Jonathan Schwartz's Blog: Rock Arrived

« Neptune Away! | Main | What Brand Means. »

Tuesday Apr 10, 2007

Rock Arrived

We just got the first silicon for our Rock systems - you have no idea how cool it is to have this arrive on my desk in a plastic sandwich bag (don't worry, with packaging like that, they don't send me the ones we put in saleable systems):

That's the front, this is the back:

The chips are running billions of instructions already (not quite at Hello, World, but not far away). Being a professor of the obvious, my first question was - how many pins are on the back? (Answer: nearly 50 on a side, or 2395 pins). Care to guess how many are used for signal (or functionality) vs. power/ground?

Answer... 812 are used for 'signal' (or real work), and 1514 are used for power/ground (the remainder are unconnected). Another interesting stat... for those fond of in-memory databases, Rock will allow us to build coherent systems supporting 256 terabytes in a single software domain (plain vanilla OpenSolaris, no less). That is an awful lot of RAM in a single system (and given the cost of memory nowadays, you'd want to post an armed guard next to that machine).

Rock is 16 cores - we haven't said how many threads per core. Nor have we said why this chip heralds the golden age of effortless parallel programming, or how it brings fault tolerance to the masses. But stay tuned, I think we're planning on talking up both in the next few weeks.

The good news about minting - I didn't say fabricating, we leave that to others - but the good news about minting your own silicon is you get to tell the transistors what to do.

Posted on 10:31AM Apr 10, 2007 | Comments[58]

Comments:

Congrats, Jonathan. That's a system that an In-Memory Data Grid like GigaSpaces can do a lot with...

Posted by Geva Perry on April 10, 2007 at 11:23 AM PDT #

Great work, Keep up the good research / work....

Posted by A Hameed on April 10, 2007 at 12:04 PM PDT #

My recollection is that this step (getting real silicon) would take about six months. This is great news!

Posted by Bob McKillop on April 10, 2007 at 12:06 PM PDT #

wooo - I'll take a load ta :) The T1 is great - can't wait to get hold of a rock. Really really neat work on the HW (don't forget the software side though please!)

Posted by Darren Chapman on April 10, 2007 at 12:24 PM PDT #

I'm old enough to remember when a megabyte of memory (core) cost $1 million ... now you have to get up to terabytes before you run into real money.

Posted by George Potts on April 10, 2007 at 12:40 PM PDT #

Thanks for the picture. The Solaris SA's at BCSSI are ready for the latest and greatest from SUN!

Posted by Rodney Feltner on April 10, 2007 at 12:55 PM PDT #

This is fantastic news! Congratulations to the engineering team. I'm sure the team is putting their baby through the paces (like all babies it'll spit all over you at times ;-)) Just the fact that it is running instructions is great.

Posted by Anantha on April 10, 2007 at 01:07 PM PDT #

Rock or Advanced Product Line? If we purchase a Advanced Product Line server from Fujitsu will Rock work in it, or will it be a Sun only offering like your current Chips? If not, then what is the point of having selling Advanced Product Line servers? If so then I guess we will look at the APL servers, if not, then I guess we better not purchase any servers from either Fujitsu or Sun until Rock gets shaken out.

Posted by Walter Anderson on April 10, 2007 at 01:28 PM PDT #

Well done Sun, isn't this ahead of schedule? With competitors roadmaps slipping, this is very good news indeed for Sun, don't get complacent though, keep fireing on all cylinders.

Posted by Peter on April 10, 2007 at 01:37 PM PDT #

This is great news. The Solaris System Administrators here at BCSSI are ready for the latest and greatest from SUN !

Posted by Rodney Feltner on April 10, 2007 at 01:41 PM PDT #

Jonathan, "I would estimate ROCK will be one of SUN Microelectronics offerings going forward." Advantage SUN!

Posted by William R. Walling on April 10, 2007 at 02:21 PM PDT #

When does it ship in a desktop? #6-) And how big is the chip?

Posted by john on April 10, 2007 at 02:58 PM PDT #

Yes, but shouldn't you give the third sample to John Lithgow? (3rd Rock from the Sun) ttfn mm mark.o.michael@boeing.com

Posted by Mark Michael on April 10, 2007 at 03:58 PM PDT #

Sorry but a picture of a chip usually means the die, not the package. Anyway as a chip guy I collect all the available die pics I can.

Posted by John on April 10, 2007 at 05:01 PM PDT #

Nothing like rubbing the nose of those Wall St analysts that suggested cutting research a few years back. Congratulations to all the great people working for this company and turning it around. Sun is coming back to the top where it belongs.

Posted by Marcelo on April 10, 2007 at 06:08 PM PDT #

That is the Future. Great progress for Network & Computer world. Congraturation !

Posted by Kyoungsu Lim on April 10, 2007 at 06:15 PM PDT #

This is great stuff... it'll be really great to see a low end UltraSPARC CPU that is both a desktop performer as well as power (green) friendly running the latest and greatest Solaris operating system on a $499 configurable system tailored for the everyday desktop. I think both SPARC and Solaris need a beautiful introduction to the masses of computing.

Posted by JuggerNaut on April 10, 2007 at 06:49 PM PDT #

You're wrong, Jonathan. I have an idea how cool it is. Congratulations to David Yen (although he has moved on to the Storage Group since Rock's beginning), John Fowler, Marc Tremblay, and the many others who worked so hard (and undoubtedly so smartly, which is just as important) to reach the tapeout milestone in January. The arrival of Rock silicon is very cool. "Cool" is the word. I for one have been eagerly awaiting certain benchmark results from Sun's next data facing processor, wondering if Solaris on SPARC will again challenge in the scientific computing market, driven by Sun's chips. Or will Solaris/SPARC in that market be left to Fujitsu? On another Rock topic, David Yen said four years ago that Rock would "provide hardware features to do Java acceleration". I hope that's in there. The Feldstein SPARC-V9 Simulator is one Java application that hopes to benefit from such technology.

Posted by Alan M. Feldstein on April 10, 2007 at 07:06 PM PDT #

Any truth to the rumors we're getting Dwayne Johnson to help promote this new CPU? :)

Posted by Dan McDonald on April 10, 2007 at 07:11 PM PDT #

"1514 are used for power/ground" I see your engineers have not found a really good clock distribution method yet either :) "supporting 256 terabytes" I have heard claims like this before, but alas, it does make me wonder time and time again... if I can address it and write to it (all) with "enough" speed, do I need a hard disk? Really? As for ram being expensive, eh, how much is that CPU there Jon? :O

Posted by number9 on April 10, 2007 at 07:34 PM PDT #

I'm sure in a few months after launch there will be another version and another and another and so on and so on and so on...

Posted by Keith L. Dick on April 10, 2007 at 09:17 PM PDT #

We will, we will, ROCK you... Couldn't resist - Congratulations on a fine achievement!

Posted by Bill R. on April 10, 2007 at 10:48 PM PDT #

Excellent job! Excellent Jonathan! I like what is Sun doing! Now is time to acquire 512 core processor technology and Atomic Chip corporation. These companies must be under Sun Microelectronics. http://atomchip.com/index.html http://www.channelregister.co.uk/2006/11/06/japan_512-core_co-pro/ Any comments Jonathan? Thanks.

Posted by Girts on April 10, 2007 at 11:08 PM PDT #

Regarding the speed wonder, how fast is each core works?

Posted by Richard on April 10, 2007 at 11:50 PM PDT #

impressive!

Posted by andrea Mannori on April 11, 2007 at 01:36 AM PDT #

Looks like after the Gigahertz war it's time for Gigacore/thread war. What about power consumption? Are you still green?

Posted by Zsolt Horvath on April 11, 2007 at 02:22 AM PDT #

"32 threads per chip", says WP.

Posted by jengelh on April 11, 2007 at 02:39 AM PDT #

So good, but what a price of this wonderful thing?

Posted by heil0 on April 11, 2007 at 04:50 AM PDT #

Nice to see it can support large memory. x86 and now x64 is fairly limited by 4 DIMMs in Sun shipped hardware (when 8 DIMMs are possible per Opteron). This will finally make it worthwhile to run everything in RAM. Be a boon to databases, web servers, OLTP, Supercomputing. A significant portion could be flash memory to store the OS and logs, and maybe the RAM could be adapted from RamSan, and fully backed by battery. No need for hard drive at all.

Posted by Amit Kulkarni on April 11, 2007 at 06:15 AM PDT #

16 cores and 256 TB of RAM, that sounds like the minimum system requirements for MS Vista Home Edition for install. If you want to run IE and Solitare you'd need to up the RAM I think.

Posted by Charlie on April 11, 2007 at 06:28 AM PDT #

Great. Chief Blogging Officer spins propeller on head while SUNW languishes below its pre-Zander level.

Posted by Rob Winlow on April 11, 2007 at 08:33 AM PDT #

Very cool Jonathan. Please pass my personal congratulations to Marc Trembley and the Rock team. Even though I'm no longer with Microprocessor Report, I've been a fan of the work Sun has done with Throughput Computing. This has been one processor I been very intrigued with. I want to see if those "hardware scout", "execute ahead", "transactional memory", and "space-time computing" patents made it into the final silicon.

Posted by Kevin Krewell on April 11, 2007 at 09:54 AM PDT #

GigaSpaces is an intel invested company ;) it would be nice to see this rock there

Posted by Shaban on April 11, 2007 at 10:36 AM PDT #

True, the stock is below pre-Zander level, but it's ever-rising... with great technologies such as this leading the way. Check Sun's charts. They're showing a steady, healthy return to ongoing profitability. :)

Posted by Maybe so on April 11, 2007 at 10:44 AM PDT #

Rock is a boon to Parallel / Grid Computing.

Posted by Lakshman on April 11, 2007 at 10:49 AM PDT #

GOOD GOD!!! I'm in SUN heaven now

Posted by David on April 11, 2007 at 10:58 AM PDT #

[Trackback] Jonathan Schwartz, CEO von Sun Microsystems hat heute in einem Blogposting die ersten Fotos und Hinweise auf den nächsten Sun Prozessor "Rock" gegeben. Gerüchte gibts ja schon lange. Rock soll mehrere Threads in mehreren Kernen (deutlich mehr als Niaga

Posted by www.widhalm.or.at on April 11, 2007 at 11:30 AM PDT #

Dear John , With such a message don't you think that Solaris users could be fed up and confused. What about APL and SPARC64 VI ? What about your technical agreement with Fujistu ? Who's gonna build Rock ? Anyway don't forget that AIX/Power5 is the enemy...

Posted by Thomas on April 11, 2007 at 11:37 AM PDT #

Always bemusing...The Professor of the obvious is just what we need in a high tech world. Keeping breakthroughs in simple terms makes them spread around the world through the small talk that runs our industries. Great job on your candid appeal for an outstanding development!

Posted by CeeRay on April 11, 2007 at 01:40 PM PDT #

Congratulations, Jonathan. I seem to remember you didn't want me to think of Sun as focusing on HW, but this is some mighty fine HW to focus on. I look forward to seeing the details of packaging, product features, and price. I'm especially interested in the idea of running BIG applications in all that memory.

Posted by Amy Wohl on April 11, 2007 at 02:11 PM PDT #

Jon, are you running a dog and pony show blogging or a dollar and cent blogging for us shareholders? Are you talking to users or shareholders? Are you gonna make money on your newfangled chip? If not, say so .. I have to decide wheter to buy Sun shares or not. Are you more in tune with your employees than with your Joe 6-pack shareholders? How much is your chip selling for? What is your fab capacity? Just one fab?

Posted by Gumbo on April 11, 2007 at 03:18 PM PDT #

Time for a reality check. You say Rock allows you to build systems supporting up to 256 TB of memory. Let's assume that in 2009 there will be memory DIMMs with up to 8 GB capacity. So it would require 32,768 DIMMs to build a system with 256 TB. I'll be impressed if you produce a system that has 512 DIMMs and 4 TB. So what's the 256 TB good for except bragging rights over something that won't be built? And you claim it heralds the age of effortless parallel programming? That's not likely to happen anytime soon. So you might just as easily argue that the first AMD dual core chip heralded the age of effortless parallel programming, too. Seems like Sun is getting pretty desperate that you have to put out such fluff.

Posted by RL on April 11, 2007 at 07:26 PM PDT #

Customers & Field team are really looking forward to seeing a real product based on UltraSparc RK, I think it would be awful cool ! Considering customer will face APL and Rock Based systems , Will they be crag-fast ?? Buy or just wait ? We need to think it over.

Posted by Zhi-Wei Willis Yao on April 11, 2007 at 08:10 PM PDT #

"SUN is desparate" , "SUN is not relevant" , "SUN will go down" , i started hearing this since mid - 2000 . Quite boring . Let us see where this Ship goes under the ROCK solid leadership of Jonathan.

Posted by Prince on April 11, 2007 at 10:54 PM PDT #

"And you claim it heralds the age of effortless parallel programming? That's not likely to happen anytime soon. So you might just as easily argue that the first AMD dual core chip heralded the age of effortless parallel programming, too. Seems like Sun is getting pretty desperate that you have to put out such fluff." "the first AMD dual core chip heralded the age of effortless parallel programming" Well, it did/does, didn't it...... ergo all the rest is most likely to happen/be true, too.

Posted by amanfromMars on April 12, 2007 at 02:01 AM PDT #

Cool, I think i would market this as Superdome on a chip ;-)

Posted by Vinnie on April 12, 2007 at 04:48 AM PDT #

This is what you get with steady investment in R&D, patience to outlast pundits and critics, and a company culture built on the premise that "innovation matters"

Posted by Adam on April 12, 2007 at 11:00 AM PDT #

Well done Jonathan! About time SUN offers something with great aplomb, away from competition.

Posted by Mariolini Von Heino on April 12, 2007 at 11:22 AM PDT #

Congratulations, Jonathan & SUN!

Posted by Sergei Plaxienko on April 12, 2007 at 06:10 PM PDT #

Cool, with this I can host google in one machine

Posted by Pepe on April 12, 2007 at 07:34 PM PDT #

It's great to see Sun carrying on in it's tradition of R&D and innovation. One Competitor just seems to wait and see what comes from Sun before making a move, and puts all its effors into designing yet another ink cartridge.

Posted by VC on April 13, 2007 at 01:47 AM PDT #

So Rock "heralds the golden age of effortless parallel programming." Just a tad overstated, Jonathan? What is under-stated here is that Sun needs new and more energetic Sparc partners beyond Fujitsu to stoke the fires. Putting David Yen back on the case is a great move, but this is a steep mountain to climb in an age of architectural consolidation. Good luck, David! --Rick Merritt, EE Times

Posted by Rick Merritt on April 13, 2007 at 10:32 AM PDT #

It's a cool posting ... very cool, but I somehow thought ROCK would be even more powerful than it apparently is. After all, you say ... "The chips are running billions of instructions already (not quite at Hello, World, but not far away)" So I'm thinking .. it never used to take billions of instructions to execute the C program that printed out "Hello World". Perhaps the next order of business should be getting the compiler suite optimized for the new chip. :-)

Posted by nvrijn on April 13, 2007 at 05:05 PM PDT #

I must be a real geek, because this gave me goose bumps…. Of course my wife looked at me like I was an alien when I went to the other room and told here. My first Sun Certification class was back in the early 90’s and I have been a loyal follower of the Sparc chip ever since—Go SUN.

Posted by M Nassal on April 13, 2007 at 05:42 PM PDT #

Congrats to the founders of high performance MP/MT/MC processor design, David Yen, Marc Tremblay, and the many others who worked so hard and kept the long view during the assaults from the financial community (reinforced by the competition selling against Sun). Unlike many other global companies headquartered in the US, Sun has continued to invest in R&D and add value to Sun's products rather than to "sell out" to the Wall Street influences demanding unsustainable growth and annual returns. If any one large company could meet Wall Street expectations for 5 years running, they would soon surpass the GNP of the US (or many other major countries) and eventually surpass the GNP of the planet. Thanks, Sun Microsystems, for keeping a rational course during these somewhat irrational financial times. Sustainable growth is nearly always more important than temporary financial vehicles created by creative investment firms which can sometimes destroy or trade long-term prospects for a short-term "cash out" or "buy out" profit scheme. Thanks to Sun's example we have a solid business example of how staying the course can increase the long-term value of a company. We live in incurious times, and Sun gives us reason to pause and reflect on their sound and responsible business management.

Posted by Ron on April 14, 2007 at 07:28 AM PDT #

Great. I guess. What does all that mean in terms of practical application for all of us?

Posted by Chas Ferris on April 14, 2007 at 04:32 PM PDT #

Nice teaser. It does make me wonder if the number of power and ground pins facilitate just powering up and down sections of the chip or do they also herald the emergence of asynchronous processing in the SPARC processor. I am thinking that virtualization in the processor could support effortless parallel programming. I still hope for cascading circuits that initiate cycles triggered by the output of the preceding circuit instead of the system clock (like a line of dominos that push each other over). There is also the possibility of cascading a single thread through all 16 cores to achieve a single thread processing speed approaching 16 X the system clock ( maybe 2 X 16 = 32 GHz ). I do hope to get one of these in a workstation some day. My Ultra 80 is actually more than I need but I just can’t help dreaming about such things. { what a huge geek I am :) }

Posted by Lee Hepler on April 14, 2007 at 07:34 PM PDT #

thanks for informations.. very nice...

Posted by evden eve nakliyat on April 15, 2007 at 05:09 AM PDT #