If geeks love it, we’re on it

Making sense of Lynnfield: Is Bloomfield really better?

Making sense of Lynnfield: Is Bloomfield really better?

Correction (15 September, 2009 @ 6:15 PM EST): The original run of this article indicated that the Lynnfield’s onboard PCIe controller depended upon the system’s bclock generator for its frequency. This is incorrect and has been redacted. All Lynnfield CPUs use an independent clock generator for the PCIe controller with a nominal frequency of 100MHz. Raising the bclock on a Lynnfield system will not push the PCI Express bus out of spec.

Original story follows:

Right on schedule, Intel has launched new Core i5 and Core i7 processors based on the LGA1156 Lynnfield core. Arriving in three flavors, are the new chips enough to unseat AMD’s price/performance crown? How do they stack up to the existing Bloomfield Core i7s? How are they different? Are they worth buying? We wanted to answer these questions ahead of our official Core i7 860 review which is coming soon.

First off, we wanted to drop all the current Core i5 and Core i7 parts in a table to show you their official specifications.

The formal specs for Core i5 and Core i7 CPUs

The formal specs for Core i5 and Core i7 CPUs

Next, we want to make some terminology very clear so as to avoid confusion going forward:

Bloomfield: This is the codename for the desktop Nehalem-based processors which launched last year. They use the LGA1366 socket, they are sold exclusively with Core i7 900 series branding, and are paired exclusively with the X58 chipset. The current roster of Bloomfield chips includes: Core i7 920, Core i7 950, and Core i7 975 XE.

Lynnfield: This is the codename for the desktop Nehalem-based processors which launch today. They use the LGA1156 socket, they are sold with Core i5 700 and Core i7 800 series branding, and are paired exclusively with the P55 chipset. The current roster of Lynnfield chips includes: Core i5 750, Core i7 860, and Core i7 870.

Now we’ll answer the million-dollar question…

How is Lynnfield different from Bloomfield?

First and foremost, Lynnfield chip is a value-conscious part. This does not mean that the two architectures won’t trade blows here and there, but it is important to know that Lynnfield is a Bloomfield that has been simplified in five big ways.

Out with QPI, in with DMI

QPI, or QuickPath Interconnect, is a new system bus architecture developed specifically for the Bloomfield-based Core i7 models. In its simplest form on a single-CPU system, QPI creates a high-speed 25.6GB/s link between the CPU and the X58 chipset.

The X58 chipset is then directly connected to the PCI Express bus, and uses a slower 2GB/s DMI link to connect to the ICH10(R) southbridge.

Intel X58 for Bloomfield-based Core i7

Intel X58 for Bloomfield-based Core i7

Lynnfield does not use an external QPI link, however. Instead, Lynnfield is launching with the P55 chipset, which condenses the north- and southbridge into a single chip called the PCH, or Platform Controller Hub. The PCH is connected directly to the CPU via that slower 2GB/s DMI link, and the PCI Express controller has been migrated onto the CPU.

QPI isn’t entirely removed, however, as it is still used internally on Lynnfield CPUs to connect the PCI Express controller to the CPU cores.

The Intel P55 chipset for Lynnfield-based Core i5/i7

The Intel P55 for Lynnfield-based Core i5/i7

While dumping the northbridge/CPU QPI lanes in favor of internal QPI lanes won’t mean diddly for your average user, it could sway the opinion of high-end gamers:

  • Tri- or Quad-SLI/Crossfire configurations are not practical on the P55. There is not enough PCIe bandwidth to reliably run them.
  • SLI/CF+Physics configurations are not possible on the P55.
  • SLI or CrossFire configurations are forced into two x8 PCIe lanes on P55, instead of 2 x16 lanes as with X58.

While the difference between 2×8 and 2×16 is immaterial at pedestrian resolutions like 1680×1050, 2×16 can be up to 7% faster at SLI/CF-friendly resolutions like 1920×1200 or 2560×1600. Seven percent is an admittedly small number, but we know that enthusiasts looking at CrossFire/SLI configurations are expecting the best.

On the opposite side of the coin, moving the PCI Express controller to the CPU die eliminates the latency penalty of going out across the northbridge to talk to the GPU. This is a major perk for users who intend to use a single graphics card:

  • Lynnfield systems are absolutely faster than Core 2 Quad systems with a single GPU.
  • The latency reduction means single-GPU Lynnfield systems with a higher CPU frequency than single-GPU Bloomfield systems will usually offer better gaming performance.
  • The latency reduction translates to a direct performance advantage over Phenom II X4 systems: Clock for clock, the new Lynnfield CPUs are much faster than Phenom II.

Changing the memory controller

The Bloomfield’s second biggest change over prior Intel chips was the implementation of a tri-channel memory controller. This swanky tri-channel controller offers Bloomfield Core i7 Extreme chips three channels of up to DDR3-1600, while vanilla Bloomfield Core i7s are limited to three channels DDR3-1066.

core_i5_i7_bandwidth

Lynnfield, on the other hand, defaults to two memory channels of DDR3-1333 and slightly increases memory latency. These changes make for an exchange of blows with no clear winner, just as we saw with GPUs:

  • Core i7 Extreme Edition Bloomfields offer significantly superior memory bandwidth to the Lynnfield.
  • Regular Core i7 Bloomfields offer moderately superior bandwidth to the Lynnfield.
  • Core i7 800-series Lynnfields also happen to support the 12x memory multiplier to enable DDR3-1600 support; this puts Lynnfield and Bloomfield neck and neck if a user knows his/her way around the BIOS.
  • Core i5 700-series Lynnfields are stuck at DDR3-1333, which makes them the least favorable choice for anyone with memory-intensive applications.

Overclocking

The Bloomfield’s overclockability is owed to four multipliers all based off a 133MHz base clock (BLCK):

  • CPU Frequency: BCLK * CPU Multiplier
  • QPI (System Bus) Frequency: BCLK * QPI Multiplier
  • Uncore (L3 Cache, Memory Controller) Frequency: BCLK * Uncore Multiplier
  • DRAM Frequency: BCLK * Memory Multiplier

All four multipliers are completely independent, but there are two minor restrictions: The Uncore multiplier should be at least twice that of the memory multiplier, and the QPI bandwidth should not exceed 8GT/s. Aside from these small hang-ups, most X58 boards happily exceed a 200MHz BCLK, push 4GHz+ on the CPU (200*20), and hum along at DDR3-2000 (200*10).

The new Lynnfield, on the other hand, locks a whole slew of multipliers, which makes it slightly more difficult to overclock than the Bloomfield.

  • CPU Frequency: BCLK * CPU Multiplier (LOCKED)
  • Uncore (L3 Cache, Memory Controller & PCIe) Frequency: BCLK * Uncore Multiplier
    • Core i7 800-series: 133*18 (LOCKED)
    • Core i5 700-series: 133*16 (LOCKED)
  • DRAM Frequency: BCLK * Memory Multiplier

These changes mean that the only way to overclock a system is by adjusting the BCLK until the system achieves the desired CPU frequency. From there, the memory multiplier can be adjusted to compensate for the rise in BCLK, but that’s it.

Don’t let that restriction fool you, though. Lynnfields are easily eclipsing 4000 and 4100MHz in testing, with DDR3 speeds fast approaching DDR3-2133. This is Bloomfield territory, so Lynnfield isn’t exactly crippled–it’s just more difficult.

A new socket

A CPU interfaces with the rest of the system via the pins on the bottom of the chip. Each pin provides voltage or carries data to other parts of the system. This means that drastic changes to the architecture of a CPU must be matched with drastic changes to the pinout of the processor.

Case in point, removing a memory channel calls for a reduction in pin count. The CPU no longer needs the pins that serve as a data channel for that third channel. Moving the PCI Express controller onto the CPU increases the number of needed pins. The processor must now have enough pins to accommodate the signaling requirements two x8 PCIe lanes.

The net result of these changes makes for Lynnfield’s new Socket H, better known as LGA1156. These new LGA1156 processors are not compatible with today’s crop of X58 motherboards, which is why the chips have launched with the P55 partner chipset.

Turbo Mode

Turbo Mode was a feature that debuted with the Core i7 that allows the processor to dynamically overclock itself if the CPU is operating within safe temperature parameters. Bloomfield Core i7s implement this feature by raising the chip’s multiplier by 1x or 2x, depending on the load demanded of the chip.

Lynnfield, on the other hand, is considerably more aggressive with Turbo Mode, and that is one of the biggest reasons why it is so damn competitive with the Bloomfield. Lynnfield more than doubles the range of Turbo Mode multipliers from 2-5x, which means the new chips might run up to 667MHz faster than their default frequencies. That’s pretty damn impressive if you ask us.

The new Core i7s land within a stone's throw of Core i7 XE chips

The new Core i7s land within a stone's throw of Core i7 XE chips

What about pricing?

Now that you know the ins and outs of what makes Lynnfield different from its big brother, let’s take a look at the real meat’n’potatoes of the launch: Pricing.

One of the biggest criticisms leveled at the Core i7 was that its price increase was not commensurate with the performance increase it offered over the Core 2 Quad. Furthermore, aggressive clock scaling from the boys in green meant that the Phenom II X4 965 put the price/performance crown squarely on AMD’s head.

Has Lynnfield done anything to change that? You bet.

Pricing in USD (Newegg.com), GBP (Scan.co.uk) and Euro (Alternate.de)

Pricing in USD (Newegg.com), GBP (Scan.co.uk) and Euro (Alternate.de)

On average, the lowly Core i5 750 turns in 8-20% better performance than the Phenom II X4 965 for $40.00 less. The price/performance win is clear in this comparison. Restricting our net to the Core i7s, the new Core i7 860 is clear price/performance champ for anyone looking to buy into the Core-series platform.

All of these chips are available immediately, and in volume, which means you can go out and buy one right now if you so choose.

The Lynnfield’s Achilles Heel

Westmere is coming. Westmere is the 32nm successor to today’s Bloomfield CPUs, and it will debut exclusively for the LGA1366 socket. Not only do Westmere chips offer six physical cores to today’s four, it will run colder, faster, and overclock better than any Nehalem we have talked about today.

Users enticed by tantalizing price/performance offers made by the Core i5 must answer an essential question: Are you really interested in opting out of more, cooler, faster cores?

That’s the big question, and it’s not for us to decide. Enthusiasts will almost certainly see the merits of buying into LGA1366, because they know that past die shrinks like Conroe to Penryn made impressive thermal and performance gains. More traditional users may not care, because they’ll own the system well into 2011, and by then both AMD and Intel will have new architectures in the form of Bulldozer and Sandy Bridge, respectively.

All that said, knowing that LGA1156 was born with limited upgrade options is a bit sobering, and it’s a big point of annoyance with us. Offering users a platform with little to no future is not taken kindly, and we were similarly critical of AMD when it did the same thing with Socket 754.

Making sense of it all

There is no clear answer for users looking to buy into Nehalem today. The upper echelons of Lynnfield completely cannibalize any value the Core i7 920 once had, which means Bloomfield’s barrier entry has really been raised to the $569.99 price tag of the Core i7 950.

At the same time, the promise of Westmere’s year-end launch makes a compelling argument for the 920. We do not believe the i7 920 is long for this world, which means it might be a great idea to prepare for Westmere by buying into LGA1366 while it’s less than $500 for a good board and a chip. Wait too long and that price tag may float towards $800 if the 950 becomes the low-end model in the Bloomfield family.

Running on the assumption that the Core i7 950 will soon serve as the lowest model in the Bloomfield family, you really only have four choices:

  1. Buy the Core i7 920 and an X58 motherboard right now to prepare for Westmere on the cheap, knowing that a Lynnfield system would have been faster
  2. Buy into Core i7 950+X58 for performance that exceeds Lynnfield, at the expense of a hefty increase in price over a Lynnfield solution.
  3. Maximize your price/performance by purchasing a Core i5 750 and a P55, knowing that you’ve beaten Core 2 Quads and Phenom II X4s, but fallen short of all other Core i-series products.
  4. Buy a Core i7 860+P55 for respectable savings and price/performance in the short term, at the expense of significant future upgrades.

Only you have the answer to that, but we hope we’ve made it a little easier to get there. For more answers and a slate of benchmarks, keep your dial tuned to Icrontic for our upcoming shootout between Bloomfield, Lynnfield and the Phenom II X4.

Comments

  1. Leonardo
    Leonardo You have packaged a great amount of disparate data into one easily comprehensible article. Great content, good packaging. This is useful.
  2. Snarkasm
    Snarkasm Wonderful comparison, Rob. Pro.
  3. GooD
    GooD Great review,

    Im going i7 920, i don't care that much about power comsomation, and i like the fact that with the 920, you can achieve good overlock ON STOCK voltage, thats not the case anymore with the Lynnfield, plus i don't like the pci-e on the uncore thing... Having to overlock the pci at the same time of the cpu , higher voltage, this could cause problem in the long run IMO.

    All of this makes me think the 920 will be much easier to work with while overclocking. But's the new perfomance of the turbo boost v.2 makes my decision harder to take thats for sure.
  4. Cliff_Forster
    Cliff_Forster Outstanding read, very in depth.

    We could debate the need for all the confusion with another socket, but I won't ;*)
  5. Thrax
    Thrax No need to debate it: <a href=http://icrontic.com/news/understanding-intels-brand-name-shuffle target="_blank" rel="nofollow">It stinks</a>.
  6. Garg
    Garg Thorough and informative, thanks Rob! I've been purusing Newegg recently, so this is just what I needed to know.
    Thrax wrote:
    While the difference between 2×8 and 2×16 is immaterial at pedestrian resolutions like 1680×1050, 2×16 can be up to 7% faster at SLI/CF-friendly resolutions like 1920×1200 or 2560×1600.

    Something I've been wondering about for awhile!
  7. Thrax
    Thrax My personal opinion on this one:

    1. Anyone who currently owns a Bloomfield should simply overclock their chip, pair it with some good DDR3-1600 (or better), and watch the Lynnfield run away crying. The two chips may offer the same core frequencies and memory speeds, but that extra memory channel and wider PCIe bus will put Bloomfield ahead.

    2. Anyone looking to buy into the Nehalem should wait for Westmere. It makes the entire Lynnfield vs. Bloomfield debate obsolete: More cores, cooler cores, faster cores, and more overclockable cores will put Westmere ahead of both Bloomfield and Lynnfield, and it's only five months away.
  8. BuddyJ
    BuddyJ The biggest conflict I see isn't Lynnfield vs Bloomfield; it's Core i5 750 vs the Phenom II X4 Black Editions. The i5 750 has the Phenom II set square in its sights.
  9. GooD
    GooD Something i wonder with the read of the first reviews of the Lynnfield,

    I don't see many chart with overlocked temp for the Lynnfield, those cores takes a lot of juice (voltage) to overlock, wich should result in more heat generation in theory, they say it runs cooler than the 920, i wonder by how much when you go higher than 3.6 Ghz.

    Plus more juice should have an impact on the PCIe uncore if 24/7, PCI doesnt like to be pushed that hard i think.
  10. Thrax
    Thrax We'll let you know very soon. :) We have a Core i5 750 and a Core i7 860 on the way, and ECS has generously provided us with a P55H-A Black motherboard.
  11. GooD
    GooD Can't wait :)

    Thankx
  12. lordbean
    lordbean
    Thrax wrote:
    My personal opinion on this one:

    1. Anyone who currently owns a Bloomfield should simply overclock their chip, pair it with some good DDR3-1600 (or better), and watch the Lynnfield run away crying. The two chips may offer the same core frequencies and memory speeds, but that extra memory channel and wider PCIe bus will put Bloomfield ahead.

    2. Anyone looking to buy into the Nehalem should wait for Westmere. It makes the entire Lynnfield vs. Bloomfield debate obsolete: More cores, cooler cores, faster cores, and more overclockable cores will put Westmere ahead of both Bloomfield and Lynnfield, and it's only five months away.

    Based on your article and my experience with my Q9450, I'm in full agreement with this assessment. I see no need to upgrade to a Nehalem based system until Westmere launches. I run a single GTX 285 as my graphics, and my CPU is not a bottleneck. It might be if I had TWO 285s, but for a single one, it's more than enough.
  13. GooD
    GooD Problem is that i found a buyer for my actual rig, and i must buy one to sold mine before he change his mind lol :P And by the time westmere comes out my actual rig wont worth any money at all so its now or never i think (single core...)

    Otherwise i would wait for westmere to take the best decision, but thats why i'll chose and a 920, because when westmere will be out for a year or so, it will go down in price and then i'll switch with the same board (hopefully, if intel does change his mind and a bios update comes out to support westmere on my board) :P
  14. lordbean
    lordbean The X58 chipset still has plenty of untapped bandwidth to spare. Westmere shouldn't have any trouble in the 1366 socket.
  15. Zuntar
    Zuntar
    Buddy J wrote:
    The biggest conflict I see isn't Lynnfield vs Bloomfield;

    It's having to explain to my wife why I NEED to upgrade when Westmere gets here!!!:eek3:


    Snarkasm wrote:
    Wonderful comparison, Rob. Pro.

    ^this
  16. mirage
    mirage Thank you, Robert, for the time, effort, and thought. Enlightening!
  17. redchief
    redchief another outstanding analysis. Thanks
  18. Darkangel Thanks Robert for this very informative article!

    From the Philippines.
  19. roschler
    roschler The OCZ FAQ on DDR3 memory claims that modules running at speeds above 1600 MHz require an architecture involving intermediary register chips to avoid serious drains and bounces. Can anyone comment on this and if so, who has put out system components that implement such an architecture, without which I would assume that the ability to actually use the DDR3 memory at those speeds to be untenable?
  20. Thrax
    Thrax The maximum "stock" DDR3 speed supported by Bloomfields and Lynnfields is DDR3-1600. Some only support 1333 (i5 750). Bloomfield and Lynnfield are perfectly capable of running at DDR3-2000 speeds if you know what you're doing with overclocking, however. I'm not sure if that answers your question or not.
  21. roschler
    roschler
    Thrax wrote:
    The maximum "stock" DDR3 speed supported by Bloomfields and Lynnfields is DDR3-1600. Some only support 1333 (i5 750). Bloomfield and Lynnfield are perfectly capable of running at DDR3-2000 speeds if you know what you're doing with overclocking, however. I'm not sure if that answers your question or not.

    Hello Robert,

    I know that the 1600+ MHz speeds can be achieved by turning up the clock multiplier, but the OCZ FAQ has me wondering if those speeds are not really usable unless you have a specialized architecture that includes intermediary register chips; which is how it appears to read. If so, then unless I read about a motherboard/chip-set that incorporates intermediary register chips, which I have not so far, I'll have to assume that you while you can achieve those speeds you can't really use them. If the i5 and i7 motherboards and perhaps or other DDR3 compatible configurations on the market already incorporate such features then someone please comment.

    Here is the link to the OCZ DDR3 FAQ and I've quoted the relevant paragraph:

    http://www.ocztechnology.com/drivers/DDR3_faq.pdf
    "It is very likely that the entire scenario of broadcasting addresses and commands to every chip will become the limiting factor for the DDR3 performance if performance is defined as frequency. It is a simple equation in which the power that is used to drive addresses and commands across the bus will cause some serious drains and bounces above 1600 MHz and the only way to avoid this is to use registers as intermediate chips between the controller and the chips on the modules."
  22. Thrax
    Thrax That Q&A section specifically refers to registered modules:
    http://en.wikipedia.org/wiki/Registered_memory

    Desktop systems do not use registered memory, so it's a non-issue. Desktop systems can go up to DDR3-2133, and it's perfectly usable.
  23. Billy The Fish Sincere thanks for the last two paragraphs, in particular the 1 - 4 listings, which made more sense to me than anything I've read in all the PC mags in the last few weeks. I have now decided to go with an i5 build(your 'Option 3'), as I want the maximum bang for my gaming buck and I want it now!
  24. Thrax
    Thrax You're very welcome, sir. :) I'm glad Icrontic could help, and I'm glad you took the time to let us know! Take care.
  25. chizow
    chizow
    Thrax wrote:
    My personal opinion on this one:

    1. Anyone who currently owns a Bloomfield should simply overclock their chip, pair it with some good DDR3-1600 (or better), and watch the Lynnfield run away crying. The two chips may offer the same core frequencies and memory speeds, but that extra memory channel and wider PCIe bus will put Bloomfield ahead.

    2. Anyone looking to buy into the Nehalem should wait for Westmere. It makes the entire Lynnfield vs. Bloomfield debate obsolete: More cores, cooler cores, faster cores, and more overclockable cores will put Westmere ahead of both Bloomfield and Lynnfield, and it's only five months away.
    Nice overview of the differences, agree with #1 as that will essentially negate the artificial gains from an improved Turbo mode on Lynnfield. I probably would've broke out the PCIE lane and bandwidth advantage to a #3 though, as the disparity in bandwidth may become more of an issue with this next-generation of GPUs (Cypress and Fermi), particularly the X2 variants.

    As for #2, I don't agree with that based on current leaked roadmaps. 32nm Westmere is looking to be the most insignificant Tick cycle on Intel's Tick Tock mfg. cadence ever from a mainstream performance POV. Currently, it only has plans on the extreme high-end with the 6C Gulftown and the low-end with a 2C Clarkdale. There's no direct transition or die shrink planned for 4C Lynnfield and Bloomfield, so unless you want to spend $1000+ for the XE Gulftown, current 4C offerings are probably as good as its going to get until 32nm Sandy Bridge, which will be the Tock sometime in late 2010 or early 2011.

    My major problem with Intel's current socket fiasco is that the feature that was most responsible for bifurcating the high-end and mainstream sockets, tri-channel over dual-channel memory, offers the least benefit in actual performance. Its understandable why they went with QPI, for the additional bandwidth in the server storage and 2P market (which also gave gamers another 20 PCIE lanes), but I would've much rather had a single unified LGA1188 (figure the QPI is about 32 pins and traces, 240 pins for DDR3, can check Intel's whitepapers for exact pin specs) or whatever instead of the two platform Intel solution we have now.
  26. Andrew Great article. Thank you.
  27. Anonym This is some quality shit.

    Thank you for making my decisions much easier.
  28. Garg
  29. _k
  30. primesuspect
  31. Thrax
  32. Mystified Pete Due entirely to my own ignorance, not what is obviously great technical writing for those with even half a clue, I now have an alternative to Kant's Critique of Pure Reason to read when I just can't get to sleep at night.

    Both are brilliant and way over my head. At some point, I might get myself to the point where I could understand your content. That makes you the clear winner! Thanks.
  33. primesuspect
    primesuspect Wahahahaha that's awesome
  34. T2 Great Article! The best summary of the iX chaos I have ever read. I don't understand why can't I find this stuff on the Intel website. They must have an interest in keeping customers confused.

    Thank Rob!
  35. Deighton Warner Excellent
    You have cleared up some great confusion with this article Mr. Robert Hallock.
    I was tormented recently about whether I should buy the 920 or switch to the 860. They are the same price!
    Thanks again.
  36. Thrax
    Thrax You're both very welcome, and thank you for reading Icrontic. :)
  37. twilightrose For those about the Westmere being only XE 6C and low-end 2C market, Intel is releasing an 'update' to their low-end Nehalem 9xx series processors branded as i7-930 to replace the 920, which may sport the 32nm die shrink for the 4C medium level, we'll just have to wait and see!
  38. Thrax
    Thrax The 930 will not be 32nm. It's a 45nm part.
  39. Adnan Khan Does the i7 860 support three DDR3 1600 mhz memory or two?
  40. Thrax
  41. Adnan Khan Wow I love this site. You guys are awesomeo
    Ok incoming dumb question:

    Even though i7 860 supports 2 1600 Mhz memory, I can still throw in 2 more 1333 Mhz memory right?

    Like I can have something like 2 x 2GB DDR3 1600 and 2 X 2GB 1333 DDR3 for a grand total of 8 GB.

    I know I am dumb when it comes to this.

    Thanks
  42. Thrax
    Thrax Yes, you can, but keep in mind that the 8GB of memory will work at the slowest memory speed, or DDR3-1333.
  43. Mr. E Great info, thanks!
  44. Wavshrdr Great write up! I just stumbled across this site while looking for info between the 860 and 920 as Core 2 quad just died. Now I have to make a quick decision and buy something to replace it. After reading this an already difficult choice has been made more difficult now that I understand the differences better. Keep up the good work!
  45. Higherstate Wasted years reading industry PC magazines and slowly walk away shaking and scratching my head. Most clearly, concise article I have read for an age. Clarity desperately needed in this industry.
  46. ProDigit the corei7 950's are sold (Christmas 2010) for $289; a good bump down of the original 500 something!

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!