backtop


Print 64 comment(s) - last by xti.. on Feb 26 at 11:39 AM

Late delivery of mobile parts cost AMD

As a perpetual underdog, Advanced Micro Devices, Inc. (AMD) has little wiggle room for lapses in execution.  Unfortunately that's just what happened to the chipmaker in Q4 2013, according to a new report from Jon Peddie Research (JPR).
 
I. Where Are the Mobile GPUs, AMD?
 
Even as AMD hungrily eyed future ARM-based server chips, it stumbled in the system-on-a-chip mobile graphics market.  According to JPR, the quarter's overall sales trends (quarter-on-quarter) were:
AMD ::::::::::::::::::::::: -10.4 %
NVIDIA Corp. (NVDA) :::  +3.4 %
Intel Corp. (INTC) :::::::  +5.1 %
 
The slump in AMD GPU sales came thanks to delivering to slowly for OEMs on the parts front.  Comments Jon Peddie:

AMD's shipments of desktop APUs (heterogeneous GPU/CPUs) jumped 15% from the previous quarter but declined 26.7% in notebooks. AMD's discrete desktop shipments increased 1.8%, and notebook discrete shipments declined 6.7%. The company's overall PC graphics shipments decreased 10.4%. Notebook build cycles are specific, and AMD was late with its new parts.

On the desktop front, AMD appeared to be on fire.  It launched its new 28 nm "Volcanic Islands" GPUs (soft launch: September 2013; shipments: October 2013), branded as the Radeon Rx 200 Series.  The new chips featured an upgraded architecture, Graphics Core Next 1.1 (GCN 1.1), which refined the original GCN architecture introduced with the Radeon 7000 series.  The new version of GCN offered some intriguing possibilities, such as using your GPU for advanced audio processing.  Desktop buyers clearly were hooked.
AMD Asus
Desktop buyers coveted AMD's Rx 200 Series cards, such as the R9 280X card pictured here. [Image Source: AnandTech]

AMD currently only has launched the 28 nm mid-to-high end components of the Rx200 Series lineup on the mid-to-high end.  The lineup is led by the flagship R9-branded GPUs, which include Hawaii, Tahiti, and Curacao chips.  In the middle are the R7-branded GPUs, which include Pitcairn, Bonaire, Oland, and -- most recently Cape Verde (launched earlier this month).  Noticeably missing are the R5 GPUs -- budget chips.  Other than Oland (the priciest R5), the other R5 are supposed to be built a 40 nm process.  The missing chips go by the codenames Caicos and Cedar.
 
AMD Volcanic Islands

Also confusing is what happened to Iceland and Maui, two other codenames that popped up on hardware benchmarking sites, stoking rumors.  While there's no concrete evidence, some have suggested that these high end chips are 20 nm R9 chips which AMD had hoped to launch by now, but were scrapped amid struggles with die shrinks at Taiwan Semiconductor Manufacturing Comp., Ltd. (TPE:2330).  Yet another missing desktop chip is the Tonga.

Aside from the missing R5 stock, AMD's laptop Rx 200 Series chips are also notably absent.  These chips were rumored to codenamed Crystal.  The higher end models would be dubbed Topaz, and would be the sister chip to Iceland.  The mid-range models would be dubbed Amethyst and would be cut down versions of Tonga.  But like their desktop brethren, there's no indication of when or if AMD is launching these chips.
Topaz and amethyst
The Topaz and Amethyst mobile GPUs were rumored to land in 2013, but have yet to launch.
[Image Source: Jewelry News Network]

Of course it's worth stating that some of these GPUs were likely renamed or rolled into other lines (e.g. perhaps Maui became one of the two Hawaii chips?).  But complete lack of new laptop product and the absence of the already announced R5 chips suggest that AMD is struggling to scale its new designs to the budget and mobile markets.
 
Last fall rumors were beginning to focus on the successor to Volcanic Islands -- Pirate Islands.  Now the rumor mill is asking -- where did all the supposed Volcanic Island chips go?
 
II. New APU Stock Also Struggles to Find Its Way to the Market
 
On the accelerated processing unit" (APU) front AMD suffered from slow execution, as well, although its plans are at least clearer.  AMD has already announced its intention to launch new A4, A6, and A10 APUs for laptops and desktops in H2 2014. 
 
AMD's upcoming APU line is anchored by the fourth-generation, 28 nm Kaveri (A10) APU, a capable design based on AMD's new Steamroller architectureSteamroller packs 3 more ALUs per core and other improvements to improve on both parallelism and single core performance.  The on-chip GPU also receives a nice bump; it features 512 GCN 1.1 SIMD cores (stream processors) packaged into 8 compute units (CUs), marking the first time AMD's APUs will have fully caught up with their discrete brethren in core design.

Kaveri v. Richland Steamroller Die
A Kaveri chip (far left) aside a unpackaged Richland chip (middle); a Steamroller core is on the right. [Image Source: Overclock.net (left); AMD via ExtremeTech (right)]

On the lighter-weight more mobile-centric side of things are Beema (A6/A4) and Mullins (A4), which pack the brand new Puma core design.  Puma replaces the Jaguar cores found in AMD's Temash and Kabini platforms -- the chips that comprise AMD's current E1/2 branded lineup, as well as much the A4/A6 branded lineup.  The upcoming chips are also expected to pack a ARM Cortex-M5 processor licensed from ARM Holdings Plc (LON:ARM).
 
That all sounds great -- the problem is that the cores aren't here yet and it's unclear when they will be.  And OEMs aren't willing to wait around.
 
III. Intel Gets Aggressive on the Embedded Graphics Front
 
When it comes to integrated GPUs, Intel was an early starter, but arguably a late bloomer.  Way back in 1999, Intel was offering integrated graphics chipsets with the Intel i810.  AMD wouldn't offer its first integrated graphics chipset until Feb. 2007.
 
Likewise, when it came to the eventual exit of the integrated GPU from the chipset and migration to the system-on-a-chip package, Intel led the way.  Its 2010 Clarkdale chip packed a 45 nm GPU die packaged along side 32 nm Westmere cores.  It followed in Jan. 2011 with Sandy Bridge, which brought the first mass-market personal computer chip with a GPU and CPU printed together on a unified die.
 
But those accomplishments were largely underappreciated in light of what AMD was doing at the time.  After teasing at Fusion (APUs) in 2007, the on-chip graphics from AMD became official in August 2010.  At the time AMD was firing on all cylinders, having just passed NVIDIA in July 2010 to seize the lead in the discrete graphics market.
 
And AMD delivered on the APU hype in Jan. 2011, launching Brazos, which was joined by the desktop-centric Llano in May 2011.  While admitted weak in CPU performance, the first-generation Fusion chips were crucial to convincing buyers that an integrated graphics chip could deliver a decent experience and even low-end gaming.  It blew away Intel's competitive GPU offerings at the time and was relatively competitive on the power front.

AMD Fusion

Now the tables have turned.  With its latest Haswell chips, Intel at last has on-die GPU (dGPU) cores that are drawing begrudging nods from enthusiasts.  While still trailing discrete GPUs in performance, Intel's new chips saw a massive performance boost with the new "Iris" branded cores, which pack a "Crystalwell" -- a unified block of memory shared by the CPU and GPU.

The JPR study reveals something rather interesting.  In Q4 99 percent of Intel's non-server chips featured built in GPUs, yet for all its hype AMD only had GPUs onboard 67 percent of its CPU chips -- roughly a third of AMD's shipped CPUs had no onboard GPU.  AMD is still using integrated graphics chipsets (IGPs) on the low end.
Iris Pro graphics
With Iris Pro, Intel stole the GPU crown from NVIDIA. [Image Source: PC Gamer]

While initial Iris products were relatively scarce, supply has started to open up with both Apple, Inc. (AAPL) and Taiwan's ASUSTek Computer Inc. (TPE:2357) push their Iris equipped products out to consumers. 

iMac EvolutionApple's new "evolved" iMac features Iris Pro inside its Intel Haswell brains.

Some would call this a recovery for Intel, but arguably it's more a case of a pioneer finally getting credit where credit is due.

III. NVIDIA Piles on

AMD meanwhile is on the defensive.

Its headaches are constrained solely to Intel.  It also has to worry about a rejuvenated NVIDIA.  NVIDIA has focused heavily on the mobile space with its Tegra cores.  While Tegra has ultimately failed to achieve NVIDIA's most ambitious mobile market adoption hopes, it's been a steady seller.

And while AMD's mobile GPUs were MIA on the mobile front, NVIDIA's mobile lineup was keeping pace with its desktop offerings.  Last April and May it diligently released a broad range of GeForce 700M chips.  AMD did release the Radeon HD 8000M series last January and would go on to score some design wins.  But as the year dragged on and NVIDIA fired back with GeForce 700M Series, AMD had no answer.

NVIDIA Tegra K1


NVIDIA was able to grow its desktop shipments as well, by a modest 0.9 percent, but when you added in the mobile Tegra chips, it reached 3.4 percent growth in Q4 for tablets+laptops+desktops.

Looking ahead NVIDIA is prepping for a huge launch with the announcement of Maxwell earlier this week.  NVIDIA is claiming doubled performance per watt, thanks to major changes to its control architecture.  Maxwell-sporting product will ship later this month, including the GeForce GTX 750 and GeForce 750 GTX 750 Ti mid-end cards.

Maxwell control logic

Last year NVIDIA, AMD, and Intel shipped 446 million GPUs (including dGPUs and IGPs). JPR expects the PC graphics market to overhaul dipped to 422 million units by 2017, as tablets displace some PC sales.  JPR also reveals that today roughly a third of PCs shipped have a discrete GPU, while two-thirds rely on on-die or integrated graphics.
 
Q4's PC GPU sales saw a 1.4 percent rise over Q3 2013, but were down 8.5 percent from Q4 2013, yet another sign of the PC's sales struggles.

Source: Jon Peddie Research [email]



Comments     Threshold


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

RE: Self-discrediting article
By lagomorpha on 2/21/2014 9:24:57 AM , Rating: 2
quote:
Obviously intel would have no marketshare in discrete gpus.


My work computer has a discrete Intel video card. :(


RE: Self-discrediting article
By lol123 on 2/23/2014 9:54:55 PM , Rating: 2
What kind of card is that?


RE: Self-discrediting article
By lagomorpha on 2/24/2014 1:46:13 PM , Rating: 2
i740


"Intel is investing heavily (think gazillions of dollars and bazillions of engineering man hours) in resources to create an Intel host controllers spec in order to speed time to market of the USB 3.0 technology." -- Intel blogger Nick Knupffer














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