s for an up and down process, this means the Cell can perform 6,400,000,000 single point operations per second. But ... that is the instruction set handing off information. Internally, when processing, the Cell runs at 25.6 gigaflops or 25,600,000,000,000 floating point operations per second per SPE. Link that to 8 SPE's and you have one fast processor when doing simple math functions.

IBM has released a prototype blade server running with two Cell processors that ran at 200 GFLOPS and is preparing to release a blade that runs at 400 GFLOPS.

In proportion, the world's largest supercomputers run at between 70 and 250 TFLOPS or ten thousand times faster than a single Cell. However, one Cell blade server running at 200 GFLOPS is

Back in 2001, a group of engineers from Sony went to IBM with an idea. At the same time, Toshiba was looking into a similar concept and was approached by IBM to join into a development consortium. Over the next four years, with a budget approaching US $400 million, the Cell Broadband Engine Architecture was developed at the joint STI Design Center.

This unique processor uses a PowerPC core, the heart of most of IBM's processors but adds to this 8 coprocessing or synergestic processing elements. In other words, a central processor that runs 8 separate sub-processors to spread the work and enable faster throughput. IBM used technologies derived from it's PowerPC4 architecture to streamline and interface the Cell architecture and make it all happen.

What resulted was a single chip that ran at a clock speed of 3.2 Ghz and could perform 9 fully independent functions at once with cross-linking so the processor could in itself shift and put more power and ability where needed as needed.

The Heart of the Cell

At it's heart, the Cell contains two primary processing cores, the PPE or PowerPC Engine (Power Processing element), and the Synergestic Processing Elements or SPE's. The SPE's are the sub-cores than run off instructions given by the PPE. Although they are fully capable of running by themselves, they will not perform unless told to by the PPE, and it load shares among all of the SPE's to ensure the best use of the processor at any given time.

The Cell is a 32 bit RISC processor that runs at an internal 64 bits and can operate within the SPE's at 128 bits. This enables the Cell to process very large amounts of data in a scalar environment that can adapt and handle varying amounts of information. With the SPE's scaling and factoring to ensure the data is processed and handled properly, no matter where it comes from.

The Cell is optimized for single precision floating point operation, which means that it does one cross-check during computation and rounds the number up to get the final answer. The Cell can be run at double-precision but at a significant speed penalty.

Another unique element of the Cell is it's Memory Coherence Architecture. In a single processor, this is not an issue as the processor does each instruction in a linear fashion, one after the other. The Cell can perform 9 simultaneous instructions per up or down clock cycle and as such, without a central control, each of the SPE's could theoretically process the same information and duplicate each other. The PPE provides that control, along with IBM's patented Memory Coherence circuitry. The main factor in using the Cell efficiently is software code, as the Cell is driven by integrated code, not by hardware commands.

The Racehorse

The power of the Cell is in some ways staggering ... The central processor runs at 3.2 ghz clock speed or 3,200,000,000 clock cycles per second. Since each clock cycle allows for an up and down process, this means the Cell can perform 6,400,000,000 single point operations per second. But ... that is the instruction set handing off information. Internally, when processing, the Cell runs at 25.6 gigaflops or 25,600,000,000,000 floating point operations per second per SPE. Link that to 8 SPE's and you have one fast processor when doing simple math functions.

IBM has released a prototype blade server running with two Cell processors that ran at 200 GFLOPS and is preparing to release a blade that runs at 400 GFLOPS.

In proportion, the world's largest supercomputers run at between 70 and 250 TFLOPS or ten thousand times faster than a single Cell. However, one Cell blade server running at 200 GFLOPS is

it all happen.

What resulted was a single chip that ran at a clock speed of 3.2 Ghz and could perform 9 fully independent functions at once with cross-linking so the processor could in itself shift and put more power and ability where needed as needed.

The Heart of the Cell

At it's heart, the Cell contains two primary processing cores, the PPE or PowerPC Engine (Power Processing element), and the Synergestic Processing Elements or SPE's. The SPE's are the sub-cores than run off instructions given by the PPE. Although they are fully capable of running by themselves, they will not perform unless told to by the PPE, and it load shares among all of the SPE's to ensure the best use of the processor at any given time.

The Cell is a 32 bit RISC processor that runs at an internal 64 bits and can operate within the SPE's at 128 bits. This enables the Cell to process very large amounts of data in a scalar environment that can adapt and handle varying amounts of information. With the SPE's scaling and factoring to ensure the data is processed and handled properly, no matter where it comes from.

The Cell is optimized for single precision floating point operation, which means that it does one cross-check during computation and rounds the number up to get the final answer. The Cell can be run at double-precision but at a significant speed penalty.

Another unique element of the Cell is it's Memory Coherence Architecture. In a single processor, this is not an issue as the processor does each instruction in a linear fashion, one after the other. The Cell can perform 9 simultaneous instructions per up or down clock cycle and as such, without a central control, each of the SPE's could theoretically process the same information and duplicate each other. The PPE provides that control, along with IBM's patented Memory Coherence circuitry. The main factor in using the Cell efficiently is software code, as the Cell is driven by integrated code, not by hardware commands.

The Racehorse

The power of the Cell is in some ways staggering ... The central processor runs at 3.2 ghz clock speed or 3,200,000,000 clock cycles per second. Since each clock cycle allows for an up and down process, this means the Cell can perform 6,400,000,000 single point operations per second. But ... that is the instruction set handing off information. Internally, when processing, the Cell runs at 25.6 gigaflops or 25,600,000,000,000 floating point operations per second per SPE. Link that to 8 SPE's and you have one fast processor when doing simple math functions.

IBM has released a prototype blade server running with two Cell processors that ran at 200 GFLOPS and is preparing to release a blade that runs at 400 GFLOPS.

In proportion, the world's largest supercomputers run at between 70 and 250 TFLOPS or ten thousand times faster than a single Cell. However, one Cell blade server running at 200 GFLOPS is

time.

The Cell is a 32 bit RISC processor that runs at an internal 64 bits and can operate within the SPE's at 128 bits. This enables the Cell to process very large amounts of data in a scalar environment that can adapt and handle varying amounts of information. With the SPE's scaling and factoring to ensure the data is processed and handled properly, no matter where it comes from.

The Cell is optimized for single precision floating point operation, which means that it does one cross-check during computation and rounds the number up to get the final answer. The Cell can be run at double-precision but at a significant speed penalty.

Another unique element of the Cell is it's Memory Coherence Architecture. In a single processor, this is not an issue as the processor does each instruction in a linear fashion, one after the other. The Cell can perform 9 simultaneous instructions per up or down clock cycle and as such, without a central control, each of the SPE's could theoretically process the same information and duplicate each other. The PPE provides that control, along with IBM's patented Memory Coherence circuitry. The main factor in using the Cell efficiently is software code, as the Cell is driven by integrated code, not by hardware commands.

The Racehorse

The power of the Cell is in some ways staggering ... The central processor runs at 3.2 ghz clock speed or 3,200,000,000 clock cycles per second. Since each clock cycle allows for an up and down process, this means the Cell can perform 6,400,000,000 single point operations per second. But ... that is the instruction set handing off information. Internally, when processing, the Cell runs at 25.6 gigaflops or 25,600,000,000,000 floating point operations per second per SPE. Link that to 8 SPE's and you have one fast processor when doing simple math functions.

IBM has released a prototype blade server running with two Cell processors that ran at 200 GFLOPS and is preparing to release a blade that runs at 400 GFLOPS.

In proportion, the world's largest supercomputers run at between 70 and 250 TFLOPS or ten thousand times faster than a single Cell. However, one Cell blade server running at 200 GFLOPS is

essor, this is not an issue as the processor does each instruction in a linear fashion, one after the other. The Cell can perform 9 simultaneous instructions per up or down clock cycle and as such, without a central control, each of the SPE's could theoretically process the same information and duplicate each other. The PPE provides that control, along with IBM's patented Memory Coherence circuitry. The main factor in using the Cell efficiently is software code, as the Cell is driven by integrated code, not by hardware commands.

The Racehorse

The power of the Cell is in some ways staggering ... The central processor runs at 3.2 ghz clock speed or 3,200,000,000 clock cycles per second. Since each clock cycle allows for an up and down process, this means the Cell can perform 6,400,000,000 single point operations per second. But ... that is the instruction set handing off information. Internally, when processing, the Cell runs at 25.6 gigaflops or 25,600,000,000,000 floating point operations per second per SPE. Link that to 8 SPE's and you have one fast processor when doing simple math functions.

IBM has released a prototype blade server running with two Cell processors that ran at 200 GFLOPS and is preparing to release a blade that runs at 400 GFLOPS.

In proportion, the world's largest supercomputers run at between 70 and 250 TFLOPS or ten thousand times faster than a single Cell. However, one Cell blade server running at 200 GFLOPS is

s for an up and down process, this means the Cell can perform 6,400,000,000 single point operations per second. But ... that is the instruction set handing off information. Internally, when processing, the Cell runs at 25.6 gigaflops or 25,600,000,000,000 floating point operations per second per SPE. Link that to 8 SPE's and you have one fast processor when doing simple math functions.

IBM has released a prototype blade server running with two Cell processors that ran at 200 GFLOPS and is preparing to release a blade that runs at 400 GFLOPS.

In proportion, the world's largest supercomputers run at between 70 and 250 TFLOPS or ten thousand times faster than a single Cell. However, one Cell blade server running at 200 GFLOPS is the equivalent of the Hitachi SR2201 that was used starting in 1996 at the University of Tokyo, Japan. Not bad for a single processor. IBM is also upgrading the world's current fasted supercomputer, Blue Gene/L to run with Cell processors and expects that system to crack the PFLOP range (1,000 TFLOPS or 1,000,000 GFLOPS) before the end of 2008.

Linux and The Cell

IBM has released the development core for the Cell that integrates in to the Linux OS. Sony has released what is called the GNU toolchain. As of Kernel version 2.6.16 Linux officially supports the Cell. Anybody can download the Linux development kit for the Cell from the IBM website along with a simulator to test code. This code can be run on the PS3. But to build games for the PS3's specific hardware, you need Sony's development kit, which is not available unless you are an actual SCEA licensed game developer.

Distributed Computing

A final unique thing about the Cell is it's ability to work with broadband or high-speed ethernet for communication. The Cell is optimised for ethernet and high-speed communications and as such can be used is a Grid computing network or in a Distributed computing environment. This means many Cells running on separate systems can communicate via the Ethernet or Internet, share processor load, and massively process information.

The PS3 has been announced as a platform for the Folding@Home distributed computing software that has become popular with home PC users. The PS3 version is called cure@home and will enable the Folding@home project to exceed a petaflop with only 10,000 PS3's running the software. This will enable the project to process protein folding in an effort to find cures and better understand the processes behind diseases such as Altzheimers, BSE(Mad Cow disease), Cancer, Huntingtons Disease and cystic fibrosis among others.

Overall, the Cell processor is performing as expected and at the specs listed. This is a true revolution in computing a it is the first tim that the power of a large scale supercomputer has ever been brought down to a consumer level and with access to tool that can actually utilize it. We expect big things for the Cell - so keep a sharp eye out - the world of computing is being turned on it's ear.