backtop


Print 17 comment(s) - last by french toast.. on Mar 25 at 12:22 PM

New ARM superchip will challenge Snapdragon 4 for market dominance

Intel Corp. (INTC) has some nifty fab-related tricks up its sleeve like its almost-here 22 nm die-shrink and its leakage-fighting (current leakage, that is) 3D FinFET transistors.  Still, it has to be a bit nervous in the face of its biggest competition in the last couple decades -- an onslaught of power-efficient ARM architecture CPUs.

Thus far we'd heard about Snapdragon 4, Qualcomm Inc.'s (QCOM) system-on-a-chip that's expected to lead ARM's push into the CPU space.  Now fresh details have leaked about a key rival design -- the Exynos 5 SoC from Samsung Electronics Comp., Ltd. (KS:005930).

I. The Process and CPU 

Exynos 5, according to a company poster photographed by Semiaccurate, will be built on Samsung's 32 nm process.  Compared to Qualcomm's 28 nm SoCs, manufactured by Taiwan Semiconductor Manufacturing Comp., Ltd. (TPE:2330), one might expect these chips to be inherently bigger and less power-efficient due to the larger feature size.  

However, Samsung's chip is second-generation 32 nm silicon as it was the only ARM chipmaker to reach the node in 2011.  And TSMC is reportedly having serious issues with its 28 nm process, causing it to suspend 28 nm production in mid-February [Source: Semiaccurate], a delay that sheds some light on the curious clock-speed drops in the latest generation of Advanced Micro Devices, Inc. (AMD) and NVIDIA Corp. (NVDA) GPUs (both companies rely on TSMC for their GPU chipmaking). 

In other words Samsung High-K Metal Gate (HKMG) doesn't exactly push the die-shrink envelope, but it's far more proven and lower risk than the TSMC process used by Qualcomm.

(Image removed at the request of Semiaccurate, see Editor's Note in addendum.)

On a CPU core side, Exynos 5, like Snapdragon 4, is a Cortex-A15 derivative (in Exynos 5's case a direct core, in Snapdragon 4's case an "Cortex-A15-like" Krait core).  Cortex-A15 is a licensed design from ARM Holdings plc (LON:ARM), which uses the tried-and-true ARMv7 instruction set.  Like the majority of Qualcomm's lineup, the CPU will be a dual-core design.

Another indication of the stability of Samsung's 32 nm node can be observed in the clock speeds.  While Qualcomm is offering up Snapdragon chips clocked at between 1.5 and 1.7 GHz, Samsung is promising 2 GHz chips.  Of course, Qualcomm recently showed off a quad-core 2.5 GHz Snapdragon 4 (the APQ8064), but this will likely be a lower volume part that relies on special binning.

II. The GPU

The Samsung SoC also packs a T-604 MP4 "Mali" GPU, an intellectual property (IP) core licensed from ARM Holdings.  The T-604 core-wise is roughly the equivalent of Imagination Technologies plc's (LON:IMG) PowerVR SGX543, so the onboard GPU is comparable to the PowerVR SGX543 MP4 found in the iPad 3, and expected to be quite powerful.

Samsung is licensing ARM Holdings "Artisan" physical IP, process improvements that help put the company's IP cores (Cortex-A15, Mali) to working silicon.

Semiaccurate also snapped a blur-tastic photograph of what it believes to be a Exynos 5 development board.  

(Image removed at the request of Semiaccurate, see Editor's Note in addendum.)
 
Expect Exynos 5 to mostly pop up in PCs due to the high clock speeds and higher power requirements.  The interesting question is whether Samsung -- a computer maker -- will allow third parties to use its chip, given that its first-hand PC sales are quite low.  

An underclocked variant of the chip could, in theory, show up in tablets as well.  Essentially it's quite similar to Apple's 1 GHz A5X (except the A5X using Cortex-A9 CPU cores and has a lower clock), found in the iPad 3.  Of course, putting such a beefy core on board would likely demand an equally beefy battery, such as the 55 watt-hour monster lurking in the iPad 3.

Editor's Note:
We were made aware of a pair of Twitter posts attributed to Semiaccurate on the site's Twitter account [1][2] complaining about us "ripping off" their site with this story.  We were a bit baffled by this, given that we cited them four times in the unedited version of this piece -- once in the "Sources" link, once in a link in the body of text, once via using their watermark on the attached images, and a fourth time in "Image Sources" text under each image.

Since this was a public critique, we wish to offer some clarification to you on these comments and why the pictures disappeared.

It appears that Semiaccurate was upset about my use of their pictures -- despite the numerous citations and live links.  Their policy (which I was unaware of) is that any reprinting of pictures requires individual email requests (read).  Having worked the tech news world for five years, I can safely say this is highly unusual as nearly every other website is just looking for fair credit -- a link and a citation.  I myself practice this policy and am proud to say that my images -- while not exactly Ansel Adams quality have popped up in a variety of reprints.  And I'm thankful for every one, just like I'm thankful to every site that links or cites DailyTech.

Our site requires permission for full reprints (as do most sites), but when it comes to pictures or small quotes we follow the industry standard -- all is good if you cite and (preferably) link.

That said DailyTech respects Semiaccurate and values them as a source.  If they do not wish their pictures to be used, we will not use them, and apologize for disrespecting their wishes by assuming they were safe to repost.

For the record, Semiaccurate wrote that it had previously "talked to" me about "ripping off [their] pictures".  To my knowledge this is semi-inaccurate (pun intended).  I keep all my emails (love and hate letters alike!), for reference sake, and have scoured them finding nary a single correspondence from the site.  About the only thing I can think of is that perhaps they mentioned something to me at a trade show like CES.  While possible, I have to say I wish (if that's the case) they followed up with me as you tend to meet a million people at CES (not literally, but close) and things tend to get lost in the noise. 


UPDATE (3/21): Glad to tell you I am not going crazy.  So I heard back from Semiaccurate and they confirmed they never actually talked me, despite that being what they wrote in the attacking Twitter post.  They have offered to post an apology/clarification, so expect that pretty soon!  Glad this misunderstanding is getting cleared up!

Or maybe the site posted a similar past Twitter complaint and expected me to read it?  Who knows. Regardless, we have not talked in depth on this topic.

We find it unfortunate that Semiaccurate raised this issue to us via a social media platform, rather in a professional capacity via email.  We hate this kind of drama as it compromises our ability to deliver the best content (properly attributed, of course) to you, the reader.  We have reached out to Semiaccurate, seeking a solution. 

We apologize to our readers for the incovenience, and for being unable (for now) to display Semiaccurate's pictures with the story.

Source: Semi-Accurate



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: So... when
By french toast on 3/25/2012 12:22:45 PM , Rating: 2
Nope,Quad core doesn't mean less battery life....it can actually mean more!

Tegra 3 quad core actually consumes slightly LESS power than Tegra 2..despite having more cores, running at a faster speed, and a more powerfull gpu..all on the same manufacturing process.
(Although Nvidia has that 'shadow core' that helps reduce the consumption)
So quad core doesn't mean more power by default..

-For a start, although there is no shadow core its on the best manufacturing process for ARM chips..Sammy's 32nm HKMG gate first process.(compared to 40nm non HKMG Tegra 3)

-On heavy threaded workloads ie...web browsing, multitaskking, (especially android) the workloads can be spread across the cores more evenly..allowing them to clock lower and reduce power consumption whilst increasing performance and 'smoothness'...

-Cores can be power gated off when not in use..reducing power consumption.
Power draw does not scale up linealy with clock speed...the higher you go up the more power it takes.

About performance, if you look at Anands review of krait, apart from linpack, if you leveled the A9's clock speed to 1.5ghz..it would perform about the same as Krait...with the quad A9's thrashing duel core Krait on heavilly threaded workloads.

The only thing that is going to be dissapointing is the GPU which will use a faster clocked variant of last years winner Mali400 mp4 @400mhz.
-But should still be faster than Tegra 3..was hoping for T604 though ;(


"The Space Elevator will be built about 50 years after everyone stops laughing" -- Sir Arthur C. Clarke














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki