran at the highest clock-rate of 1.5Ghz. Remember the CPU and GPU architectures in Menlow and Moorestown are the same although Moorestown has a 200Mhz advantage here, can support faster memory and has a faster GPU clock.

The difference is very noticeable with the X70 rendering pages much more quickly, even with Flash enabled. Move to Firefox and disable flash and the difference is even bigger.

I see real-world advantages here. Faster, full Internet experience and a huge advantage for web-based applications and compressed or encoded content although it has to be said that in this high CPU-load scenario, battery drain on the Moorestown platform is likely to be slightly (although not considerably) more.

Remember, we’re ONLY comparing CPU platforms here and in this simulation, the Moorestown platform is showing great potential. It can deliver web pages, process script, decrypt HTTPS, GZIP and images much faster than the best ARM-based solution out there. It also adds multitasking and large memory support too. As a platform, if it delivers on the battery life claims, Moorestown is going to be a great, high-performance smartphone, tablet and even netbook option.

Recommended reading – Why social netbooks have a ‘lock-in’ opportunity. In this article I talk about key features of a smartbook. Many of these apply to a Moorestown/MeeGo-based product.

Recommended reading – Anandtech on Moorestown. Detailed with good background research and knowledge.



Recommended Sites

  • Big Beach Our marketing advisors. Located in UK.
  • Carrypad Tablets and consumer mobile products
  • JKKMobile JKKMobile – Mobile computing and hacking – Partner of UMPCPortal
  • Liliputing Netbooks and small-form-factor laptoping – Partner of UMPCPortal
  • Ultrabook News Ultrabook products, specs and news