Sun dropping support for the ROCK processor?

I just came across a link from OSNews about Sun dropping development on the ROCK processor. At first I was a bit surprised by this, given the ROCK feature set Sun had been touting for several years. But in practice, this decision doesn’t surprise me for a few reasons:

1. By the time the ROCK was feature complete and ready for market, Intel and AMD should be able to match its performance when you took cost into account.

2. The cost of the ROCK processor would be significantly more than the Intel or AMD counterparts, especially when you factor in things like system redundancy.

3. While the ROCK has some killer technology (e.g., transactional memory), software isn’t quite ready to take advantage of these features.

4. Few applications can truly take advantage of large numbers of hardware execution threads (and using virtualization to run multiple apps on these hosts has its own drawbacks).

5. More and more people are migrating away from SPARC, so I’m not entirely sure a market would exist for the product when it eventually shipped.

While this is definitely a bummer for the SPARC followers, this may actually be a good thing for Sun. I would love to see them focus more on their X64 product line, and use their top notch hardware designers to get more bang for the buck out of their X64 product line. At the end of the day things always boil down to the cost/performance ratio, and given recent advancements in X64 hardware, it’s extremely hard to justify spending an order of magnitude more for SPARC gear when you can get X64 gear at a fraction of the price. I would love to hear what others think about this topic.

3 thoughts on “Sun dropping support for the ROCK processor?”

  1. I disagree that most people are moving away from SPARC chips. x86 chips make sense for web servers but I would choose SPARC for Oracle database servers. I could not imagine taking my current database off a 15k to put it on some x86 box.

  2. “I’m not entirely sure a market would exist for the product when it eventually shipped.”

    Current CMT (T1, T2, T2+) processors (that are growing continuously in sales) are highly threaded (i.e. 32, 64, 128, 256 threads.)

    These processors have octal crypto engines built in, to accelerate SSL traffic – a single processor keeps up with 3-4 x86 processors in web-tier traffic with quad processor boxes needing 12-16 processors per chassis to keep up (hard to find, very expensive.)

    Growth in this market has been reasonable!

    Significant applications are highly-threaded today (Apache, Tomcat, Oracle, Remedy, Middleware, Serach Engines, etc.) These UltraSPARC CMT processors are very well suited for highly utilized web-tier applications and application servers – which is basically the entire internet.

    “Few applications can truly take advantage of large numbers of hardware execution threads (and using virtualization to run multiple apps on these hosts has its own drawbacks).”

    UltraSPARC RocK processor has far fewer threads (32) where the user community does not have to program them (the 16 cores has 2 threads apiece, one of which is a scouting thread, whose purpose it so bring instructions & data into the cache to eliminate cache misses.

    The secondary thread (per RocK core) is not used by the coders directly – rather it is a “helping” thread which makes a single thread of execution run faster.

    In other words, UltraSPARC RocK is not aimed at highly threaded applications that the UltraSPARC CMT (T1, T2, T2+) processors are aimed at – rather the UltraSPARC RK processors are aimed at workloads which are not highly threaded and workloads which are stupid (i.e. non-thread optimized applications!)

    If the UltraSPARC RocK chip is canceled, it may be an acknowledgment that there is no lightly-threaded market left to shoot for.

    With Intel & AMD both releasing new processors targeting 8 & 12 threads) and increasing their thread counts every 6 months – it may be hard for anyone to imagine that a chip which runs fewer threads exceptionally fast is really needed.

    I disagree with this assessment, I am an advocate of choice in the marketplace. I believe the consumer should decide whether there is a need, the same way Sun had provided early choice with the most highly threaded UltraSPARC CMT processors in the market (T1 running 32 threads) and started the entire industry down this road.

    I can see databases be the major benefactor of a chip like RocK – in order to keep per-socket and per-core license costs down. With Oracle purchasing Sun, Oracle would have an incentive not to overprice SPARC, so perhaps this is yet another reason to possibly cancel RocK.

    We will have to wait to find out the details, since RocK cancelation is still speculation at this point.

Leave a Reply

Your email address will not be published. Required fields are marked *