Further Intel Penryn details emerge

Discussion in 'Beyond3D News' started by Arun, Jan 27, 2007.

  1. Arun

    Arun Unknown.
    Moderator Legend Veteran

    Joined:
    Aug 28, 2002
    Messages:
    5,023
    Likes Received:
    300
    Location:
    UK
    <a href="http://dailytech.com/Life+With+Penryn/article5869.htm" target=_b3dout>Dailytech</a> and <a href="http://www.hexus.net/content/item.php?item=7754" target=_b3dout>Hexus</a> both have a bunch of new mini-facts on Penryn today, among which what they claim is the first official die shot of 'Penryn'. Apparently, most of the extra transistors (410M vs 298M for Conroe) will be used for extra cache (6MiB vs 4MiB), although the chip also implements SSE4 - but it is unclear to us whether most of these are actually new native instructions or not.

    Given how much of incremental improvement Penryn seems to be, it wouldn't be too surprising if this was mostly a way to improve the addressable market of SSE4, while the biggest boosts from these instructions would be for Intel's upcoming Nehalem microarchitecture. At the same time, Intel's CEO recently said in a conference call that Penryn would get specific performance boosts for floating point code. Given that K8L is said to have substantially improved FP64 performance and SSE performance hopefully matching that of Conroe, it would be quite interesting if Intel decided to further boost their FP units, in addition to adding these new SSE4 instructions.

    At this point, nobody really seems to know what Intel actualy did (if anything!) for FP with Penryn - but this will surely have a significant impact on Intel and AMD's respective competitiveness in the server market. Furthermore, Intel is marketing high-k and metal gates extensively for the 45nm transition, and righly so - it's likely to be one hell of a nice boost to performance/watt. Even with SOI, AMD is very unlikely to match Intel's 45nm leakage levels for some time, so it will also be interesting to watch how much that affects the bigger picture, and whether AMD can remain competitive in the notebook space anyway - at least until they <a href="http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=197001065" target=_b3dout>catch up</a> with 45nm high-k metal gates in 2008, as announced today by IBM. TSMC, on the other hand, is not planning to introduce high-k metal gates before the 32nm process node.
     
  2. Ateo

    Newcomer

    Joined:
    Sep 30, 2006
    Messages:
    178
    Likes Received:
    3
  3. INKster

    Veteran

    Joined:
    Apr 30, 2006
    Messages:
    2,110
    Likes Received:
    30
    Location:
    Io, lava pit number 12
    TheAlSpark likes this.
  4. TheAlSpark

    TheAlSpark Moderator
    Moderator Legend

    Joined:
    Feb 29, 2004
    Messages:
    20,980
    Likes Received:
    6,183
    Location:
    ಠ_ಠ
  5. bbot

    Regular

    Joined:
    Apr 20, 2002
    Messages:
    745
    Likes Received:
    11
    Will it be able to match the FP performance of the Cell Broadband Engine?
     
  6. trinibwoy

    trinibwoy Meh
    Legend

    Joined:
    Mar 17, 2004
    Messages:
    10,481
    Likes Received:
    500
    Location:
    New York
    Obviously no, but how and why is that relevant in any way?
     
  7. Arty

    Arty KEPLER
    Veteran

    Joined:
    Jun 16, 2005
    Messages:
    1,906
    Likes Received:
    55
    I dont expect a significant performance boost from Penryn, but from Nehalem.
     
  8. Arun

    Arun Unknown.
    Moderator Legend Veteran

    Joined:
    Aug 28, 2002
    Messages:
    5,023
    Likes Received:
    300
    Location:
    UK
    Same here, but I would actually expect some nice clockspeed increases and performance/watt improvements in Penryn. How likely is it that K8L is more efficient per transistor and per clock (by more than 5-10%) than Penryn? Not very, in my book. NGMA isn't Netburst, and yet it looks like Intel is going to have a clockspeed and power advantage. Ouch?

    It just seems to me AMD's management is not preparing for the worst or being paranoid in enough, and that's rarely a winning strategy in this industry... Hopefully this is just an outsider's view of the situation, and they'll do just fine. We'll see.


    Uttar
     
  9. Arty

    Arty KEPLER
    Veteran

    Joined:
    Jun 16, 2005
    Messages:
    1,906
    Likes Received:
    55
    I actually meant that in comparison to Conroe itself. Sorry for not being clear. :oops:

    More than a performance boost I view Penryn as Intel's step towards making the Quadcore a mainstream processor.
     
  10. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,193
    Likes Received:
    3,134
    Location:
    Well within 3d
    Per core, K8L might not be competitive with Conroe in integer code, much less Penryn.
    In FP, K8L might be competitive with Conroe. AMD's bandying numbers that put it at a significant advantage over Opteron, and possibly a per-clock advantage in sustained throughput over Conroe. Unfortunately, the chips at introduction are likely to clock 600 MHz less than current Conroes, with Intel likely sandbagging a few speed grades in case AMD's process improves.

    Against the Penryn derivatives, K8L might be dealing with an even larger frequency gap.

    Theoretically, AMD has a better platform, and even current K8s far much better in multi-threaded comparisons, even if they get hammered in single-threaded.

    It's sort of what IBM has done with POWER vs Itanium, where individually weaker cores can be performance competitive when mated with a strong platform.

    This isn't necessarily a winning proposition, either, but it's not an outright loss.

    AMD is in the unenviable position of adopting this strategy, but without all of IBM's advantages. IBM owns the entire system stack, AMD's going IBM's route without the necessary broad scope.
     
  11. Arun

    Arun Unknown.
    Moderator Legend Veteran

    Joined:
    Aug 28, 2002
    Messages:
    5,023
    Likes Received:
    300
    Location:
    UK
    I'm interested in how the platform can help for the dual-core (and single-core!) solutions. Sure, it's going to help AMD a lot for the quad-core market, and that might be significant in the server area. But unless I'm missing something, I don't see how it helps for the rest of their product line.


    Uttar
     
  12. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,193
    Likes Received:
    3,134
    Location:
    Well within 3d
    The platform in that case does little or nothing at all to help, which is partly why IBM's POWER isn't in that market anymore.

    From some statements from the company, it appears AMD is ceding the upper echelon leadership in those segments to Intel, at least until K8L's successor comes out (which if AMD's execution does not improve will be facing an Intel processor a node ahead and a full core revision enhanced).

    AMD's strategy seems to be hunkering down and trying to maintain share in the more lucrative low/mid-end server space. The high end is something Intel may likely maintain leads in, in part because the IMC on K8 with current DRAM is likely to constrict the number of DIMMs per processor.

    Desktop won't be a blowout for Intel, but the margins are going to suck as AMD shifts to be a value proposition like in the pre-Athlon days and those bad quarters where K7 had those insanely inflated PR ratings against much better Northwoods.

    For mobile, AMD will likely cede the high-end mobile to Intel as well. The mid to value proposition looks salvageable.

    Intel might allow AMD some breathing room for a while by concentrating on the high-margin stuff, if only to keep investors happy with fatter margins.

    AMD is going to be suffering for quite some time. We'll see if its gamble that a convergence with ATI's tech will play out when the products finally become available, assuming Intel doesn't beat them to it because of the classic AMD and classic ATI habits of missed execution.
     
  13. TheAlSpark

    TheAlSpark Moderator
    Moderator Legend

    Joined:
    Feb 29, 2004
    Messages:
    20,980
    Likes Received:
    6,183
    Location:
    ಠ_ಠ
  14. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,193
    Likes Received:
    3,134
    Location:
    Well within 3d
    Interesting that Intel has opted for something other than Nickel-silicide for the gate material. I think the AMD/IBM/etc. papers stated they were planning on using that for their future nodes.

    I wonder if they're sticking to that, or if they will be shifting the materials used as well.
     
  15. The_Wolf_Who_Cried_Boy

    Newcomer

    Joined:
    Feb 18, 2005
    Messages:
    172
    Likes Received:
    9
    Location:
    Floating face down in the stagnant pond of life.
    Having stumbled over this link at Real World Tech I quoted from and posted at AMDZone but it hasn't drawn much response.

    RE: PENRYN
    Plus these estimations;

    Code:
    [b]AMD[/b]      Barcelona core    65nm    25 mm2
    [b]Intel[/b]    Penryn core       45nm    24 mm2   
    
    Granted I have no knowledge of CPU design but it strike me that the latest prediction of AMD's imminent implosion under the seemingly wizard like abilities of Intel are yet again grossly exaggerated. It's all good and well to claim 20-30% increases in transistor switching speed but nobody is making mention of any matching reduction in wire delay. I gather logic pipelines are designed with a specific balance between wire and transistor delays, likewise I gather improving the performance of one but not the other simply results in design imbalances, your absolute clock limit will be defined by the weakest link. Is it likely the main advantage of Intel's new node will be power reduction? While performance per watt is nice and all, absolute performance at a given price point is what gets peoples attention and it seems that will be a very tight race. Perhaps dual core Barcelona will have a larger die, but Penryn as the vanguard 45nm product would have node transition expenses amortized into its costs. Perhaps the initial stepping of Barcelona will only clock to 2.6 GHz, but Penryn will be debuting against the second or possibly even third stepping, and taking AMD's 90nm efforts as a guide these will be running faster on appreciably less power. Perhaps I'm blinded by fanboi devotion but I'm not seeing a run away victory for Intel.

    I can't fathom why IBM/AMD using immersion lithography would give a process node longer legs relative to Intel sticking with air, if you can define a feature at a given size then you can define a feature at a given size I would have thought, but I certainly put enough faith in Mr. DeVries knowing his subject matter to take his comments at face value. Anyone like to give that one an explaination?




    PS 1 down, 865 to go; Appologies to Geo for the inane, pointless, redundant and infantile spamfest...:wink:
     
  16. TheAlSpark

    TheAlSpark Moderator
    Moderator Legend

    Joined:
    Feb 29, 2004
    Messages:
    20,980
    Likes Received:
    6,183
    Location:
    ಠ_ಠ
    IBM demonstrated ~32nm using immersion lithography last year. See how far Intel can go with air then. ;)

    http://domino.research.ibm.com/comm/pr.nsf/pages/news.20060220_nemorelease.html
     
  17. The_Wolf_Who_Cried_Boy

    Newcomer

    Joined:
    Feb 18, 2005
    Messages:
    172
    Likes Received:
    9
    Location:
    Floating face down in the stagnant pond of life.
    I took Han's comments to be specifically aimed at each camps 45nm node (?).
     
  18. TheAlSpark

    TheAlSpark Moderator
    Moderator Legend

    Joined:
    Feb 29, 2004
    Messages:
    20,980
    Likes Received:
    6,183
    Location:
    ಠ_ಠ
    Well, with immersion lithography, they'd be able to still use DUV. I guess they figure for the future they'll implement immersion litho for 45nm and they'll still be ready for the move to 32nm without replacing the laser system.
    edit:
    hm...I think I should just use this in my signature. That way every....
     
    #18 TheAlSpark, Feb 4, 2007
    Last edited by a moderator: Feb 4, 2007
  19. Arun

    Arun Unknown.
    Moderator Legend Veteran

    Joined:
    Aug 28, 2002
    Messages:
    5,023
    Likes Received:
    300
    Location:
    UK
    Please keep in mind I'm replying to this as the Devil's advocate, I just want to give you another possible perspective on this, which might or might not be more accurate in the end.
    You talk a lot clock limitations, but that has been fairly irrelevant for Conroe, and will also be so for Penryn. Remember how high people can clock their E6300s if the motherboard's FSB can manage it? And yet, that's supposedly the low-end bin.
    If Intel can reduce the power by 40% for a given clock rating, they can also very easily incease the clocks for a given power rating. In the end, Netburst was massively limited by its thermal characteristics. While this is not *as* true for NGMA, it's also worth considering. Intel doesn't need further clock improvements now, because that's not their bottleneck. You'll see much higher clocks with Penryn if Intel feels it is necessary, and the lower power dissipation is what'll make that practically viable for all of their markets.
    It obviously will, and it'll also have a smaller number of gates in the core (which could point towards a lower IPC, assuming Intel's design is as efficient as AMD's, which is very questionable) and only half the cache.
    And unless AMD suddenly takes over the market, it'll also have 3x+ times the volume to amortize those very same transition expenses...
    Steppings are an interesting perspective. It would be quite interesting if AMD could improve perf/watt further before Penryn emerges. I'm skeptical the results of this will be quite as dramatic as on 90nm, let alone because they already have another part that was testing the waters on the process, which is Brisbane. The last time around, unless I'm mistaken, these steppings were on the first chip that came out of the fab for 90nm.

    Personally, I'm not willing to call it either side's victory before I see the K8L IPC in action. If AMD can't match Intel on an IPC basis for their respective dual-core offerings, then it's fairly easy to see AMD is in trouble for the desktop market (excluding the server segment, where they've currently got an interconnect advantage). If AMD can beat Intel in terms of IPC, then that might put them in an interesting position, possibly compensating Intel's 45nm power advantage too. If they can just match Intel, then we'll have to watch the budget market, and consider how good they can compete there and if they can maintain acceptable margins. Although you'd still expect AMD to lose on perf/watt if they can't match Intel's IPC, which would be a tad problematic for the laptop segment they seem to cherish so much nowadays...
     
  20. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,193
    Likes Received:
    3,134
    Location:
    Well within 3d
    AMD's statements are to the effect that the new cores will not beat Conroe in integer code IPC. Barcelona won't lag as far behind as K8 does, but it isn't as aggressive as Intel's chip in memory reordering, prefetching, and cache size.

    Barcelona will likely lag nearly as badly as K8 in clock speed until AMD's process goes through some gradual tweaking to increase performance. According to AMD, 65nm transistors at the outset will perform identically to the 90nm transistors they replace.

    For FP vector code, there may be a much better chance at higher sustained throughput, if some of the core changes that were made to fix certain bottlenecks prove to be effective. I'd imagine a narrowed clock-speed gap might make Barcelona or a future revision even more competitive.

    In the single/dual core desktop market, Intel's offering will likely keep the premium top spot.

    AMD has repeatedly stated its emphasis on the server market for "K8L" or whatever chips fall in the same area of the code name, where the supposed superiority of the platform can come into play.
     
    #20 3dilettante, Feb 6, 2007
    Last edited by a moderator: Feb 6, 2007
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...