AMD Details Upcoming 32nm Fusion Processor, Codenamed "Llano"
AMD revealed more information on its first Fusion-class product at ISSCC this week. The new chip will combine both a CPU and GPU on a single package, but don't use either nickname—AMD is calling the new chip an "APU" or Application Processor Unit as part of its bid to shift product focus from the core components in a system to the tasks and workloads a system is capable of handling. Done properly, this could solve certain nomenclature confusions that have plagued the computer market from the very beginning, but we aren't holding our breath. Both AMD and Intel have tackled this type of rebranding before (think AMD Live! and Intel ViiV), with minimal success. For now, however, we'll roll with the APU designation.
The first thing to know is that Llano isn't Bobcat. In an article last month, we speculated that AMD's delayed transition to 32nm could impact the company's plans to launch a 32nm Phenom II-class processor, but this has not occurred. Llano is a 32nm derivative of the company's current architecture. Given the new details AMD has released, this makes good sense—Llano will target a wide range of operating environments (2.5 - 25W) compared to the 1-10W range AMD will target with Bobcat. The 2.5-25W operating envelope still represents a substantial step forward for AMD; the company's lowest-power 45nm mobile processors have a TDP of 25W (dual-core) or 15W (single-core).
One core of AMD's Llano. The chip doesn't have an L3 cache but allocates 1MB of L2 per core.
According to Senior AMD Fellow Samuel Naffziger, Llano will be a quad-core processor with 1MB of L2 cache per core, a 3GHz-or-higher operating speed, support for DDR3 memory, and an on-die DX11-class graphics processor. The die is also relatively svelte, at ~35 million transistors and 9.69 sq mm (excluding the 1MB of L2 cache, as shown above.) The big news on Llano, however, concerns the chip's power management and throttling technology. Mobile power consumption has always been a relative weak spot for AMD compared to Intel; an issue Llano should address when it
ships in 2011.
The new 32nm chip will introduce core power gating (a method of disconnecting power to the core when it's not in use); AMD claims its method of implementing this feature will reduce leakage power by 90 percent. Llano will also measure CPU power consumption digitally, by tracking a series of 95 signals from various parts of the die. At present, all CPUs, including Intel's, rely on analog thermal sensors that provide information on how hot the chip is running (and therefore, how the CPU clock should be adjusted). AMD claims that this digital approach gives it much more fine-grained control over clock rates than traditional analog measurements are capable of reporting. Finally, AMD reports that it's been able to "dramatically reduce the amount of metal and buffering in this system to reduce clock switching power by an estimated factor of 2."
AMD's Upward Mobility
Llano is currently scheduled to sample to OEMs during the second half of 2010 and hit the market 'sometime' in 2011. While we don't know anything about Llano's GPU as of yet, AMD has always emphasized the importance of a strong graphics processor in any system; we expect the company's DX11-class part to significantly outperform the IGP Intel currently includes with Clarkdale and Arrandale. Depending on when Llano drops, however, AMD could actually face competition from the mobile flavor of Intel's Sandy Bridge, which will integrate a 32nm GPU as opposed to the current 32nm CPU / 45nm GPU design.
Llano may not do much to close the CPU performance gap between AMD and Intel, but the new design could deliver substantially improved battery life as well as modest performance gains compared to AMD's current 45nm mobile offerings. Given the consumer market's current focus on long battery life and low power systems, Llano should be a definite step forward, particularly if its GPU performance can compensate for a relatively weaker CPU. AMD's decision to focus so aggressively on improving Llano's power consumption implies that the chip is more than a quick patch to carry the company between 45nm mobile parts and 32nm Bobcat. Given the degree of overlap between the two architectures' targeted TDP ranges, it's reasonable to think we'll see Llano scaling into smaller/lighter notebooks that might compete directly against Intel's CULV products, while Bobcat targets the netbook market currently owned by Intel's Atom.
If you want more information on how AMD and Intel are positioned for the coming quarter and what we expect from the two companies in 2010, check our reports here and here.
The first thing to know is that Llano isn't Bobcat. In an article last month, we speculated that AMD's delayed transition to 32nm could impact the company's plans to launch a 32nm Phenom II-class processor, but this has not occurred. Llano is a 32nm derivative of the company's current architecture. Given the new details AMD has released, this makes good sense—Llano will target a wide range of operating environments (2.5 - 25W) compared to the 1-10W range AMD will target with Bobcat. The 2.5-25W operating envelope still represents a substantial step forward for AMD; the company's lowest-power 45nm mobile processors have a TDP of 25W (dual-core) or 15W (single-core).
One core of AMD's Llano. The chip doesn't have an L3 cache but allocates 1MB of L2 per core.
ships in 2011.
The new 32nm chip will introduce core power gating (a method of disconnecting power to the core when it's not in use); AMD claims its method of implementing this feature will reduce leakage power by 90 percent. Llano will also measure CPU power consumption digitally, by tracking a series of 95 signals from various parts of the die. At present, all CPUs, including Intel's, rely on analog thermal sensors that provide information on how hot the chip is running (and therefore, how the CPU clock should be adjusted). AMD claims that this digital approach gives it much more fine-grained control over clock rates than traditional analog measurements are capable of reporting. Finally, AMD reports that it's been able to "dramatically reduce the amount of metal and buffering in this system to reduce clock switching power by an estimated factor of 2."
AMD's Upward Mobility
Llano is currently scheduled to sample to OEMs during the second half of 2010 and hit the market 'sometime' in 2011. While we don't know anything about Llano's GPU as of yet, AMD has always emphasized the importance of a strong graphics processor in any system; we expect the company's DX11-class part to significantly outperform the IGP Intel currently includes with Clarkdale and Arrandale. Depending on when Llano drops, however, AMD could actually face competition from the mobile flavor of Intel's Sandy Bridge, which will integrate a 32nm GPU as opposed to the current 32nm CPU / 45nm GPU design.
Llano may not do much to close the CPU performance gap between AMD and Intel, but the new design could deliver substantially improved battery life as well as modest performance gains compared to AMD's current 45nm mobile offerings. Given the consumer market's current focus on long battery life and low power systems, Llano should be a definite step forward, particularly if its GPU performance can compensate for a relatively weaker CPU. AMD's decision to focus so aggressively on improving Llano's power consumption implies that the chip is more than a quick patch to carry the company between 45nm mobile parts and 32nm Bobcat. Given the degree of overlap between the two architectures' targeted TDP ranges, it's reasonable to think we'll see Llano scaling into smaller/lighter notebooks that might compete directly against Intel's CULV products, while Bobcat targets the netbook market currently owned by Intel's Atom.
If you want more information on how AMD and Intel are positioned for the coming quarter and what we expect from the two companies in 2010, check our reports here and here.
