Bobcat: AMD's Answer to Intel Atom, ARM Movement
8/24/2010 by: Theo Valich
Last week, AMD held a series of talks with journalists and analysts in regards to their upcoming CPU architectures known as Bobcat and Bulldozer. While Bulldozer is a successor to Phenom and first new architecture from AMD since SledgeHammer/K8/AMD64/x86-64 in 2003, Bobcat is something the company desperately needed in order to compete with Intel and ARM ecosystem in the sub-20W computer range.
AMD's Future Architecture Overview: Bulldozer and Bobcat
We've followed AMD closely and it is good to see that after years of roadmap slips and departures of important staff, the company is finally bringing Bobcat and Bulldozer to market. Needless to say, both of these architectures were supposed to launch in 2009, while the reality of AMD's execution pace is that we'll see products based on these architectures in 2011. This continues long line of architectural delays, as the original SledgeHammer was also 18 months late [architecture debuted in late 2001, product was launched in 2Q 2003].
What is Bobcat?
First and foremost, Bobcat architecture has one key differentiation between itself and the competition. Bobcat is not an in-order execution core such as ARM and Intel Atom architectures, as AMD opted for an Out-of-Order execution [OOO]. This decision, which is costlier in terms of transistors and power consumption - brings one major advantage, especially in processors equipped with just one core - performance.
AMD Bobcat Micro-Architecture
According to AMD's documents, Bobcat packs Aggressive OOO load/store engine with dual x86 instruction decode, 32KB instruction and 32KB data L1 cache, as well as 512KB of L2 cache. The core naturally supports pairing into multi-core setup, or CPU+GPU setup. AMD disclosed the codename of their first Bobcat-based CPU+GPU i.e. APU [Accelerated Processing Unit]: Ontario.
Bobcat is a full 64-bit processor, supporting AMD64 i.e. x86-64 ISA [Instruction Set Architecture], with all commodities of modern CPU world - SIMD extensions from SSE1, all the way to SSE4A [firstly introduced with K10 "Barcelona" architecture]. Intel already has a 64-bit architecture inside Atom processors, but they suffer from limitations of in-order execution.
Virtualization inside a Watt?
There are no secrets that the big trend in the server industry is virtualization and as such, both key CPU vendors are putting significant resources to make virtualization as fast and reliable as possible. Recently, ARM started to look into the HPC and virtualization space with their own architecture, as demonstrated by a recent $48 million investment in an ARM-based server processor start-up. Investors? ARM themselves, ATIC [majority owner of GlobalFoundries] and Texas Instruments.
Thus, AMD has to react or fall behind. Bobcat architecture packs a good part of AMD-V set of advanced virtualization technologies known from Barcelona [K10] and Shanghai [K10.5] architecture, supporting Nested Page Tables [NPT PDF Download], ASID and world switch acceleration.
Meet the First Bobcat APU: Ontario
AMD Bobcat Core plan: Add an 80-core Cedar GPU and you have - Ontario
Ontario is the codename of the first AMD APU [we'll rather call it SOC]: as you might have guessed, sometimes during 2009 AMD learned that Bobcat core is speeding ahead of their conventional Fusion project, i.e. K10.5 "STARS" core paired with a mainstream, 400-core Redwood GPU. As a result, Bobcat was stayed on the roadmap for launch in 2010, while Bulldozer and Fusion got pushed into 2011.
Ontario is manufactured in 40nm node over at TSMC, and features monolithic design pairing a Bobcat core and Cedar GPU, i.e. Ontario gives you DirectX 11 graphics in only 18W TDP. If that isn't impressive, we don't know what is.
Performance-wise, German site Hardware Infos got their hands on a leaked table that showed a 1.4GHz Ontario achieving 3.05 billion instructions per second [GIPS] and 1.35 billion floating point operations per second [GFLOPS]. Given that competing Atom D510 at 1.66GHz achieves 1.89 GIPS and 0.74 GFLOPS, these results put AMD in the lead, ahead of Intel and ARM-powered parts.
To put things into perspective, AMD's low power Athlon II X2u running at 1.6GHz gives out 3.47 GIPS and 1.59 GFLOPS. Still, the fact that a similar power envelope now gives you DX11 capabilities and additional computing performance through ATI Stream and OpenCL is certainly worth mentioning.
Conclusion: 64-bit enters the embedded stage
Even though Bobcat and Bulldozer are inexcusably late, which has mostly to do with Dirk Meyer's statement made to several executives that Intel Atom does not pose a threat [which in turn resulted in several key executives leaving AMD], Bobcat looks pretty good on paper. Core is able to go into sub-1W state [C6], the architectural improvements pitch the part above K8 efficiency and so on and so forth.
But most importantly, AMD is bringing 64-bit and full SSE support into a part that will see most of its life in netbooks, tablets and - embedded electronics. Given the recent movements in automotive, airline, cell phone and consumer electronics industries - the world needs this part and to kickstart the Atomic war [pun intended].
Tags:
AMD, Bobcat, HotChips, HotChips22, HotChips conference, Atom, Intel, INTC, ARM, Qualcomm, Freescale, Low Power, TDP, Thermal Design Power, Cedar, DirectX, DirectX 11, DX11, OpenCL, STREAM, GPGPU, APU, CPU, SOC
© 2009 - 2011 Bright Side Of News*, All rights reserved.