Sun Enters the Commodity Silicon Business

We’re
announcing the fastest microprocessor
we’ve ever shipped this week – delivering 89.6 Ghz of parallel computing power on a single chip – running standard Java applications and open source OS’s. Simultaneously, we’ve said we’re entering the commodity marketplace, and opening the chip up to our competition. It’s probably worth explaining.


I’ve made the following statement before, but it bears repeating – what do banks, oil companies, and telecom operators all have in common?


They represent some of the largest industries on earth, yield some of the wealthiest companies, and are all oriented around monetizing… commodities. Commodity markets are the largest and most valuable on earth (despite those who seem to think they have no value).


Commodities are defined by goods or services in near perpetual, universal demand. Like financial services, oil, and network access (and increasingly, internet search and water, both vital ingredients for life as we know it in Silicon Valley).


In my view, computing is also a commodity – as is storage and networking. They’re all in demand globally – and the market’s growing (especially in the developing world).


Now, companies that do well in commodity markets don’t do so exclusively on price. They differentiate via R&D. In fact, the industries I mention above are among the world’s largest investors in R&D – oil companies, banks, telcos and network service companies all have mammoth technology budgets. As does the technology industry (at Sun, we invest nearly $2,000,000,000 (billion) a year on R&D).


Why do we invest? To differentiate, of course.


So when we announce (via this webcast) the fastest microprocessor the industry’s ever seen (the benchmarks are staggering) – and say we’re entering the “commodity microprocessor market,” what does it really mean? It means we’re no longer limiting ourselves to serving an internal market, inside Sun. Instead, we’re opening ourselves up to the broadest market possible – where the opportunity’s largest.


Despite having what’s arguably the single biggest competitive advantage our systems business has ever had, we’ve separated out our microelectronics business – and told them to win on the open market, as well. We signed our first OEM agreement with Marvell, through which we’ll be collaborating to take our networking advancements to the marketplace. More broadly, our microelectronics team is free to sell to our competition. For the record, we’d be thrilled to supply, for example, a Niagara blade to HP or IBM for their blade servers – as well as to the larger market of networking, storage, automotive and industrial applications. It’s a commodity market, after all – vast, growing, and in search of differentiation.


You’ll recall we followed this path with our software business – decoupling Solaris from its exclusive focus on Sun hardware. That experience validated the obvious: the market for Sun’s innovation is always larger outside of Sun, than inside. When we opened ourselves to the market, our business grew faster (Software grew 13%, year over year, faster than Sun overall). Now we’re following that path with our microelectronics business.


To add fuel to the fire, the blueprints for our UltraSPARC T2 (I personally like the moniker, “Niagara 2” – named after Niagara Falls, btw, and the great volumes of water that pass over them), the core design files and test suites, will be available to the open source community, via its most popular license: the GPL. Making Niagara 2 the only commodity silicon whose core designs are available to the open source community – whose strength, and market power, only grows by the day.


These are all huge changes to our business. Driven by a simple philosophy: the open market is bigger than any internal one.
But ultimately, why now? A simple reason: because customers building infrastructure for the internet have been asking us to do so. And they’ve all basically said the same thing – in building their own internet infrastructure, from telecom equipment to consumer devices, they want commodity economics without commodity performance.


Stay tuned for more details.


In the interim, on Monday a reporter accidentally violated our news embargo, which set off a flurry of press coverage. I have to admit my favorite part of the day was reading a quote from a few of our (hopefully erstwhile) competitors – who referred to the internet as a “niche.” Hmm…

48 Comments

Filed under General

48 responses to “Sun Enters the Commodity Silicon Business

  1. This new processor sounds promising. It is very brave of sun to releases technical information under an open source license. Congratulations.

  2. Thanks a lot for your post. For a Open Source Analyst it is very interesting times indeed. I look forward to the combination Niagara 2 and your Project Indiana that is said to see it’s light in late 2008 – then you have hit the REAL commodity market. Thanks for all transparency!

  3. Congratulations to Jonathan and the Niagara 2 Team on yet another milestone in the history of microprocessors! Actually, Niagara 2 is a microsystem: Just add RAM and you’re all set!

  4. I appreciate your comments on how to make money from commodity markets. Always a concern for those in any computer-related field as innovation and automation drive down prices. Excited to hear about 10Gb Ethernet technology entering the main stream as well.

  5. Tobias

    Great, I liked the T2000 systems already. Hope to get a hand on a Niagara II system via Try&Buy.
    One question remains open: Who’s the leader to ask about the future direction of Niagara? Will there be a T3? Where will it take us?

  6. Abdul Hameed

    I beleive SUN has taken the right step towards the right direction but bit late, never mind it has happend…Open Market, Open Software and Now Open Microchip…..now SUN is setting the IT trend…
    Congratulation Jonathan and all folks @ SUN Microsystems…Keep up the good work…….

  7. The comments by HP are too funny. The T1 was a great first step and now the T2 make significant improvements on these industry leading and shifting designs. How they can say the T1 & T2 are proprietary CPUs and target only a small niche and still keep a straight face is incredible. Given the GPLing of the designs, binary compatibility with previous Sparc chips it’s amazing the only mainstream user of the "great selling" Itanium chips even has the gall to make false statements like they did.

  8. Larry Chen

    Correct me if I’m wrong but this photo appears to have been taken from the NY side. I Love NY too!!

  9. howlingmadhowie

    this chip is going to revolutionize the market place for server chips. i expect to see changes on netcraft.com over the next few years. i can hardly wait to see it as an option when compiling a linux kernel.

  10. This new chip could easily slay the IBM Bluegene supercomputer ( http://www.research.ibm.com/bluegene/ ) on every single measurable point on the graph; in terms of throughput, in terms of up front costs and then in day to day operating costs when one considers power usage and cooling. More importantly this is not some specialized hybrid monster but a standardized production run where anyone can order it. I look forward to tossing Radiance from Lawrence Berkeley National Labortory on that machine and taking the top performance spot. Again. Solaris 10 on Opteron already holds the fastest benchmarked results with Radiance ( a massive floating point load I may point out http://www.blastwave.org/articles/BLS-0059/index.html ) in both the single chip and the multi-core results. I fully expect that the Niagara T2 will repeat that winning position. Congratulations and I look forward to the numerical throughput!

  11. Gumby

    Jon
    You better hurry and buy out AMD and its three newfangled Fabs! You need all the Fabs you can take your hands on…I am happy to know that you are finally breaking out of Apple sauce mold!

  12. Interested Party

    Jonathan, did you and Linus ever have dinner? If so, what did you serve him?

  13. Jon, read the WSJ article – congratulations! we’d love to get our hands on the technology to test with it. Cheers, Jason

  14. Michael McClary

    Two (related) questions:
    1) Is it GPL2 or GPL3?
    2) Can anyone just take the RTL, do the backend work, and manufacture this chip and/or derivatives of it with no further licensing and no more patent problems than Sun itself, or are additional licenses required to actually build such silicon?

  15. Sounds impressive. However, to be absolutely clear:
    Can a third party manufacture a T2 with the open documentation? Is one allowed to without making a spesific deal with Sun, wrt. Sun’s possible patents? (I’m not quite sure how the GPL2 implicit patent license language should apply to manufacture of depicter hardware.)
    If not, it’s not all it’s cracked up to be, though educational certainly. If yes, let’s hope it revives the SPARC architecture a bit; it needs the boost, and having a _truly_ open modern CPU architecture in that way would certainly be a good thing for the free world.
    Thanks.

  16. Mr. Little

    The 89.6GHz number you state is completely wrong in so many ways. Even if you go with a fake "GHz" rating based on the number of threads running at a certain frequency, you are still wrong because each core in the T2 processor *cannot* simultaneously execute an instruction from all 8 threads in the same cycle. The true fake GHz number is closer to 8*1.4 = 11.2 GHz.

  17. Eric

    If it doesn’t run x86 and x64 compatible instruction sets…how is it a commodity? I know Sun lives in its blue skies world where Microsoft doesn’t exist, but I wouldn’t say its much of a commodity until you get x86 and x64 support…is there a fast emulation environment that will run on top of this platform to support that perhaps?

  18. Prince

    The biggest joke from a competitor was this chip serves a "niche" market called internet / web serving. How pathetic.
    On another note we have to see how much T2 can catchup as a commodity chip when compared to AMD and Intel chips. That is a formidable challenge, i think.

  19. Rock on Sun!
    This is a truly fantastic move on your part, and defiantly step in the right direction. You clearly have a solid understanding of the rationale behind why a business would open source their core IP. The most fundamental being an understanding of how it contributes to, and can grow the bottom line.
    Congratulations and thank you for your leadership in this space.

  20. Fantastic news!
    Now I have something to announce!
    < The OpenSolaris Magazine. > My own project I am working on over at http://moonshine.opn4.org/
    Keep a eye out for things to come..
    Regards,
    Edward.

  21. Raghu

    I feel there should have been some better support for memory controller by including it on the chip 65nm should have helped that.
    this chip will not do better than 4 dual core option, which has eight threads.
    all though there are 64 threads, with out optimizing data transfer path between CPU and memory, getting better results than AMD is difficult.

  22. Anon

    Jonathan,
    PLEASE drop the "89.6 Ghz" FUD — Niagara is a great processor as it is. But all of us in the processor community wince when we hear the frequency multiplied by the number of threads to give the total Ghz. It "mega, giga, tera hurts" to hear something which we labour over for years and years trivialized like this. Soon the marketers will take over and we engineers will stop trying to achieve higher real frequencies if you make it meaningless.

  23. aggelis aggelis

    Well i couldn’t expect anything less from you Jonathan. I hope others in SUN are clever enough to share your vision. (otherwise no SUN stocks for me🙂 )
    I will give an example. Pythagoras theorem was tought until the
    19th century (i think, IIRC,but the point is valid) in University classes.Now its "Commodity Knowledge" tought in highschools.
    Same story for TCP/IP networking. Nowdays almost anyone can
    configure a DSL router.
    "Hardware/Software" industry has already reached the Commodity point
    of its history.
    So where is the "money"?
    It’s in the mass market in a broader sense. Not in the elitism.
    You cannot be an elitist selling cars (well you can, but Ferrari is owned by FIAT, Lamborgini by Volkswagen and that is an irony and the truth).
    ——————————
    My English is better than most people’s German.

  24. Regarding the memory bus… while you can rightly say that the per-thread bandwidth is relatively low, the per-core bandwidth is still quite high. These chips offer 50 GiB/s, which is 6.25 GiB/s per core! Yes, that equates to "only" 781 MiB/s per thread, which is at comparable to what Intel is pushing on its dual 4-core systems.
    Personally, I think that 64-threads at 781 MiB/s per-thread is much better than Intel’s 8 cores at 750MiB/s per single-threaded core! This is actually quite decent in comparison, and actually sound like they can handle some real work.
    My only complaint is that they don’t run x86 code, which kills this option for me… plus, unless we see this truly become "commodity" and see some commodity pricing, it will be out of my price range as well.
    (btw, I apologize if any of my math is wrong…)

  25. Kebabbert

    Studies of Intel shows that a typical x86 server idles ca 60% under FULL LOAD, because of cache misses. It waits for data 60% of the time. A 3GHz Intel CPU does work 40% of the time, hence it equals 1.2GHz CPU with no cache misses. Therefore Intel tries to decrease the cache misses, by using larger cache, by using complicated prefetch cache logic, etc. But never the less, there WILL be cache misses – no matter what you do unless the CPU can see into the future which data will be needed next.
    This ancient problem is solved by Sun’s new family T1 and T2 CPUs. This family has no elaborate cache logic at all, instead the CPUs doesnt try to avoid the cache miss problem at all. When a miss occurs, the CPU switches to another thread in ONE clock cykel (something an ordinary CPU can not do). As soon a miss occurs, the CPU continues work on another thread. Therefore this family of CPUs dont idle 60% of the time, its more like less than 10%, even 5%. They work 95% of time with no waiting. If you do this at 1.5GHz, it would be equivalent to a x86 CPU of roughly 3.8GHz.

  26. Mark

    Mr. Little,
    As I understand it, a Niagara 2 can execute two threads per core simulaneously. Each core has two integer execution units, each supporting four threads. See the preso on OpenSPARC.net (http://www.opensparc.net/pubs/preszo/06/HotChips06_09_ppt_master.pdf). See pages 8, 10, and 13. So that would make it 22.4 GHz.
    The truth is, GHz is irrelevent as a singular measure. It is like speaking of the RPM of an internal combustion engine, without reference to number of cylinders, displacement, or added performance features (superchargers, etc.).
    Eric,
    You are missing Jonathan’s point. Computing is a commodity, computers are not. Niagara will run many of the same apps as x86 processors (or PowerPC processors, for that matter). Oracle, Apache, IBM WebSphere, BEA WebLogic, SAP, MySQL, and all of those cool Web 2.0 scripting languages.
    Raghu,
    The Niagara 2 processor has four memory controllers on chip. See the above mentioned preso.
    And while a Niagara 2 may not do as well as four, dual-core Opterons on some apps, it may do better than such a system on other apps such as serving encrypted web pages. Niagara 2 is more than just 8 cores and 64 threads. It is a system on a chip with two 10 Gigabit Ethernet controllers, a PCIe bus, and on-chip crypto processing. And don’t forget, if you want a system with four, dual-core Opterons, Sun can sell you that, too.

  27. Long Island Linux User

    Wow this if very promising and exciting news. Finally my old UltraSparc workstation may have company besides the Dual G5 Mac and AMD boxes littering my office space. I might even be willing to watch my stock shares in Sun for a change.

  28. William R. Walling

    Jonathan,
    "Kudos to ALL at SUN Microelectronics division for their continuing ‘open’ based efforts within computing hardware."

  29. It sure looks like a great chip!
    Though calling it a 89.6 GHz chip is like calling a 10 coach train running at 100 km/hr a 1000 km/hr train.🙂

  30. Jonathan, do you will send come CPUs to several reviews sites? i want to see how the niagara2 performs on a Anandtech, XBitLabs or ARSTechnica Test😉 i want to see some independient benchmarks soon, if 64 threads are a solution or a problem (memory controller, etc.)
    open your core blueprints in GPL means that AMD or Intel can use part of your design? of course they shall open their own blueprints in GPL too, but that never gonna happen😉

  31. Rock on!
    I’m glad to see that 64 bit RISC is still making progress, and that Sun is very committed to open source. I hope that you will also open the specs for other components around the hardware, so that open source projects such as openbsd can port to them.

  32. Dear Jonathan Schwartz,
    When Sun wanted to widen the reach of Solaris it embraced X86, and Solaris, which was on the exlusive territory of Sparc found its way to other platforms. One reason why Sun did this was because Sparc wasn’t reaching wide enough.
    T2 appears very advanced and the price looks attractive too. What is more, the strategy to open source the hardware would cause wider developments revolving around the Sparc architecture
    Has the wheel has turned a full circle? Is the reverse going to happen? Solaris moved out to X86 yesterday, and would the Operating Systems FROM x86 converge onto Sparc this day? Beginning with Network products?? Longhorn on Sparc, perhaps?
    Could happen. And, T2 still appears to be the beginning. T2 would scale up and could at the same time scale DOWN for lower end computer products. With production outsourced to TI, perhaps even a Sparc processor for the phone with as many threads…
    What puzzles me about Sun is where SUNW still is. Why is the market slow to react? It is prolonging its unfair judgement of SUNW just for the simple reason that the stocks slided once a long time ago. The investing world still continues with with its characteristic refusal to look no further nor deeper. Extended opportunity for the wise for good fortune !

  33. How does all this make communications better, faster, richer? To me, that’s what it’s all about. If Sun can’t relate to communications, what it develops, then it fails at pushing The Network IS the ComputerTM. If you’ve noticed lately, the head of Mexican telecom has recently passed Billy G as the world’s richest man. Surely, with the Internet, communications is no fluke. Tell us about communications, Jonathan, that’s what IT’s about.

  34. Bob

    "delivering 89.6 Ghz of parallel computing power"..What a crock of &^%. Reminds me when our 3-4 year old Pentium workstations were blowing the doors off brand new UltraSPARC machines that cost 4-5 times as much and Sun kept denying the SPARC was slow. They’d make up some kind of B.S> like this 89.6 Ghz nonsense. It may have a lot of threads, but it’s a SPARC at heart and is therefore dog slow. As soon as IBM and Intel/AMD catches up on the technology they will far surpass in speed , this will be relegated to SPARC zealots. Sun simply has no idea how to build a CPU to be fast. Sun would be better off pushing Solaris on AMD/Intel has hard as possible and building more of it’s own x86/x64 systems that appeal to a broader audience.

  35. Great news.
    I miss the pizza-box workstations (I have a small collection of them) and truly would love seeing (and buying) a new SPARC-based workstation that had the same distinctive non-PC looks of the old purple-feet Sun boxes.
    I know… The desktop computer market is overly crowded and has obscenely low margins, but I think Apple demonstrates a honest OS running on great looking computers can command slightly higher prices.
    And a sexy computer with the Sun logo on it is a great PR.

  36. Steven Arnold

    I am proud of Sun for its commitment to open-source and the community in general. Sun has done many things over the years to support our community and to become an integral part of it. Thank you to Sun for their efforts!

  37. Sebastien Stormacq

    Bob,
    These days, computer speed is not measured in raw performance for a single task, but for the capacity to handle many tasks at once.
    Today performance is all about parallelism !
    When you’re browsing the web, as an end user, you will not notice the difference if your page is delivered in 200ms or 300ms.
    However, on the server side, if you can only serve a single user at a time or 16 users at the time, the throughput is not exactly the same.
    A quick math will explain
    16 x 200ms single thread = 3200 ms
    16 x 300ms on 16 threads = 300 ms
    OK, this is over simplified, but you get the idea.

  38. Anonymous

    "As soon as IBM and Intel/AMD catches up on the technology they will far surpass in speed"
    Er right, that’s kind of the point. They haven’t yet, and by the time they have, Sun will have come out with something faster again. That’s how competition works.

  39. Didn’t Sun already release the T1 as GPL? And nobody built that one either. So what’s the point? Looks like T1 is available on opensparc.net. I would love to see someone start producing this but economies of scale seem to doom us all to x86 forever.😦

  40. Peter

    Dawn of a new computing era, makes it hard to justify unreliable PC crap now doesn’t it. Oh can you make something like Thumper with a Niagara 2, then when Solaris 11 comes out with ZFS boot, the computing world will be on its knees.

  41. Bob

    Sebastien..Niagara IS a niche CPU. It will be useless for the many, many single threads apps out there and certainly more than useless for scientific computing.
    194.125.39.94..Sun has never been able to keep with IBM and Intel. UltraSPARC always lagged way behind..and Sun will quickly be passed with this technology as well…and the only ones buying them will be folks who still think their Blade 100 is a fast workstation…

  42. e

    Niagara 2 is true art.

  43. UltraSPARC Reality Check Guy

    Jonathan,
    There’s a really good reason the techies (can’t remember how exactly you referred to them in the presentation) cringe when you say "89.6 GHz". I’m no computer architecture expert (aspiring perhaps), but it’s fairly obvious after a little investigation that it’s only 11.2 GHz, not 89.6 GHz. Here’s why. Each core is executing up to 8 threads (ideally) with its 1.4 GHz which prevents pipeline bubbles due to common memory latency conditions. The result then is that each thread is only getting its fair share (every 8th cycle), which is only 175 MHz. That is the number that I’ll let you multiply by 64. The other number is just plain wrong.
    That’s ok though, 11.2 GHz is nothing to be ashamed of. Granted, you can get a lot more throughput out of those 11.2 GHz than your competition can, but that’s because each thread is only actually doing 175 MHz "worth" of work!! It’s easy to deal with slow memory when each thread is only accomplishing as much work as it did roughly 10 years ago.

  44. Sandvika Dude

    Hey, IBM already has its power.org for its silicon and blade.org for its systems. So whats new ? Also, whats the tpm-c per core of this chip ? Does it beat the best of the best ? I think, Sun should go back to where its strongest………Java and give up on Silicon.

  45. Winlow

    Jonathan, it’s all puffery until you show some revenue growth from this and any number of other innovations that you claim are so good.

  46. Bob Weiler

    Jon,
    The T2 guys deserve kudos for what looks like another terrific engineering effort. The problem is that by the time these chips turn into actual servers, the prices somehow come out 50 to 100% higher than Sun’s own AMD based systems with roughly similar performance. You can see the effect now with the far less capable T1 systems. T2 systems will sell like hotcakes (or coolcakes, if you prefer) but not if the buyer has to pay a 50% SPARC premium. They’ll just buy the Opteron or Xeon box and instead of making a profit on the sale of both a chip and a box, Sun will just get the low margin box sale. Or worse yet, prospective customers will buy somebody else’s commodity x86_64 box, and Sun will get nothing. My current employer could really use T2 (and even better, Victoria Falls) boxes, but they just aren’t going to pay 50% more no matter how much power they save. Bean counters being what they are a Cap-ex dollar in hand is worth 2 in the bush. Getting SPARC back on the performance curve solves 1/2 of the problem, what is it going to take to get SPARC boxes back on right side of the price curve?

  47. Dan

    Will I be able to upgrade my existing T1 processors with the new T2 processor? Perhaps something like processor in a box.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s