Results 1 to 23 of 23

Thread: Nehalem 101 part2 - 1156 and P55/P57

  1. #1
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147

    Nehalem 101 part2 - 1156 and P55/P57

    Hey guys,

    the launch is still almost half a year away, but i thought id post a second thread with some details so your already prepared for "Core i5" and the new 1156 platform and how to clock it. Ill try to post more details here bit by bit as i have the time and new rumors appear on the net
    First lets look at the parts, intel is apparently following the same strategy as with Core i7
    They will only use very few SKUs, 3 cpus for socket 1156 at launch with more to follow, but like with i7, with each introduction 1 will be EOL'ed most likely.
    So there will be an overlap of more SKUs but its rumored that there will only be 3 non igp and 3 igp cpus (plus some LV versions of those) maintained at the same time.

    "Core i5 935" 2933MHz
    4cores 8threads VTT TXT
    Bclock 133
    QPI Multiplier 16x *3(2133MHz/4.3GT/s)
    CPU Multiplier 22x (2933MHz) 23x (3066MHz)*1 27x (3600MHz)*2
    Uncore Multiplier 15x (2000MHz)
    Memory Multiplier 10x (1333MHz)

    "Core i5 925" 2800MHz
    4cores 8threads VTT TXT
    Bclock 133
    QPI Multiplier 16x *3(2133MHz/4.3GT/s)
    CPU Multiplier 21x (2800MHz) 22x (2933MHz)*1 26x (3466MHz)*2
    Uncore Multiplier 15x (2000MHz)
    Memory Multiplier 10x (1333MHz)

    "Core i5 915" 2666MHz
    4cores 4threads
    Bclock 133
    QPI Multiplier 16x *3(2133MHz/4.3GT/s)
    CPU Multiplier 20x (2666MHz) 21x (2800MHz)*1 24x (3200MHz)*2
    Uncore Multiplier 15x (2000MHz)
    Memory Multiplier 10x (1333MHz)

    *1 Turbo Mode with 4 cores active
    *2 Turbo Mode with 1 core active
    *3 according to some sources all i5 cpus still use QPI, either to connect the igp chipset to the cpu on the cpu package, or to connect the pciE interface to the rest of the cpu inside lynnfield. The multiplier seems to be 16x however, which would allow bclocks of up to 250mhz compared to 222mhz with an 18x multiplier, which is the lowest qpi multiplier the 1366 platform supports


    1156/P55 compared to 1366/x58:
    lower cpu clocks
    lower uncore clocks
    Less memory channels (2 vs 3)
    higher memory clocks possible (uncore has to be 1.5x mem clock, not 2x like on i7)
    higher bclocks possible (min qpi multiplier is 16x, which allows up to 250bclock without breaking the 4ghz/8GTs limitation)
    igp cpus dont have an imc (imc is inside the igp chip on the cpu package, and its P45 based, not i7 based)
    non igp cpus have an i7 based imc
    Less pciE lanes (16 vs 32)
    new hdd cache (only on p57?)
    lower to identical cpu price
    lower to identicl mainboard price

    1156/P55 compared to 775/p45:
    higher cpu clocks
    better igp performance (~2x g45)
    Better uncore to memory ratios (memory multipliers)
    new hdd cache
    identical to higher cpu price
    identical to higher mainboard price

    so you get the picture, its about taking 1366/x58 and making it cheaper so it can go mainstream.
    time will tell whether the edges have been cut too much or just right
    In the end 1156 will be in between 775 and 1366, so there will be 3 sockets in the desktop segment actively supported at the same time, so far we always had one actively supported together with the last gen socket fading out, and in recent years intel did an amazing job keeping 775 alive, which means we have lived with one socket for the recent years, so adding 2 more sockets is quite a brave and questionable move.

    Btw, the name i5 is not decided yet it seems, it will most likely be called something entirely diferent, not even core, but ill go along and call it i5 like everybody else in this thread. The new platform will be the mainstream to entry level version of 1366 and i7. actually the positioning isnt exactly clear and intel might not even have decided it yet. it could go all the way from entry level up to performance or even highend and only leave the ulra highend to 1366, or it could only be the entry level up to mainstream and leave the performance up to ultra highend to 1366. I suspect something in between...

    intel definately plans to keep 1366 alive in the highend desktop/workstation segment and as such will release the fastest cpus for 1366 only. so 1156 cpus will always be one bin below that of the fastest 1366 cpu.

    cpu prices for i5 should be between 200$ and 300$ at launch, possibly above that depending on how intel plans to position it. there will be 100$ and even cheaper i5 cpus eventually, but dont expect them soon after the launch. mainboard prices for i5 should be between 100$ and 200$, so 50-200$ cheaper than 1366 mainboards.

    Core i5 has an integrated pciE controller, so 16 lanes will directly go to the cpu, while extra pciE lanes are available through the south bridge that is directly connected to the cpu. this connection is the dmi bus, which intel has used to connect the sb to the nb since the 915 chipset many years ago. this dmi bus is actually a slightly modified pciE bus and is pretty much 4pciE 1.1 lanes bundled together to one link. thats not a lot of bandwidth considering this bus is used for all onboard devices, hard drives, sound, lan, and tv cards or other add in cards. so if you really push things with raid or tv cards etc, expect this bus to hold you back.
    dont expect a performance boost from the integrated pciE controller on the cpu, it actually does help performance, but the improvement is tiny.

    another thing is that some boards youve probably seen have 3 pciE 16x slots. unless they use a pciE switch like PLX/NF200 or hydra the third slot will be very limited since it actually comes from the soutbridge. the southbridge ironically has 4 pciE 2.0 lanes, but those are actually 1.1 lanes which are 2.0 device compatible...
    so its more of a marketing 2.0 thing... dont plan to use those third pciE 16x slots for tri vga setups, i doubt nvidia would allow it anyhow.

    which brings up the next thing, sli on P55... will it work without NF200 chip? its still not set in stone but it seems nvidia will continue the same strategy they had with x58 and license sli on a per board fee.

    what else is new, there will be intel turbo cache2 pretty much, that will serve as a flash buffer/cache to your hard drives, which should improve hdd performance significantly, unless your running an sdd, in which case i doubt there will be much of a boost if any.

    another thing is the uncore to memory multiplier limitation, wich will change. right now on i7 you have to run double the memory multiplier for the uncore, on i5 you will only have to run 1.5x the memory multiplier for uncore.

    so that means with 4000Mhz uncore on i7 you are stuck to DDR3 2000, on i5 you can run DDR3 2666. that will probably still limit the best memory out there, but its better than i7 for sure, where you need subzero cooling to max out your memory clocks.

    There will be i5 igp cpus with an igp northbridge under the hood, on the same package, but thats not really interesting for people here... the igp performance will be far from specatcular and the added heat doesnt help. another reason i dont recommend those cpus to xs readers is that they use QPI to connect the igp to the cpu, which will most likely limit overclocking.
    Last edited by saaya; 04-26-2009 at 12:15 AM.

  2. #2
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    it seems 775 will live longer than most expected, with support throughout 2012, so there will be 3 sockets, 1366 for the highend, 1156 for the mainstream and 775 for the entry level.

    i suspect 1156 will come at 2133Mhz 2400Mhz and 2666Mhz initially, so it will replace the low end of 1366 and the highend of 775.

    If you guys have any questions, post here or send me a pm!

  3. #3
    all outta gum
    Join Date
    Dec 2006
    Location
    Poland
    Posts
    3,390
    About the memory:uncore ratio - I wonder if [i5 uncore @ 2000 with dual channel memory @ DDR-2333] is going to be faster than [i7 uncore @ 4333 with 3-channel DDR-2166 memory].
    www.teampclab.pl
    MOA 2009 Poland #2, AMD Black Ops 2010, MOA 2011 Poland #1, MOA 2011 EMEA #12

    Test bench: empty

  4. #4
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    same timings? havent tried, but i dont think so... i7 has a much higher uncore clock in your comparison and that should help.
    i havent benched much mem though on i7 since there doesnt seem to be a big benefit from tweaking it.

    same uncore, same mem clock, same timings, dual channel mem, i5 is slightly slower than i7
    intel said by adjusting the buffers to get the 1.5x ratio they lost some perf... maybe it can be tweaked to have the same perf as i7, but idk if its worth the effort as memory perf is really not whats holding these cpus back
    Last edited by saaya; 03-25-2009 at 02:15 AM.

  5. #5
    Memory Addict
    Join Date
    Aug 2002
    Location
    Brisbane, Australia
    Posts
    11,651
    Interesting about the hdd cache... wonder how well that will play with MLC SSDs hehe
    ---

  6. #6
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    like i said, doubt itll help with sdds... its supposedly up to 200mb/s read iirc.
    whats interesting tho is that you can format a part of that cache to hold the os!
    so you can boot directly from the os... and another interesting thing is that you can use a partition of the flash memory to cache certain apps, so the load immediatly.
    but i think that basically means its compatible with microsofts turbo boost or whatever they call it.

    so the really interesting part is loading the os into it and speeding up sata and sas hdds by basically upgrading their cache.
    its kinda like upgradeable L5 cache if you see the cpus system memory as L4 and hdd as L6
    Last edited by saaya; 03-25-2009 at 02:56 AM.

  7. #7
    Memory Addict
    Join Date
    Aug 2002
    Location
    Brisbane, Australia
    Posts
    11,651
    DDR5 memory based cache would be nice LOL
    ---

  8. #8
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    well response time of nand will always suck compared to dram so i think its actually better for cache, and since it can read and write so fast beeing volatile is ok since you can reprogram it quickly at start.
    but the bw will be limited by dmi, like i said, the sb is connected to the cpu via 4 pciE 1.1 lanes, for everything... thats not that much bw...

    the best would be if we would have 16gb modules, then we could run everything off the memory...
    during boot just load the os+apps image into the mem, and then resume it and run from there
    Last edited by saaya; 03-25-2009 at 03:29 AM.

  9. #9
    all outta gum
    Join Date
    Dec 2006
    Location
    Poland
    Posts
    3,390
    Saaya, could you perhaps shed some light on FDI (Flexible Display Interface)?
    www.teampclab.pl
    MOA 2009 Poland #2, AMD Black Ops 2010, MOA 2011 Poland #1, MOA 2011 EMEA #12

    Test bench: empty

  10. #10
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    dont know much about fdi, afaik its just a link routing the igp frame buffer on the cpu package to the display driving hardware in the ibex peak chipset.
    i guess other graphics adapters can use the same bus and send data to the ibex peak chipset to use the integrated display output... but thats only interesting for mobile parts or multi display setups...

    i updated the first post in this thread again

  11. #11
    I am Xtreme zanzabar's Avatar
    Join Date
    Jul 2007
    Location
    SF bay area, CA
    Posts
    15,871
    with the IGP can it do things now, like playback 1080p without killing the cpu like amd can
    5930k, R5E, samsung 8GBx4 d-die, vega 56, wd gold 8TB, wd 4TB red, 2TB raid1 wd blue 5400
    samsung 840 evo 500GB, HP EX 1TB NVME , CM690II, swiftech h220, corsair 750hxi

  12. #12
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    one important thing many people overlook about i5 is that the igp cpus are actually just like the current intel platform.
    they have an igp northbridge, a southbridge and a cpu, just like p45, the only difference is that the igp northrbidge is integrated into the cpu package, and connected to the cpu via qpi. so from i7 intel took a step forward to non igp cpus which integrate pciE, and took a step backwards for igp platforms which are a real 3 chip solution again, just that one chip is "hidden" on the cpu package.
    Last edited by saaya; 04-23-2009 at 11:16 PM.

  13. #13
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    updated the first post

    i5 are rumored to still use qpi, even the non igp cpus.
    the igp cpus use a G45/P45 based igp chipset which is on the cpu package. the cpu apparently either doesnt have an imc, or its not used. instead the igp chipsets imc is used, and the igp and cpu are connected via qpi.

    the non igp cpus seem to still use qpi internally, apparently intel reduced the x58 ioh pciE hub to half the pciE lanes and then moved this entire block inside the cpu silicon. so the pciE hub is not connected to the cpu directly, but is connected to a qpi interface. so the data goes from pciE to qpi and then to the cpu, this means extra transistors and possibly heat, and worse efficiency, but its a safe step which is less likely to cause bugs apparently.

    both cpus seem to have the qpi multiplier set to 16x, which is lower than on 1366 and should hence allow higher bclock overclocking without maxing out the qpi clock.
    if the qpi interface maxes out at around 4ghz or 8gt/s like on x58 and i7 cpus, then we can except bclocks of up to 250mhz on i5. UNLESS there is another limiting factor.

    how much the OFF cpu die memory controller on igp cpus influences the performance is unknown, and its based on P45/g45 instead of i7, which might reduce performance compared to i7 as well. id suspect that igp cpus have a worse memory performance than non igp cpus, how much is hard to tell...

  14. #14
    Registered User
    Join Date
    Dec 2008
    Posts
    63
    What's the difference between P55 and P57?

  15. #15
    Xtreme Addict
    Join Date
    Nov 2008
    Location
    Heart of Europe
    Posts
    1,992
    I think that P57 is meant for HTPCs and such, at least boards I saw looked like that. Maybe it somewhat ready for CPU with IGP.

    Reminds me that I saw P55 - Foxconn Quantum Force - Infernal Katana .. looks very very nice. (sister of Bloodrage )
    i7 930 D0 - 4,2 GHz + Megashadow
    3x4GB Crucial 1600MHz CL8
    Foxconn Bloodrage rev. 1.1 - P09
    MSI HAWK N760
    Crucial M500 240GB SSD
    SeaGate ES.2 1TB + 1TB External SeaGate
    Corsair HX 850W (its GOLD man!)
    ASUS STX + Sennheiser HD 555 (tape mod)

    Old-new camera so some new pics will be there.. My Flickr My 500px.com My Tumblr

  16. #16
    3D Team Captain Don_Dan's Avatar
    Join Date
    May 2007
    Location
    Munich, Germany
    Posts
    4,199
    Quote Originally Posted by saaya View Post
    how much the OFF cpu die memory controller on igp cpus influences the performance is unknown, and its based on P45/g45 instead of i7, which might reduce performance compared to i7 as well. id suspect that igp cpus have a worse memory performance than non igp cpus, how much is hard to tell...
    Does that mean it "could" be possible to make DDR2-boards?

    Quote Originally Posted by chew* View Post
    You can never have enough D9's.

  17. #17
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    Quote Originally Posted by salad View Post
    What's the difference between P55 and P57?


    i heard some things have changed, but im not sure which... there is a rumor that p55 will support robson (ONFI flash controller similar to the one on intels ssd's integrated in the chipset to boost hdd performance)

    Quote Originally Posted by Don_Dan View Post
    Does that mean it "could" be possible to make DDR2-boards?
    of course... but intel would have to support it cause the G45 based chipset is on the cpu package, and i think the pin routing would have to be slightly diferent for ddr2 support, and there would have to be bios support which intel wont unlock/allow for sure... so the hardware actually can run ddr2 afaik, but it wont be possible...

    same with i7, which i heard actually does support ddr2... but intel wont support it... afaik ddr3 is pretty much ddr2+, its mostly the same with some logic added, and depending on how you design a ddr3 memory controller it can support ddr2 with almost no effort and extra logic. thats why amds cpus actually dont have 2 memory controllers, its the same memory controller on the cpu that supports ddr2 and ddr3...

    and actually pre deneb cpus already supported ddr3 afaik, but amd never unlocked it cause they wanted to tweak it more and wait for the ddr3 prices to come down.

    anyways, there are a lot of sceptics when it comes to robson...
    while i cant say for sure how the demo worked that intel showed me at computex, it DID work amazingly well... they ran a script of photoshop and outlook and other hdd intensive apps and it ran much faster with robson enabled than using just the hdd, and they used a 7200rpm 16mb recent sata2 hdd, so it wasnt a slow hdd they used to compare...

    the script finished in half the time or even less with robson enabled, and the script ran for 30-60 seconds... there were two scripts actually iirc... what i remember vvidly is how outlook launched immediatly as if it was already running and just in the background before... really amazing if it works the way intel showed it in the demo...

    i asked if those apps need to be preloaded into the robson flash modules, but intel didnt clearly answer that... anyways, the modules will be 8 to 16gb for starters and thats more than enough to preload any important app for desktop users, and for workstations, hopesfully there will be bigger modules... but well... workstations without ssd's are kinda a weird idea nowadays ^^

  18. #18
    3D Team Captain Don_Dan's Avatar
    Join Date
    May 2007
    Location
    Munich, Germany
    Posts
    4,199
    Thanks for the elaborate answer!

    Quote Originally Posted by chew* View Post
    You can never have enough D9's.

  19. #19
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    your welcome

    is there anything else you guys want to know?
    about the branding, well im sure you all heard, from now on core i7 i5 and i3 will be used, but for platforms... and not for cpus themselves...
    so a nehalem 1366 cpu at low clockspeeds might be called core i5 in future, and a "fast" nehalem 1156 will be called core i7, and a slow one core i3... it will all get very very messy... not to mention that they also added a 5star rating on top of that, so there will be core i7 5 star, core i7 4 star etc etc...
    so whats better, a core i7 1366? or 1156?
    is a core i7 4 star better or worse than a core i5 5 star?
    and then there are several core i5 4 star chips at diferent clockspeeds, so which is better?
    this is going to get messy... intel is copying nvidias naming/renaming sheme here in the worst possible way...

    overall, theres not much new for core i5... at all...
    the igp performance of 1156 cpus will be better than past igp chips, duh!, obviously , but itll still stink... and the memory performance of igp 1156 cpus seems to be severly handicapped, so much that it event affects cpu performance and reduces it by a few percent... so stay away from those igp 1156 cpus! well hopefully the final chips have this problem fixed, but even then id recommend everybody to stay clear of those igp cpus since they will consume more power thanks to the on-cpu-package-northbridge and will overclock and perform slightly worse than the non igp cpus...

    overclocking on 1156 could be quite annoying if you think of the huge turbo mode multipliers... so youll either have to pump huge vcore through your chip even though it actually doesnt need it with 2 or more active threads since the clocks drop down a lot, or you will run low vcore but have to disable turbo mode cause otherwise the cpu will jump to huge clocks with only one thread, crashing the system... maybe itll be possible to adjust vcore and turbo multipliers properly and we wont have those annoyances... but i havent heard or seen anything like that yet, unfortunately...

    so i still recommend the same thing... get a good basic 1366 board with a 2.66ghz xeon...
    Last edited by saaya; 06-21-2009 at 08:31 PM.

  20. #20
    Banned
    Join Date
    Apr 2008
    Location
    Brisbane, Australia
    Posts
    3,601
    =____=

    I never thought about the turbo issue Saaya. . . . damn. i7 still looks good

  21. #21
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    funny to look back on this now
    my speculation about 2933, 2800 and 2666 was right, but the naming was 8xx and 7xx despite the chips performing about the same as 9xx chips... but intel clearly positioned 1156 lower than 1366, and even went for 2 lower clocked chips, 2.53 and even 2.4ghz... ironically, they clearly positioned 1156 below 1366 according to ratings, yet pricing is about the same... its definately confusing, i think calling the 2.66ghz 1156 part 925 would have made a lot more sense... but oh well

    1156 will become interesting when 32nm dualcore parts arrive in early 2010 with clocks going up to 3.46ghz, and thats without turbo! but not much after that VRM 12 will be introduced and there will be new boards and new cpus, maybe 1367 and 1157 to diferentiate... we will see...

    its really a shame intel wont bring 32nm dualcore chips to 1366... there might be dualcore 32nm xeons for 1366 though... that would definately be sweet

    in the end, 1156 will be interesting because of 32nm dualcore chips and lower cpu prices... which intel could have done on 1366 as well... its really a pity that intel didnt go for one platform and instead split it all up...

    lets hope those 32nm dualcore chips will have an unlocked uncore multiplier unlike the current 1156 cpus and 32nm ES chips... and lets hope there will be 1366 versions too

  22. #22
    Xtreme Enthusiast
    Join Date
    Feb 2007
    Location
    So near, yet so far.
    Posts
    737
    Quote Originally Posted by saaya View Post
    funny to look back on this now
    my speculation about 2933, 2800 and 2666 was right,.....
    Yeah, a bit funny "now". I guess its time to tidy up this sticky dude..
    [[Daily R!G]]
    Core i7 920 D0 @ 4.0GHz w/ 1.325 vcore.
    Rampage II Gene||CM HAF 932||HX850||MSI GTX 660ti PE OC||Corsair H50||G.Skill Phoenix 3 240GB||G.Skill NQ 6x2GB||Samsung 2333SW

    flickr

  23. #23
    Xtreme X.I.P.
    Join Date
    Nov 2002
    Location
    Shipai
    Posts
    31,147
    nah, time to let it go and fade away
    it served its purpose of giving people an idea of what was coming with 1156

Bookmarks

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •