Rory Buszka
Partition Master
From the Techgage News: Things have certainly been rough for AMD as of late. Today, AMD's CEO, Hector Ruiz, stepped down from his position at the company. There is no way of knowing whether or not his departure is a result of the company's performance, but after Jerry Sanders' and Phil Hester's leave, you have to wonder what the plan is for the future at the company.
With all of that said, I, as well as many others want to see AMD truly step up in the consumer CPU industry. While AMD probably won't leave us any time soon, with Intel all but dominating the space, the migration back over to Intel CPUs will leave the company in a hard position to compete with. -- Matt Serrano
The CEO of chip maker Advanced Micro Devices Inc. is stepping down.
Hector Ruiz had been just the second person to lead AMD after company founder Jerry Sanders. He'll be replaced by the chip maker's No. 2 executive, Dirk Meyer.
My Take: For quite some time, now, AMD's former CEO Hector Ruiz has drawn much ire from many tech industry analysts, who blame his unresponsiveness for many of the firm's current troubles (which at present have primarily to do with their Athlon- and Phenom-family x86 CPU products.) Now that he's stepping down, AMD's board of directors would do well to replace him with someone who can tap into the company's engineering prowess to produce CPU products that have broader market appeal and push the envelope of innovation. AMD will need to do something truly radical to unseat Intel's current market dominance with their Core 2 architecture. "Execution" has been AMD's real weak spot lately, and as a result, many who were once loyal to the AMD brand have jumped ship for Intel on the basis of superior performance from comparable products (number of cores, clock frequency, heat generation, etc).
One thought I've had recently, to the end of radically new CPU architectures: A new company, LucidLogix Technologies (http://www.lucidlogix.com), recently debuted a new technology called Hydra for integrating multiple GPUs together without relying on any special software modifications to allow for this parallelism. Explained simply, you wouldn't need Crossfire or SLI in order to make use of as many GPUs as your heart desired and your budget had room for. Lucid's product is the Hydra Engine, which essentially acts as a 'master' GPU which farms rendering tasks out to individual GPUs installed in the systems and then collates the results. All the software 'sees' is the single Hydra Engine in the place of a single GPU, so there's no need to re-develop the software for graphics parallelism.
In AMD's architecture lineup, you may have wondered about the absence of the 'K9' designation. Did AMD simply skip an architectural designation? Not at all. The ill-fated K9 project was a CPU architecture built on the assumption that the consumer software industry would rapidly adopt the type of multi-threaded processing that makes quad- and even octal-core systems practical today. However, no such massive shift to parallel processing has occurred, outside of a few specific games and other software already developed with high-performance computing in mind. Many current programs still only run between one and eight simultaneous threads. Thus the K9 project was scrapped -- CPUs based on the K9 architecture did not perform competitively when the number of concurrent threads was low.
If you haven't yet figured out where I'm going with all this, think of how a current PC is set up to handle the massively parallel task of 3D rendering -- a CPU with between one and four execution cores uses an outboard graphics processor with many cores (in the current iteration of AMD Radeon GPUs, up to 800 simultaneous tasks) to handle just the tasks related to rendering a 3D image. With NVIDIA's CUDA and AMD's Stream SDK, we're beginning to see the performance potential when developers write their code to take advantage of the GPU's massive parallelism for processing other data, most notably with NVIDIA's adoption of PhysX processing on GeForce 8-series and GTX 200 series GPUs. The GPU itself essentially becomes a powerful accelerator for tasks that benefit from massively-parallel processing - a General-Purpose-upon-Graphics Processing Unit.
What is needed, I believe, is a new approach to CPU design that learns lessons from both the GPGPU movement and Lucid's "Hydra" technology. On such as CPU as I've envisioned, four fast serial execution cores are provided for handling serial processing tasks, as well as a large number of slower-clocked cores for handling parallel processing tasks, and another processor (similar to Lucid's Hydra Engine) which determines if processing tasks can be translated to parallel tasks and then collated into a single result. Essentially, it would be a CPU with an inboard parallel-processing accelerator -- like a CPU with an Ageia PhysX PPU on-die. Yet this CPU would also have the ability to translate some serial processing tasks into parallel tasks (such as SIMD instructions) and feed them into the large bank of slower cores for parallel tasks. This would reduce the need for single-threaded programs to be rewritten from the ground up for multithreading in order to realize the gains of parallel processing. The CPU would then have two parts, operating at different clock speeds, in the same way that a GPU's stream processors can operate at a different clock speed from the rest of the GPU. I mentioned K9 earlier to point out that highly parallel CPU processing is somewhere that AMD's engineering staff has been before.
Of course, there may be some technical flaws in the CPU architecture I've just described, but it's an example of how radical ideas could revolutionize the way we think about CPUs in general. And AMD needs to be willing to bet some cash on these ideas in order to out-innovate Intel and leave them scrambling. While Intel figures out how to accomplish the task of 3D graphics on a smaller number of cores optimized for serial instructions (Nehalem), AMD should be focused on building GPGPU functionality into their upcoming Fusion architecture, and then placing the control of that added parallel-processing functionality squarely in the hands of the other CPU cores, rather than relying on the software itself to make GPGPU calls. Bam! Intel is sent back to the drawing board to come up with a similar massively-parallel processor architecture, essentially from scratch, to compete with Nvidia Tesla and FireStream GPGPUs.
The point I'm trying to make in this roundabout way is that AMD's new CEO needs to be willing to identify these potential ways to gain a competitive edge, and then change course rapidly on projects that are already in the works in order to transform seemingly unrelated efforts (integrating a GPU on-die, vis a vis Fusion) into instant competition-killers (integrating a GPGPU with direct hardware control on-die) when the opportunity presents itself. AMD needs a CEO who will respond swiftly, and who will inspire the rest of the company to respond with him, to take the lumbering Intel giant down at the knees.
Comments? Flames? Should I quit my day job and do this kind of industry analysis for a living? Hate for Hector? I invite all comers to respond.
Addendum: In an ironic juxtaposition of corporate fortunes, Matt's other news article for the day highlighted a 25% increase in Intel profits. This just further underscores the need for a radical shake-up at AMD.
With all of that said, I, as well as many others want to see AMD truly step up in the consumer CPU industry. While AMD probably won't leave us any time soon, with Intel all but dominating the space, the migration back over to Intel CPUs will leave the company in a hard position to compete with. -- Matt Serrano
The CEO of chip maker Advanced Micro Devices Inc. is stepping down.
Hector Ruiz had been just the second person to lead AMD after company founder Jerry Sanders. He'll be replaced by the chip maker's No. 2 executive, Dirk Meyer.
My Take: For quite some time, now, AMD's former CEO Hector Ruiz has drawn much ire from many tech industry analysts, who blame his unresponsiveness for many of the firm's current troubles (which at present have primarily to do with their Athlon- and Phenom-family x86 CPU products.) Now that he's stepping down, AMD's board of directors would do well to replace him with someone who can tap into the company's engineering prowess to produce CPU products that have broader market appeal and push the envelope of innovation. AMD will need to do something truly radical to unseat Intel's current market dominance with their Core 2 architecture. "Execution" has been AMD's real weak spot lately, and as a result, many who were once loyal to the AMD brand have jumped ship for Intel on the basis of superior performance from comparable products (number of cores, clock frequency, heat generation, etc).
One thought I've had recently, to the end of radically new CPU architectures: A new company, LucidLogix Technologies (http://www.lucidlogix.com), recently debuted a new technology called Hydra for integrating multiple GPUs together without relying on any special software modifications to allow for this parallelism. Explained simply, you wouldn't need Crossfire or SLI in order to make use of as many GPUs as your heart desired and your budget had room for. Lucid's product is the Hydra Engine, which essentially acts as a 'master' GPU which farms rendering tasks out to individual GPUs installed in the systems and then collates the results. All the software 'sees' is the single Hydra Engine in the place of a single GPU, so there's no need to re-develop the software for graphics parallelism.
In AMD's architecture lineup, you may have wondered about the absence of the 'K9' designation. Did AMD simply skip an architectural designation? Not at all. The ill-fated K9 project was a CPU architecture built on the assumption that the consumer software industry would rapidly adopt the type of multi-threaded processing that makes quad- and even octal-core systems practical today. However, no such massive shift to parallel processing has occurred, outside of a few specific games and other software already developed with high-performance computing in mind. Many current programs still only run between one and eight simultaneous threads. Thus the K9 project was scrapped -- CPUs based on the K9 architecture did not perform competitively when the number of concurrent threads was low.
If you haven't yet figured out where I'm going with all this, think of how a current PC is set up to handle the massively parallel task of 3D rendering -- a CPU with between one and four execution cores uses an outboard graphics processor with many cores (in the current iteration of AMD Radeon GPUs, up to 800 simultaneous tasks) to handle just the tasks related to rendering a 3D image. With NVIDIA's CUDA and AMD's Stream SDK, we're beginning to see the performance potential when developers write their code to take advantage of the GPU's massive parallelism for processing other data, most notably with NVIDIA's adoption of PhysX processing on GeForce 8-series and GTX 200 series GPUs. The GPU itself essentially becomes a powerful accelerator for tasks that benefit from massively-parallel processing - a General-Purpose-upon-Graphics Processing Unit.
What is needed, I believe, is a new approach to CPU design that learns lessons from both the GPGPU movement and Lucid's "Hydra" technology. On such as CPU as I've envisioned, four fast serial execution cores are provided for handling serial processing tasks, as well as a large number of slower-clocked cores for handling parallel processing tasks, and another processor (similar to Lucid's Hydra Engine) which determines if processing tasks can be translated to parallel tasks and then collated into a single result. Essentially, it would be a CPU with an inboard parallel-processing accelerator -- like a CPU with an Ageia PhysX PPU on-die. Yet this CPU would also have the ability to translate some serial processing tasks into parallel tasks (such as SIMD instructions) and feed them into the large bank of slower cores for parallel tasks. This would reduce the need for single-threaded programs to be rewritten from the ground up for multithreading in order to realize the gains of parallel processing. The CPU would then have two parts, operating at different clock speeds, in the same way that a GPU's stream processors can operate at a different clock speed from the rest of the GPU. I mentioned K9 earlier to point out that highly parallel CPU processing is somewhere that AMD's engineering staff has been before.
Of course, there may be some technical flaws in the CPU architecture I've just described, but it's an example of how radical ideas could revolutionize the way we think about CPUs in general. And AMD needs to be willing to bet some cash on these ideas in order to out-innovate Intel and leave them scrambling. While Intel figures out how to accomplish the task of 3D graphics on a smaller number of cores optimized for serial instructions (Nehalem), AMD should be focused on building GPGPU functionality into their upcoming Fusion architecture, and then placing the control of that added parallel-processing functionality squarely in the hands of the other CPU cores, rather than relying on the software itself to make GPGPU calls. Bam! Intel is sent back to the drawing board to come up with a similar massively-parallel processor architecture, essentially from scratch, to compete with Nvidia Tesla and FireStream GPGPUs.
The point I'm trying to make in this roundabout way is that AMD's new CEO needs to be willing to identify these potential ways to gain a competitive edge, and then change course rapidly on projects that are already in the works in order to transform seemingly unrelated efforts (integrating a GPU on-die, vis a vis Fusion) into instant competition-killers (integrating a GPGPU with direct hardware control on-die) when the opportunity presents itself. AMD needs a CEO who will respond swiftly, and who will inspire the rest of the company to respond with him, to take the lumbering Intel giant down at the knees.
Comments? Flames? Should I quit my day job and do this kind of industry analysis for a living? Hate for Hector? I invite all comers to respond.
Addendum: In an ironic juxtaposition of corporate fortunes, Matt's other news article for the day highlighted a 25% increase in Intel profits. This just further underscores the need for a radical shake-up at AMD.
Last edited:
