Slashdot Mirror
IBM's Cell Processor — Not Just for PS3 Anymore
TechFreep writes to tell us that IBM has released a new line of QS20 Blade Servers based on the processor they developed for the Playstation 3. From the article: "Today IBM announced a new line of high-powered QS20 Blade Servers intended for use in seismic research, encryption, digital image rendering and military surveillance applications. Each QS20 will include two nine-cored Cell Processors clocked at 3.2Ghz apiece, which were developed along with Sony and Toshiba for Sony's upcoming Playstation 3 console. As Playstation 3 isn't scheduled for release until November, the QS20 will mark the first application in which the highly-touted Cell will be available to consumers."
Not necessarily about this announcement of course as most of us have seen it coming for some time. However, the concept of the core processor is quite strong and I think that it will play an important role in computing in the future.
h tml
A little tid bit about all that: http://www.blachford.info/computer/Cell/Cell0_v2.
Justin - Don't be afraid of my blog, it won't bite.
Maybe take a peek at the Wiki entry about the cell processor for a good background of what it is capable of.
Neutiquam erro
I'm not a major corporation but I could order a server from Dell and they would ship it to me. That's consumer enough for me.
I think you were just being a dumbass, but in case you weren't... according to this article about the same announcement, IBM is recommending Fedora Core as the operating system to use. So yes, linux does run on it.
Regards,
Steve
Servers aren't consumer products so I don't see how this gets consumers any closer to a Cell or any sooner.
Depends on which definition of consumers you are using. Taken more literally, even large companies are "consumers". I think the intent of the submitters statement was simply that it would be the first shipping product that would house the cpu. After all, if you were willing to pay the cost, you _could_ purchase it. No different than say a Ferrari Enzo, it's a consumer product, but one in which obviously only a small percentage of consumers could actually afford.
it is a server with blades.
That's begging the question. A blade is a modular computer on a board. You team these up into a chassis and have them communicate together and you've got a blade server.
.. paranoid crackpot leftover from the days of Amiga.
(Score:5, Informative) for a simple wikipedia posting is just bullshit people. We all know that if we don't know what something is, google is your friend. If I simple type in CELL, it's the fifth fucking thing listed. ARG!
As Ex Amiga owner I can say 68k has nothing to do with Power architecture. One is CISC (68k) and other is RISC from beginning.
:)
Motorola joined IBM producing RISC CPU instead of upgrading their 68k line.
I can understand your confusion since Apple and Amiga (via extension board) moved to PPC 603 from Motorola 68k.
I am not a programmer but RISC chips are famous for their reduced registers and commands, to run instructions in less cycles.
Better give you pointer written by real programmers:
http://en.wikipedia.org/wiki/68k -->68000 family
http://en.wikipedia.org/wiki/Powerpc --> PowerPC
As ex Amiga person, reading the first line, "32bit from the start" on 68k still makes me mad to Wintel monopoly and Apple jumping... Anyway. What is done is done.
What a load of crap. Please cite your reference where Apple claims 5x against a Quad 2.5 GHz G5. On Apple's Mac Pro pages I see: 1.8x, 1.8x, 1.6x, 1.4x and 1.4x on content creation. On SPECInt_rate_base2000 I see 2.1x and on SPECfp_rate_base2000 is see 1.6x.
Apple-Mac Pro-Intel XeonApple dumped PowerPC because IBM couldn't get to 3.0 GHz and they couldn't get to 2.0 GHz with a low power version for notebooks. It isn't complicated and no conspiracy theory is necessary.
The Potential of Science With the Cell Processor
http://science.slashdot.org/article.pl?sid=06/05/
It reference a second article:
Researchers Analyze HPC Potential of Cell Processor
http://www.hpcwire.com/hpc/671376.html/
This discusses research at Lawrence Berkeley National Laboratory on using the STI Cell processor for scientific computing. From the article quoting the LBL paper:
"Overall results demonstrate the tremendous potential of the Cell architecture for scientific computations in terms of both raw performance and power efficiency,"
and
"We also conclude that Cell's heterogeneous multi-core implementation is inherently better suited to the HPC environment than homogeneous commodity multi-core processors."
The paper went on to say that while the Cell processor was designed for single-precision 32-bit floating performance but with some simple changes to the design it could be optimized for double precision 64-bit floating performance.
This makes a lot of sense if this is the same Cell processor that IBM is using in their blade servers.
Really cheap, really fast 9 core processors!
An interesting read.
RLH
multi-core Cells are IBM's answer to Sun's chip multiprocessor (CMP)
Um. No, sorry, you can't compare Cell and Niagara (let alone Rock). Very different beasts, designed for very different workloads.
Cell has a central core, which uses the PPC instruction set. It then has a number of specialised vector units hanging off the side. In other words: one general purpose core; eight vector units.
Niagara, on the other hand, contains up to eight cores, each of which is designed as a multi-purpose, multi-threaded CPU (up to 32 threads can run simultaneously on an 8 core Niagara chip.)
Cell excels at tasks that don't require much branching, and/or lots of vector math. Scientific apps, for example. Niagara excels at tasks that can easily be divided into multiple simultaneous tasks. There is some overlap between the two, but not as much as you might think. I wouldn't, for example, use Cell for a web server, whereas Niagara would do very nicely for that purpose.
The iSeries is Power5. The PPC970 is a castrated Power4.
A chronology might help:
Power4: Moderately crappy core, great system architecture/SMP fabric
PPC970: Power4 minus the great system architecture/SMP fabric plus shitty high-latency chipset.
Power5: Redesigned Power4 fixing many instances of crappiness, with even better system architecture.
A deep unwavering belief is a sure sign you're missing something...
As the others have noted, its 3.2GHz per processor. It doesnt make the system 6.4GHz, however. As we all know, 1+1 does not equal 2 in the processor world. When you get into multi processor systems, the amount of processing speed depends on a lot of factors, such as the subsystem, software tuning, etc.ac
But Niagara has 8 cores, each capable of runnning 4 threads. Importantly, they all use the same instruction set and have access to external memory. They all have MMUs too.
9 423
http://www.theinquirer.net/default.aspx?article=1
It's an SMP beast.
So you could run a traditional OS on it, and all the cores would be used if needed. On a Cell, there is a single PowerPC core, which is a bit underpowered with no out of order execution for example. The bulk of the processing power is in the SPEs. But these have a different instruction set, no access to external memory and no MMU.
If you're writing a game with procedural textures, you can live with this by partitioning the algorithm amongst the SPEs and using the PowerPC essentially for housekeeping. But it would take a lot of work to partition a general purpose OS like this. So the two machines are not really aimed at the same market.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
Even with a 20% yield the cell is going to cost between US$15 to US30. The cell will be made on a Silicon on Insulator (SOI) wafer (300mm diameter) costing approx US1200. In huge quantities (and the for PS3, Wii and Xbox 360 there will be huge quantities) this will bring down the cost of the wafer and thus the overall cost of the chips on SOI will be cheaper.
On an interesting note the Cell chip is approx 250 square mm so taking the 20% option on a 300mm diameter SOI wafer (assume US$1300 for wafer and assembly) the overall cost of the chip is US$23. The fab machines are about $11M each and are capable of producing completed wafers between 25,000 to 35,000 per year. Great if you control the process and IBM, Toshiba and Sony (to a lesser degree) do.
Do a Google search on "Silicon On Insulator" as a starting point.
There ain't no such thing as proprietary standards only proprietary formats. Standards are by definition open.