Penryn

Hello, this is Dileep Bhandarkar blogging from Beijing, where it is already tomorrow today! There is a lot of buzz here about Penryn. So, what exactly is Penryn?

Penryn is an unincorporated community in Placer County, California, in the United States. Geographic location is 38.852° N 121.169° W.

Penryn granite is noted for its beauty and strength. Mottled in more-or-less equally sized specks of black and white, it appears a medium-to-dark gray in color, at first glance, but takes on an almost bluish-gray hue when viewed in a subdued light or, when wet or polished. This unique stone can be seen in the foundations and walls of a number of California landmarks including The State Capital and the old U.S. Mint in San Francisco.

Source: http://en.wikipedia.org/wiki/Penryn,_California

You could say Penryn rocks!

Penryn is the code name for Intel’s Next Generation Intel® Core™ 2 Family processor microarchitecture. It is also the name of the industry’s first 45 nm microprocessor. It is an improvement on the Core Microarchitecture introduced last year. It delivers more performance at the same clock speed compared to our Core 2 Duo processors. It is a dual core product with a shared 6 MB L2 cache.

Intel’s 45 nm High-K Metal Gate Silicon Technology allows Penryn to run at higher clock frequency than its predecessor. It features 47 new Intel SSE instructions designed to enhance Media, Graphic and Gaming. We have measured >20% perf improvement on existing games compared to today’s fastest dual core processors, and more than 40 percent for gaming and video encoding with Intel SSE4 optimized video encoders. But of course, as with any performance claims “Your Mileage May Vary”.

Penryn-based processors provide faster divide performance with a Fast Radix-16 Divider, roughly doubling the divider speed over previous generations for applications such as scientific computation.

Enhancements to Intel Virtualization Technology speed up virtual machine transition (entry/exit) times by an average of 25 to 75 percent.

Up to 6MB of L2 cache per die (50% larger) enables greater performance across workloads; and increased associativity ( 16 way to 24 way) improves the utilization of the larger cache.

All of this fits in a compact 107 mm2 die featuring 410 million transistors!

The Penryn family enables optimized products for each of the target market segments from the high end Xeon all the way to mobile.

It will enable mobile optimized processors that provide greater performance within the existing mobile form factors and support lower processor idle power with Deep Power Down Technology. It also supports increased performance on single threaded apps with Enhanced Dynamic Acceleration Technology, which uses the power headroom of an idle core to boost performance of the non-idle core.

For the Desktop, there will be two versions that will ship at greater than 3GHz. A second generation Quad-core with up to 12 MB of L2 cache that will fit into existing QC power envelopes, and an energy efficient dual-core version with up to 6MB of L2 cache that will maintain the 65 W envelope.

For the Server and Workstation segment, the Penryn family will extend the leadership that the Xeon Processor family delivers today with dual and quad core products. We expect the combination of new 1600 MHz FSB for HPC segments, >3 GHz clock speed, microarchitecture enhancements and larger caches to deliver up to 45% improvement on bandwidth intensive floating point workloads compared to today’s fastest quad core processors.

Some Java workloads exhibit up to 25 percent increase.

I guess the March of the Penryns has begun!

If you are in Beijing, plan on attending Steve Fischer’s talk on Wednesday titled “Technical Overview of the 45nm Next Generation Intel® Core™ Microarchitecture (Penryn)”.

That’s all for today!

38 Responses to Penryn

  1. Ronny says:

    Hello,
    It’s great that Intel is doing so much to improve our life.
    I would like to know why is the next generation of the mobile processors is dual core and not quad core?
    Does Intel is stops to manufacture the quad core processors?
    If not, will some day mobile pc’s will have a quad core processor? When?
    Thanks,
    Ronny

  2. Steve Barger says:

    Hi There, I grew up in Penryn and I’m curious who decided to give our sleepy little town it’s 15 mins of fame? Any chance of giving the inside scoop on how it got named that way?

  3. rbtbob says:

    Are you playing with using phase-change cells as switches to allow re-configuration of processor architecture on-the-fly?

  4. Dileep Bhandarkar says:

    Thanks for your interest in quad core for notebook compuetrs, Ronny. The current quad core processors are supported only down to 50W which is too high for notebooks today. More work is needed on power management before quad core can be offered in notebooks. Expect to see quad core in notebooks sometime in 2008.

  5. Dileep Bhandarkar says:

    Steve, I am sure that you are not the only one who is curious about why the name Penryn.
    Typically, code names at Intel are based on some geographic location – city, river, mountain etc.
    The design team proposes names that they would like management to approve. One of the development sites for Penryn is in Folsom, CA. Generally, “obscure” or little known places have a higher probability of being approved. Sacramento would not have worked!
    The project started originally in Israel under the name Hagar, but one quick look at the comic section of the paper convinced us that that was a Horrible name!

  6. rbtbob says:

    What about this from the release titled “Introducing 45nm Hi-k Next Generation Intel® Core™ Microarchitecture”
    …Scalable and configurable system interconnects and integrated memory controllers…
    That phrase “configurable system interconnects” sounds like non-volatile switches. (or re-setable fuses)
    ((Don’t worry, I’ll go into lurking mode when the real techies find your blog.))

  7. thalwil says:

    Any news on when Penryn will be available for mobiles? I can’t wait to get my hands on a MacBook Pro with this beast inside!

  8. Dileep Bhandarkar says:

    The Penryn family of products will ship in the second half of this year in all segments – mobile, desktop, and servers.
    Check with Apple on when they will launch it in their products.

  9. JeffG says:

    For Penryn Quad core you’re still doing MCM. Is this just production and timescale related or does it have anything to do the difficulties of the shared cache being shared by more cores? When you do go monolithic quad core can we expect a new cache hierarchy, something like each pair of cores sharing an L2, and all cores sharing a common L3 pool?

  10. Dileep Bhandarkar says:

    Yes, the Penryn Quad Cores are dual die in a package just like the current 65nm quad core products. It seemed like the right tradeoff for our first quad core and since Penryn is a enhanced compaction of the previous core, we continued the same approach. You will have to wait a while to see exactly how we do a monolithic quad core in the future.

  11. JeffG says:

    Right, but are there technological limitations to the architecture that make monolithic Quads unfeasible at this time? (such as poor scaling on shared L2) Or is the decision based purely on other factors like production and time to market?

  12. Dileep Bhandarkar says:

    No, there are no technological limitations. It is a matter of finding the right business case. The architecture for a monolithic design is likely to be different than a dual die design. I am sorry if I sound a bit evasive, but I cannot disclose what a monolithic design might look like until we are ready to disclose such a product at some time in the future!

  13. Tony T says:

    Hello Dileep Bhandarkar, I have a few questions about this new multi-core era, since I can’t find the info myself maybe somebody knows from inside Intel. Who invented the multi-core system? How long has this technology been around? Can you give a brief history of how the multi-core architecture became so popular?

  14. Dileep Bhandarkar says:

    The idea of multicore is a natural progression of the computer industry. As Moore’s Law allows more transistors on a single die, more and more functions migrate on to the processor silicon. Examples from the past are caches and floating point units. Dual core processors started appearing about 5 years ago (I may be off by a year or two) at the high end. IBM and HP introduced dual core processors in their high end RISC products. PC processors tend to use smaller die size since they are more cost driven. In 65 nm technology, our first monolithic dual core processor (code named Yonah) fit in 90 mm2.
    The rest, as they say, is history.

  15. Van says:

    Multicore is a fine idea but how do you increase memory bandwidth to keep up with more cores, or keep processors busy instead of waiting for memory? One idea for the latter was proposed years ago by Rishiyur Nikhil, a colleague of Arvind at MIT. He proposed a “split-phase” memory (and I/O) transaction, plus two queues (active and waiting) of register files, managed by three new instructions: fork (which creates a new register file copy in the queue), join (which decrements a counter and destroys a register file copy if the counter isn’t zero), and send (which sends a transaction to memory or I/O and moves the register file from the active queue to the waiting queue; when the transaction is done the memory or I/O moves the register file from the waiting queue to the active queue). The processor executes the process at the head of the active queue until it is destroyed by a join or until it moves to the inactive queue. The Tera/Cray MTA machine embodied some of these ideas. This seems to me to be a very productive way to use the enormous number of transistors Intel can put on a chip these days.

  16. Igor says:

    Dileep, why not be honest and admit that the speed difference between Core 2 Duo and Penryn in the above mentioned tests comes from carefully selected memory bound tests and 25% higher FSB? If you look carefully, that is exactly the speed difference in many tests. Where the difference is lower, it is obvious that the test is compute bound and it scales with clock speed which is ~7%. As for DivX 6.6, it uses SSE4 so the comparison is not fair.
    Don’t get me wrong, I am not saying that Penryn sucks, I am looking into getting one because I am a developer and I need it, but it is far from the hype Intel is trying to create and the sooner you admit it, less disappointed people are going to be about it when it comes out.

  17. Igor says:

    While you are at multi-core issues, I sincerely hope that you will fix the DTS reading mechanism before Nehalem comes out because the whole idea about MSR register holding DTS value is incredibly stupid.
    On a Quad Core it takes four context switches to read all four sensors. It would be way better if there was some way to set the index of a core you wish to read and read single shared register instead of having to change thread affinity four times for such a simple operation.
    Moreover, Tjmax should have been numeric constant in the same register, not a single bit in completely different and entirely undocumented register.
    Finally, “reading valid” bit is also stupid. Reading should be valid all the time and if it is not valid at the moment it should return previous valid reading from a latch register because this is what you end up doing in software when you get “not valid” response.

  18. rbtbob says:

    Are ya’ll keeping an eye on Cswitch? They had 3 patent applications published today:
    Multi-terminal phase change devices 20070096071
    Content-addressable memory having phase change material devices 20070097740
    Methods for fabricating multi-terminal phase change devices 20070099405
    These were posted by a deep digger by the id of olconnom on the ENER message board.
    If you are not careful these Transmeta guys will cut you off at the pass.

  19. harit124 says:

    Van: What you are suggesting (from MIT work) seems like a variant of Hyperthreading to me. Hyperthreading & HW multithreading, while hiding latency & improving throughput, do not solve the basic problems of memory access latency (aka cache miss penalty) and memory BW scalability for multicore architectures. It would be interesing to see how the available memory BW will scale as the number of CPU cores serviced by specific memory controller increase.
    Hari T

  20. Dileep Bhandarkar says:

    I will no longer be contributing to this blog topic and will not be reviewing and approving comments.
    Please consider this topic to be closed.

  21. James Beal says:

    Hello,
    If you need a consumer to test out that bad boy I’m your man. It sounds like another revolutionary product that Intel is creating and I can’t wait till its out on the market so I can look at it!
    Have a good day,
    J. Beal

  22. Nick says:

    Penryn is I believe derived from the Welsh Penrhyn which means promontory or headland.
    The Cornish town in England of Penryn gets its name from the Cornish word Pennrynn also meaning ‘promontory’.
    Is the new Intel Penryn family the highland of
    chip technology?

  23. Luke Scott says:

    Just to let you know Penryn is also an ancient Town in the UK which stands at the top of the Penryn river.

  24. Allen M. says:

    How does the new high-density interconnect (HDI) PCB boards for the next generation Montevina notebook platform, impact, change, challenge, the motherboard capacitor decoupling scheme?

  25. John says:

    Ronny, the only reason quad cores will NOT fit in laptops is due to the heat it generates; laptops were maxed when the dual cores went into play with a second cooling vent on the rear. The more cooling vents, the less room for ports. PC designers such as dell, hp, and gateway are trying to target the MOST, widest group of consumers who need connectivity. With us, we can just slide in a pcmcia card to add more ports and we’re fine with the card sticking out the side, affecting appearance; the others who want a sleek looking laptop.

  26. Rakesh says:

    The blogs are good. But i get nervous Because i will buying a laptop in jan 2008. and i loose both The new Intel’s Penryn and low rates of core 2 duo!

  27. Ed G. says:

    Thanks for the insight to Penryn’s naming source! I drove out to Penryn, CA last Monday looking for a produce stand that wasn’t open and would have paid more attention to the rocks had I read this blog beforehand. Will be returning to pick up some produce but most likely will have to get the processor elsewhere!

  28. Sadri says:

    It look like Intel has delay to send out Penryns processor. Is there anybody know the ETA? For example I want to know when 6460DTU which is T61 series will be in market (In Canada)

  29. Robert Caldwell says:

    Hello,
    This new technology is amazing and quite interesting. We first came across it in the March 6th issue of Machine Design and so we have sought out to discover more information. I work for a manufacturing company that specializes in custom design and fabrication of clean room components. We currently serve Intel in and outside of the U.S.; they are one of our largest customers. With this new technology will we see any new Fab processes? How might our company obtain more opportunities to supply and support Intel’s new processes? Any insight that you could supply us would be greatly appreciated.
    Thanks,
    Robert Caldwell
    Pro-Fab, Inc.
    ISO90001-2000 Certified
    PH: 505-869-2222
    FX: 505-869-3333
    E-Mail: robertc@pro-fab.com
    http://www.pro-fab.com

  30. alex addison says:

    i’am living in turkey. i will i notebook. but i don’t know anything about ” PENRYN INTEL CORE 2 DUO T8300” . i want learn it’s PENRYN series is double cpu or 4 cpu. and whats 2.1. performance

  31. kay says:

    The original Penryn was founded in 1216 in Cornwall, England and was given it’s charter by the Bishop of Exeter in 1236