Slashdot Mirror

We also checked out both options for our lab at the university. At that time (3 years back), a PS3 was EUR 600. The only way to get a "cell computer" was via IBM blades.

That's odd. Mercury Computer has had a "cell accelerator board" for $8K since the last quarter of 2006. Basically its a cell processor in a PCIe slot.
Second generation is here: http://www.mc.com/products/boards/accelerator_board2.aspx

Maybe they had export problems with it, although they announced it at a singapore trade show.

Mercury had a PCI-e cell expansion card for over a year now.

Unlike the leadtek one, the mercury version has the full version of the cell processor, with 8SPEs. Dont think it comes with any prebuilt codecs though.

Mercury Computers has had a card with a cell on it for quite some time. It is, I believe, very expensive (~$10k?).
Link to the card.

http://www.mc.com/products/productcategory.aspx?id=246&Processor=246

Mercury Computer Systems will sell you systems/boards with Cell processors even if IBM won't.

So far the only way to get a Cell processor is PS3.

Not true

Check out Mercury systems. These aren't exactly for home use, but they are shipping Cell + Linux computer systems. However, given the economics, it might just be better to go with a PS3, if you can live with the memory limitations. There may be other companies, but checking at TerraSoft (one vendor for PowerPC Linux software and hardware) takes you right back to IBM and Mercury hardware.

Its very interesting reading the paper linked to the link http://www.umiacs.umd.edu/users/vishkin/XMT/spaa07 paper.pdf. It reminds me of Mercury Computing Programming Toolkit for Cell Processor Programming. They too have a spawn and join method of concurrent programming see: http://www.mc.com/uploadedImages/MCF-FOE-model.jpg at http://www.mc.com/microsites/cell/ProductDetails.a spx?id=2824. Notice the worker/manager similarity to the spawn/join semantic. It would appear that this chip is fundamentally the same, but provides implicit engine allocation. Very interesting....

Its very interesting reading the paper linked to the link http://www.umiacs.umd.edu/users/vishkin/XMT/spaa07 paper.pdf. It reminds me of Mercury Computing Programming Toolkit for Cell Processor Programming. They too have a spawn and join method of concurrent programming see: http://www.mc.com/uploadedImages/MCF-FOE-model.jpg at http://www.mc.com/microsites/cell/ProductDetails.a spx?id=2824. Notice the worker/manager similarity to the spawn/join semantic. It would appear that this chip is fundamentally the same, but provides implicit engine allocation. Very interesting....

I was at TGS and a) it was about 1000 degress inside; b) the PS3s were in tiny plexiglass containers with no visible ventilation

All the following links are from www.mc.com (Mercury Computer Systems)
either their cell evaluation system, which is designed in the IBM bladecenter form factor:
http://www.mc.com/products/view/index.cfm?id=27&ty pe=systems
the turismo gangable computers: http://www.mc.com/products/view/index.cfm?id=34&ty pe=systems
http://www.mc.com/products/view/index.cfm?id=96&ty pe=boards mercury computing has been offering/releasing Cell based products for a couple months. Their boxes are just as affordable for "consumers" as an IBM blade system. I don't work for them, but mercury computing has some pretty awesome stuff.
-Scott

All the following links are from www.mc.com (Mercury Computer Systems)
either their cell evaluation system, which is designed in the IBM bladecenter form factor:
http://www.mc.com/products/view/index.cfm?id=27&ty pe=systems
the turismo gangable computers: http://www.mc.com/products/view/index.cfm?id=34&ty pe=systems
http://www.mc.com/products/view/index.cfm?id=96&ty pe=boards mercury computing has been offering/releasing Cell based products for a couple months. Their boxes are just as affordable for "consumers" as an IBM blade system. I don't work for them, but mercury computing has some pretty awesome stuff.
-Scott

All the following links are from www.mc.com (Mercury Computer Systems)
either their cell evaluation system, which is designed in the IBM bladecenter form factor:
http://www.mc.com/products/view/index.cfm?id=27&ty pe=systems
the turismo gangable computers: http://www.mc.com/products/view/index.cfm?id=34&ty pe=systems
http://www.mc.com/products/view/index.cfm?id=96&ty pe=boards mercury computing has been offering/releasing Cell based products for a couple months. Their boxes are just as affordable for "consumers" as an IBM blade system. I don't work for them, but mercury computing has some pretty awesome stuff.
-Scott

Yes, IBM did pretty much all the work, the STIDC is in Austin, on an IBM site, if I remember correctly. However, in terms of business purposes, very little of IBM had any plans for the Cell. They, Sony and Toshiba put in $400 Mill to develop the chip but IBM didn't have a roadmap for further development until the past year or so. From what I've seen, the STI guys were heads-down Cell development, and weren't paying too much attention to other potential applications until the thing finally went to fab. At that point, people at IBM started looking around and saying: "Hey, this is actually a pretty sweet chip. Maybe we can sell this to somebody." Mercury ( http://www.mc.com/cell/ ) was on board for actually making a product before IBM got around to slapping them into a blade-center chassis. Just because the initial announcements didn't say PS3 doesn't mean that it wasn't for the PS3 that it was originally developed. Most companies play things like that a bit close to the vest until a certain level of risk has been abated.

<sarcasm>But you are right, I "don't know anything about the Cell". The 2 blades (and 2 Cells per blade) that I've got at work are just collecting dust, and we never actually TALK to people at IBM.</sarcasm>

http://www.mc.com/cell/products/view/index.cfm?id= 96&type=Boards

Thats about the closest your going to get at this point.

Here you go. Enjoy.

http://www.mc.com/cell/

Actually, this is probably being used today by people who buy one of Mercury's Linux based Cell blades.

Don't tell Mercury that!

If you like real action with the IBM Cell processor in the next-gen war, why not try this baby:
http://www.mc.com/powerblock200/
It knocks the stuffing out of any Sony PS3.

Check this bad boy out: http://www.mc.com/products/view/index.cfm?id=64&ty pe=boards

Its a Dual Cell BE Processor blade, and if you wanna count the SPE's as processors this would give you 9x2x14 = 252 cpu's in a blade center enclosure. :)

By the end of Q1 2006 (or thereabouts), we expect to see shipments of Mercury Computer Systems' Dual Cell-Based Blades; Toshiba's comprehensive Cell Reference Set development platform; and of course the Sony PlayStation 3.

Well, to be fair, there aren't that many PowerPC based PCs. In the server and embedded market there are quite a few PPC based machines, though. In this market, you can definitely find PPC machines with HT. For instance, Momentum boards have HT.

why? because apple engineered and designed the silicon, had the boards fabed for them

Apple designed an ASIC for their own use, but they could have also used IBM chips.

ibm has shifted over to being a video game console cpu maker

Yes, they do make the Cell, Xenon, and Broadway processors, but saying that they have shifted their core business is a bit disingenuous. One of the things IBM does is make custom chipsets for whoever is willing to pay for it. Furthermore, IBM makes a lot of chips for many different applications, and both Apple and the video game market are a small fraction of their output. I'll give you another example: Hitachi also makes custom CPUs for video game consoles, but no one is claiming that Hitachi's core business is video game consoles.

they hardly care about apple's business, and as such aren't focusing on the design work apple needs done to stay in the game. intel was willing to make a sweet deal, and amd would have had the same production volume issues that ibm was having. since intel has the fab capacity

Ars Technica has a good article on pretty much exactly what you said.

1) design a low-power-consumption high-performance PowerPC chip that would be ideal for Apple to use
2) keep the development so secret that spouses are kept in the dark
3) launch the product after Apple has already abandoned PowerPC
4) Ignore Apple because they are irrelevant. Instead, sell stuff to the many companies who consume more PPC chips than Apple ever could now or in the forseeable future.
5) PROFIT!

This is great news for customers like the US Navy who rely on Linux-on-PowerPC for important tasks like sonar imaging systems.

IBM just teamed up with a company called Mercury to build Cell-based computers for (military) applications:

As a result, demanding applications such as radar, sonar, MRI, digital X-Ray, and many others can be taken to new levels of sophistication and performance.

And as we all know, the Cell is basically a Power processor.

I'm not sure how Sony comes into play, but IBM made a deal with a company called Mercury for them to use the Cell. According to their press release, they will be using it to build stuff in "medical imaging, industrial inspection, aerospace and defense, seismic processing, and telecommunications".

comes from building hardware for a specific task. Unfortunately most of you can't access this little bit of nerd heaven but some incredibly cool hardware architectures are being described at the High Performance Embedded Computing conference. Sky and Mercury have some of their hottest new designs here. How about a machine that can do a 256 mega-sample FFT in real time?, or a self configuring supercomputer on a chip? Of course most of these tricks will never escape the lab except for the speed-ups for rendering engines...one place where gamers and the DOD are driving technology in a dead heat race with lots of winners. Besides, in a few months, something will come along that will go even faster than blue gene.

I wrote some demo fibre channel block-device drivers for vxWorks a few years ago, and I found that VxWorks's FAT FS was buggy. The bug only affected one flavor (I don't remember if it was FAT12, 16, or 32), but it was clearly reproducible and clearly an OS fault. It was a corner case and we found some way around it (like avoiding an writes with a length of 1MB).

Here's the usual rant you see here on slashdot, and it's true: since it was closed source, we couldn't verify that we'd caught all the bad cases, and we couldn't submit the fix to back to WindRiver.

Bill Joy could join Mercury, Venus, Earth, Mars, Jupiter, Saturn, Neptune or Pluto.

(Oddly enough, there isn't a company that bears the name of the missing planet you're thinking of ...)

If you were the general in charge of weapons procurement for the Department of Defense and you needed to select a high-end server to process radar information tracking enemy missiles targetted at the USA, then which server would you select?

Either way, the entire chip surrounding the Altivec engine is 32-bit, so this would be an improvement on that.

check out Mercury Computer Systems

one of the co-designers of RapidIO, they make ppc shared memory multicomputers which will use RapidIO in the future.

think 320 PPC nodes, 256MB RAM each, the whole thing in one 9u 19" rack mount chassis, running on standard power with no special cooling requirements.

It all depends on what part of the "agency" you're talking about. Certainly, there's a lot of push to move to MS on desktop-type systems.

But mission systems are a different matter. To start with, you're talking in some cases about systems that have been deployed for decades. What my company (and we're not alone) has been doing for the last 7 years is migrating these custom OS/HW systems to COTS platforms. In the sonar arena (think the sonar workstations in The Hunt for Red October), we have progressively moved the signal processing systems from custom systems to embedded (VxWorks and Mercury) to Solaris and SGI prototypes to Linux on Intel. Currently deploying systems are using Linux on Compaq Proliant 8500 8-way boxes. The next refresh will be to multiple dual-CPU P4 Xeon boxes communicating over Gigabit Copper Ethernet. Expect to see Itanium-based units in a couple of years. All of this is saving the Navy a lot of money while dramatically improving the capabilities of the fleet.

Do a quick Google search for Acoustic Rapid COTS Insertion.

And it's not stopping there. Plenty of other onboard systems, both in the surface and undersea communities, are moving from outdated one-of-a-kind systems to commodity hardware encapsulated in survivable enclosures.

my company has several g4 systems an order of magnitude more powerful (320 CPUs) both in house for testing and in customers hands. and they only eat up 9u in a 19" rack.

and as far as ppc architecture clusters go, just about any IBM RS-6000-SP2 system is more powerful, and probably takes up less space too.

Diskless nodes are much better from the point of view of maintenance and upgrading software on the system. Do it once, on one node (the "master" node with the disk), and all nodes are updated.

Caveat: Using the network for temporary storage hurts performance and doesn't scale if you have a lot of temporary data.

High Performance Computing Hardware

If your parallel process has a lot of communication, you would be using Myrinet, and/or getting a CSPI cluster running Linux, because ethernet is too slow. (You would avoid Mercury and RACEway because they are avoiding Linux.)

Appropriate Parallel Applications:

In short, this system would be good for embarassingly parallel tasks like a brute force search through key space for breaking encryption, or encoding mp3s. In such tasks, each processor only has to communicate at the beginning and ending of the task, and no processor has to communicate with any other processor during the task.

Inappropriate Parallel Applications

This system would not be so good for STAP in RADAR or SONAR.

Kenneth J. Hendrickson

US energy has a relatively low cost of about $1/watt-year. Non-US energy costs are generally higher. For many energy efficiency is not a significant individual cost. Yet, consider the following...

CASE 1. Embedded High Performance Computing.

Sustainable PERFORMANCE/WATT can be a strong processor selection criteria for *embedded* processor selection including embedded high performance computing. Hence, vendors in the niche embedded high performance market tend to use PowerPC and DSP chips, not Alpha, Pentium or SPARC for dense compute solutions.

Vendor examples: CSPI, Mercury, Sky

Did folks notice that Black Lab Linux (see optional software) is working with MPI Software Technologies? Look close, MPI Software Technologies is actively supporting *embedded* MPI & MPI-RT.

CASE 2. Not Too Large Clusters (50 or so processors)

Also, notice the HPC Wire news article posted on the Top 500 site where...

"Several problems were caused by the power consumption and the heat in that small laboratory. The Paderborn people needed the fire brigade to pump cold air into the room. The power consumption was about 10 - 12 KWatt. Switching on only the power supplies of the nodes, the electric fuses switched off."

CASE 3. Blue Mountain...

6,144 processors -- 10,000 KW always online; 2,600 KW average usage.

SIXTEEN A/C Units plus FOUR 750-ton chillers plus TWO 10MW power stations (major UPS feature)

CASE 4. Aggregate World Energy Consumption (Green perspective)

Now consider 100,000,000 processors on desktops (far less than one per world capita)

100,000,000 * 30 watts-year * US$1/watt-year = US$3 Billion each year (without consideration for cooling costs, higher average world energy prices, CPU fan MTBF costs & such)

Hey, what's US$30 / year per CPU for a typical US home wallet??? Well maybe not much to many reading this... however, the energy costs can add up. Embedded system designers, large scale system installations, and environmentally sensitive people know what I'm talking about.

-- Think Global, Act Local --