Haswell-E reveal: 8 Cores, DDR4, X99

Discussion in 'Intel Processors' started by phrozenfayte, Jun 15, 2013.

  1. drescherjm

    drescherjm [H]ardForum Junkie

    Messages:
    14,442
    Joined:
    Nov 19, 2008
    No but there will only be 4 dimm slots. And DDR4 will ship with a 16GB max per unbuffered dimm at release date so 64GB will be the max at least at the start.
     
  2. Sunin

    Sunin [H]ard|DCer of the Month - August 2008

    Messages:
    3,421
    Joined:
    Dec 27, 2005
    And that will most likely cost about what 800?
     
  3. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    I'm hoping that there will be 8 DIMM motherboards and that the 1 DPC that has been mentioned thus far is the quoted speed for 1 DPC, with 2 DPC possible, but at a lower speed. But an artificial limitation to limit memory capacity on Haswell-E by locking Haswell-E to 1 DPC fits with typical Intel behavior, unfortunately...wouldn't want those new overclockable 8 cores to start competing with the Xeons by allowing them to have 128GB of memory, now would we...:rolleyes:
     
  4. Bluesun311

    Bluesun311 2[H]4U

    Messages:
    2,523
    Joined:
    Sep 21, 2013
    Is anyone else disappointed by 40 PICE lanes? Maybe this is stupid but I was expecting 48... Don't we figure full Maxwell is going to need 16x? I had my eyes set on going TriSli this year/next year.
     
  5. RanceJustice

    RanceJustice [H]ardness Supreme

    Messages:
    5,991
    Joined:
    Jun 9, 2003
    I too thought there would be a maximum amount of PCI-E lanes on X99 - that's one reason to buy the platform, I thought!

    As far as Maxwell is concerned, will it really take up a full PCI-E 3.0 x16 bandwidth? I was under the impression that even the most powerful cards of the day were not able to saturate it enough to actually require no less than such a slot., especially with things like dual-GPU On Single Card (290x X2, GTX 790 etc..), and would not for quite some time. If Nvidia was one generation away from requiring a single PCI-E 3.0 x16 slot on their high end single GPU card, I'd imagine there would have been a huge amount of push for them to roll out PCI-E 4.0 or some such, which doesn't seem to be coming anytime soon - final specs, last I heard, won't be even released until late this year at the earliest?
     
  6. cennis

    cennis Limp Gawd

    Messages:
    375
    Joined:
    Aug 15, 2013

    http://forums.guru3d.com/showthread.php?t=375347
    at high resolutions a 680 shows improvement between PCI-E 2.0 x8 vs PCI-E 3.0 x8
    if the highend Maxwells are twice as fast as 680 which is likely, it may make a similar situation with PCI-E 3.0 8x (equal to PCIE 2.0 16x) vs PCIE 3.0 16x
     
  7. Zomoa

    Zomoa Limp Gawd

    Messages:
    242
    Joined:
    Nov 28, 2011
    Saw this recently:

    [​IMG]


    If true, this is awesome. We might finally be able to realize the benefits of the smaller architechture cooling / OC wise. It's silly that my 3770k takes as much to cool as my i7 970, despite the 970 pulling 3x as much wattage under OC load...simply due to Intel cheaping out on the thermal solution recently.


    Edit: found a video of the GDC unveil. Actually a really really good watch if you have the time. The presenter addresses our concerns directly. A guy asks a question about the TIM and tries to dig for super-specifics, but the presenter said we won't see super specifics till closer to launch. Given the hype as a selling point of the part, I'm hoping a direct metal contact solution like solder, or better than solder, isn't out of the question... http://www.youtube.com/watch?v=-SucnCc2te4

    Not sure if there is a higher quality recording out there. I'm lazy.
     
    Last edited: Apr 2, 2014
  8. OrangeBic

    OrangeBic [H]Lite

    Messages:
    124
    Joined:
    Dec 5, 2011
  9. iTune

    iTune n00b

    Messages:
    1
    Joined:
    Aug 30, 2013
    Intel Core i3-4330 Dual Core Processor 3.5 2 NA (BX80646I34330)

    What is 2 NA in the description?
     
  10. Bluesun311

    Bluesun311 2[H]4U

    Messages:
    2,523
    Joined:
    Sep 21, 2013
    Can't help wondering about the Z97 mobos... Asus Maximus VII Hero will be the go to guy again?

    Do I have this right: current Haswell procs will be compatible with 9-series chipset?

    But Devils Canyon will not be compatible with z87 (this I know)

    I'm ready for a new motherboard. It would be nice if EVGA would get their shit together on a properly spaced ATX dual SLI z97 board.
     
  11. ep0x73

    ep0x73 2[H]4U

    Messages:
    2,563
    Joined:
    Sep 5, 2013
    Most likely North America, could be wrong though and speculating.

    Many global manufacturers have different products for different markets.

    Generally Europe, Asia and NA.
     
  12. silk186

    silk186 [H]ard|Gawd

    Messages:
    1,601
    Joined:
    Feb 26, 2008
    I'm surprised by the lack of buzz. This is the most significant release in years.
     
  13. chrcoluk

    chrcoluk [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Jul 7, 2008
    so intel degraded the TIM so they can make money later by reupgrading it again :D brilliant.

    Do i feel bad about buying a normal haswell now? probably not as I expect these chips and boards will be way out my budget anyway.
     
  14. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    I'll speak for myself only and say It's tough to get excited about a quad core, regardless of how much it's warmed over...:rolleyes:

    As for the 8-core annoucement, this is a product that Intel had over two years ago. They had an awesome product then (E5-2687W) that they ignorantly chose to hard lock. I'm not going to get excited and cheer for a two year old product, which is what this product amounts to, even considering improvements of the Haswell architecture. I should be excited that Intel has finally decided to give us an overclockable 8-core, 2.5 years after they could have easily done so?

    When Intel releases something truly [H] worthy, I'll be the first one to get excited...;)
     
    Last edited: Apr 3, 2014
  15. DejaWiz

    DejaWiz Oracle of Unfortunate Truths

    Messages:
    18,940
    Joined:
    Apr 15, 2005
    E5-2678W = 3.1/3.8 GHz @ 150W TDP, business/enterprise segment targets. $1800+.

    These = 130-140W TDP, likely higher stock base and turbo clocks, unlocked, PCIe 3.0 plus a wealth of other improvements or additions, and meant for the consumer segment. Probably no more than $1000.

    In essence, I look at it as being able to get a two year old Xeon 8 core with more IPC, less TDP, upgraded capabilities, and about half the price.
     
  16. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    The sample Haswell-E pic that was passed around there around Christmas had a clock speed of 3.0 GHz on its IHS. May not be the final speed, but should be a ballpark figure. I can't imagine the max turbo being much higher than 3.8-3.9 GHz. All in all, the clock speeds between Haswell-E and the E5-2687W should be similar. Personally, I think the Haswell-E 8-core will be very similar to the E5-1680V2 in clock speeds.

    The E5-2687W also natively supports PCI-E 3.0. The "wealth" of other improvements or additions that'll come with Haswell-E are modest at best and are mainly incremental. An added "feature" of Haswell-E that the E5-2687W doesn't have is the restriction to 1 DPC with Haswell-E (a very significant drawback for some).

    The price for the top Haswell-E 8-core is still up for debate, but personally, I believe it'll be no less than $1499. Time will tell, I guess. The only chip that may have merit price-wise is the neutered 8-core, but I wouldn't expect to see it until next year (at the earliest).

    While the E5-2687W does wear a TDP label of 150W, I've found that in real world usage, it runs incredibly cool. The dies that Intel will be using for Haswell-E will doubtlessly NOT be the best dies that Intel will producing, with the best dies instead being used for Xeons. Haswell-E will almost certainly be made of 8-core dies that are the least desirable (but still fully functional) examples. I'd guess that the temperature of Haswell-E and the E5-2687W should be similar when running at the same clock speed, but in all honesty, it wouldn't surprise me to see the E5-2687W run cooler, especially under load, given its silicon pedigree and what is sure to be Haswell-E's lack thereof.

    All in all, I think Haswell-E is finally providing a chip to enthusiasts that a near equivalent to the E5-2687W, but is providing it 2.5 years afterward. The only real improvement that Haswell-E offers, IMO, is an unlocked multiplier and BCLK, a feature that Intel could have easily bestowed upon the E5-2687W 2.5 years ago with the flick of a switch.

    Bottom line: Haswell-E is a product being provided far too late for me to care too much about.
     
  17. Zomoa

    Zomoa Limp Gawd

    Messages:
    242
    Joined:
    Nov 28, 2011
    It has all the features intel brought with ivy and haswell.

    It's not a generation behind like previous "high end" intel chips. It will be the current refresh, up to date.

    If intel isn't lying, it should overclock better than current haswell chips.

    It has DDR4.

    It's back to what EE chips were like in the gulftown days. If you're the target audience, this chip is actually gonna be great. If you're not able to make use of all the features, sure, it wont be as cool for you.

    For someone like me that likes to do extremely heavy video encoding tasks, the prospect of running 8 haswell cores @ 5ghz under water is going to be a huge leap in performance over what is currently available at normal consumer prices.

    Given the locked nature of server chips, it's not really a good idea for me to pay 2-3x more for something that will probably perform slower than the new x99 platform for what I need. DDR3 -> DDR4 is going to be 10% or more alone for what i do. AVX2 another 10%. Add on the IPC improvements and the core count improvement and you see why the hype is real for some of us.
     
  18. DejaWiz

    DejaWiz Oracle of Unfortunate Truths

    Messages:
    18,940
    Joined:
    Apr 15, 2005
    I didn't realize the E5 SB Xeon supported PCI-E 3.0, thought it was limited to 2.0 as the consumer SB is.
     
  19. SpeedyVV

    SpeedyVV [H]ardness Supreme

    Messages:
    4,212
    Joined:
    Sep 14, 2007
    Actually really meh about this annoucement.

    Was getting all excited about that Intel Engineer's poll about a really high end future chip for enthusiasts.

    Something along 12 cores or sonething like that. Oh well.
     
    Last edited: Apr 3, 2014
  20. RamonGTP

    RamonGTP [H]ardness Supreme

    Messages:
    7,714
    Joined:
    Nov 9, 2005
    I suspect you're the type of person who always sees the glass half empty.
     
  21. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    Valid point.

    I'm not so optimistic on that. I'm guessing 4.5-4.6GHz will be the norm under water.

    Not necessarily a good thing...DDR4's gonna be stupid expensive for mediocre kits until production ramps up early next year. I question the need when memory bandwidth isn't the constraining factor for performance. DDR4 may even decrease performance, as each iteration of successive memory has done before (the initial, slow-clocked DDR3 sticks were slower than the fast DDR2 sticks, due mainly to increased latency, IIRC).

    I must disagree with this statement. Gulftown was a fully fledged chip, equal to the DP Xeons in core count, most features and size. The Xeon single processor chips were awesome...unlocked and highly overclockable and the only neutered part was the snipped QPI link. Also, if one wanted, the DP Xeons could be used instead on X58 boards, and overclocked to the limits of the CPU. The only dies that Intel made at the time that were superior were Beckton and Westmere-EX, both chips that were never made into LGA1366 variants.

    Today, we have three tiers of Xeon DP dies, with a high likelihood that this strategy will continue into Haswell-E/EP. The i7 gets the smallest slated die size, and probably the leakiest dies that are produced (with the better ones becoming lower-leakage Xeons). For me to consider your statement as correct, we'd have to be able to take the best Haswell-EP chips (14 core), pop them into an X99 board and then overclock them to our hearts content or have an equivalent EE chip (neither of which is going to happen).

    Haswell-E should perform well, although clock-for-clock, although I'm not convinced it'll be more than 10% faster in most tasks than the E5-2687W is. Again, I'm not optimistic on these chips hitting 5 GHz on normal water, and think they'll be in the 4.5-4.6GHz range for the good ones. I could be very wrong...that's just my hunch. Overclocked to this speed, Haswell-E will no doubt be strong, and we'll finally get to see an overclocked 8-core in action, albeit far later than we should have seen one.

    The frustrating thing is that there is little justification in Intel locking the top SKUs of Xeon. If they were unlocked, it may well be worth it to buy one for the 2-3x the price you mentioned when one factors in the extra performance that could be teased out of such a high-quality, high-core count CPU. Intel already unlocks the single processor Xeons, so refusing to unlock the top SKUed DP Xeons is quite simply a petty and purile policy. No one at Intel seems to have the courage to try something different by unlocking the top SKUs. If they were to try unlocking the top SKUs, it just might result in higher sales of their most profitable products, with virtually no risk or issue.
     
  22. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    You and I both. However, Intel has occasionally surprised us before with unexpected SKUs during CPU launches. I'm still hanging onto a sliver of hope that Intel may release something along these lines when Haswell-E drops, or decides to dispose of leftover Ivy-EP HCC Xeon dies in this fashion this summer after they ramp up the production of Haswell-E/EP. I must admit though that either act is highly unlikely considering Intel's recalcitrance toward offering a truly top drawer enthusiast CPU. Unfortunately, all we can do at the moment is hope...and continue to keep letting Intel know that we want such a chip...badly.
     
    Last edited: Apr 3, 2014
  23. Zomoa

    Zomoa Limp Gawd

    Messages:
    242
    Joined:
    Nov 28, 2011
    I get where you guys are coming from and agree with you. Intel is holding back, and getting excited over an 8-core CPU when 15-core CPUs are already mainstream server CPUs seems kinda silly.

    I'm just glad we're not getting the SB-E/IB-E treatment again. There's no reason we shouldn't expect a core-count increase every generation, given intel already has the silicon with server processors. It'll be good to have an upgrade that is significantly better than the previous generation, rather than the incremental crap we've been getting. Sure, it's not as big of a jump as they could give us..not even close. But it's getting better.

    Intel has the capability to give us more than 8-cores, why do the engineer poll if they don't. I'd gladly pay 2.5k for an unlocked 12-core today. Give it the GTX Titan treatment. It's a ton of money, but I wouldn't feel salty about the purchase at all, as it would be a huge leap in power.

    Here's hoping for a surprise [H] edition SKU :cool:
     
  24. drescherjm

    drescherjm [H]ardForum Junkie

    Messages:
    14,442
    Joined:
    Nov 19, 2008
    I believe this is only because the 14nm process was delayed.
     
    Last edited: Apr 4, 2014
  25. chrcoluk

    chrcoluk [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Jul 7, 2008
    Well consider this.

    90% of desktop software is able to saturate single core utilisation (meaning single core perfornmance is important), but not much software will scale to multi cores to saturate it, yes we have video encoding but that is a small part of the market.

    PC games wont be optimised for 8 cores, not a chance, they made for mainstream hardware. $1000 cpus are not mainstream. Games will be optimised for 2-4 cores or more likely single core than 8 core.
     
  26. drescherjm

    drescherjm [H]ardForum Junkie

    Messages:
    14,442
    Joined:
    Nov 19, 2008
    Last edited: Apr 4, 2014
  27. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    I think we are getting the SB-E treatment again. While it's a good thing that we're finally being given an 8-core CPU, I highly doubt Intel is giving us this 8-core CPU out of any desire to be benevolent. I think with Haswell-EP, I think there will be more models with 8-cores or more than with fewer than 8-cores, due largely to the fact that folks buying mid to high-end servers have had 8-core options for over two years now. This mid to high-end market is likely becoming a bit saturated, and these folks are unlikely to upgrade to another 8-core, and instead are looking to a higher core count CPU for their next purchase. Intel is very likely less and less able to get the prices they obtained previously for an 8-core Xeon CPU, due to this saturation in the mid to high-end markets. This means that the 8-core Haswell-EP will be occupying the lower end of the server market at lower price points (more entry level), much like the 6-core Ivy-EP is now. Combine this reality with the fact that the i7-4960X was distinctly underwhelming to enthusiasts and likely sold much more poorly than the i7-3960X did makes it a logical progression to finally move to an 8-core i7. A native 8-core Haswell-EP die would be capable of servicing the lower end Xeon market, with the higher-leakage, poorer quality dies that are unsuitable for use as a Xeon being made into i7s. The bottom line is that we are only getting an 8-core CPU due to Intel's progressively increasing inability to sell 8-core dies as higher priced Xeons, not out of any desire by Intel to do the right thing by enthusiasts.

    I think the one thing that the GTX Titan has proven is that the highest end hardware sells quite well, even at a high price. Nice things cost money, and the folks who can afford these items know this. Intel stubbornly refuses to acknowledge this reality, but I'm hoping with continued success of the GTX Titan and other similar high-end hardware that they get their heads out of the sand and acquire the courage to try releasing some top shelf hardware of their own.

    $1000 GPUs aren't mainstream either. But they still sell very well, in the word of NVidia's CEO, "like hotcakes". The fact that the Titan Z has been announced at all is indisputable proof that high-end hardware sells. The GTX 780 Ti is the more powerful gaming card in most situations, yet the GTX Titan still sells very well. This is due to the fact that people use the Titan for more than just gaming. The same would be true for a 12-core unlocked, ultra high-end enthusiast CPU from Intel. Some folks would use them for gaming sure, but many would be sold for use in many other tasks as well, with these other tasks far exceeding the amount of gaming done on these CPUs. The thing is with this caliber of CPU is that the profit margins are obscene and are well worth the effort to make such a product available.

    I agree that the 15-core Xeon E7 would be way out of the budget of most people and than an enuthsiast-based 12-core makes the most sense fiscally. However, if Intel would simply unlock the top SKU of the Xeon E5, this situation would resolve itself immediately, with no new SKUs needed. While an unlocked E7 Xeon would be nice, the lack of suitable motherboards presents a considerable problem for this strategy to be a success. The best solution at present would be a Xeon E5 with 15-cores at the extreme top end of the Xeon E5 SKU stack at a price point appropriate to a 15-core CPU. Both this theoretical 15-core CPU would be unlocked, as would the 12-core below it. Buyers could then decide which model they want based on their needs and budget. The 15-core Xeon E5 would sell to server/workstation buyers, as well as the occasional ultra high-end enthusiast user.
     
  28. chrcoluk

    chrcoluk [H]ard|Gawd

    Messages:
    1,210
    Joined:
    Jul 7, 2008
    I am not disputing it sells enough (still be fraction of market tho), just dont think gaming software will be optimised for it.
     
  29. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    Gaming software won't take full advantage of the CPU, no, but an overclocked 12-core running at 4.5 GHz under good water will be just as fast as an i7-4770K running at the same speed, with the ability to do much, much more at the same time. One could launch video encoding and play a demanding game at the same time with little to no detriment to either task.

    If the chip was dual-capable, it would enable a complete monster to be built. With 24-cores overclocked to 4.5GHz, you could throw anything you wanted at it, and the system would largely shrug it off. A clock speed such as that is 50% higher than the default max turbo clock on the E5-2697V2...just imagine the increase in performance. The only apt word that readily comes to mind is "sickening". Just totally stupid levels of performance...:D

    It's an absolute crying shame we aren't permitted to have such a processor, not at any price...:(
     
  30. JaiWebb

    JaiWebb Gawd

    Messages:
    601
    Joined:
    Feb 8, 2014
    So when are these newer CPU's coming out? Is there an official release date, yet?

    (Only skimmed this thread but didn't see anything on it)


    Thanks
     
  31. gigatexal

    gigatexal [H]ardness Supreme

    Messages:
    7,237
    Joined:
    Jun 22, 2004
    yeah but who does that? triple monitor gaming with one displaying a movie, the other doing a compile and the center playing crysis?
     
  32. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    The best info available at present is Q3, likely in September.


    That was just an example of usage. The horsepower that such a system would provide could be used in many different ways. As for your specific usage scenario, I raise my hand. A single E5-2687W can do it, but the system I described would have much more power on tap to do it much quicker. A ton of threads running at a very high clock speed...the best of both worlds...;)
     
  33. gigatexal

    gigatexal [H]ardness Supreme

    Messages:
    7,237
    Joined:
    Jun 22, 2004
    Hah, I must admit, I did that on my dual titan rig, a movie on the right, SC2 on the mid, and the third monitor with afterburner running, and other diag screens showing temps and stuff still seems like a very niche thing to do
     
  34. SoulCreatr

    SoulCreatr Limp Gawd

    Messages:
    229
    Joined:
    Mar 3, 2008
    Any guesses on what the 6C version will cost?

    I would love to get a hold of 6C/12T haswell-e with DDR4 and pcie 3.0. Should let me game while also doing work through multiple Citrix applications that are required for remote login. Not to mention it won't drag me down doing those things while also having the internet browser open with several tabs, Pandora going, etc. My current machine chokes up when I get too many Citrix strings going, and god forbid I attempt to play a game at the same time.
     
  35. drescherjm

    drescherjm [H]ardForum Junkie

    Messages:
    14,442
    Joined:
    Nov 19, 2008
    ~$600
     
  36. Michaelius

    Michaelius [H]ardness Supreme

    Messages:
    4,684
    Joined:
    Sep 8, 2003
    Considering there's no 4 core version so far it should be cheaper than 600$.
     
  37. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    Tough to say for sure, but $600 sounds likely to me as well. Considering the flagship quad core will be near $400, a $200 premium for a 6-core doesn't sound at all implausible, especially considering the added features that X99 has over Z97.
     
  38. Michaelius

    Michaelius [H]ardness Supreme

    Messages:
    4,684
    Joined:
    Sep 8, 2003
    That really depends on how Intel wants to distinguish them.

    socket 1150 Haswell has bigger die than Ivy-Bridge-E due to size of gpu
     
  39. lutjens

    lutjens Gawd

    Messages:
    792
    Joined:
    Feb 18, 2013
    Sure, but Intel has a penchant for charging more whenever they can. Haswell-E (and Ivy-E before it) are marketed as a high-end desktop (so-called, IMHO) and the quad core 1150 Haswell is marketed as mainstream. Intel doesn't set their prices based on die production costs, they price based on the market for the CPU in question and the price that market will bear. Considering that Ivy-E has a smaller die than Haswell 1150, that simply means their profit margins are higher and the cost of production lower on Ivy-E. But the market for Ivy-E is significantly smaller than for Haswell 1150, hence the tradeoff. The increased volume of Haswell 1150 makes up for the smaller profit margin with that CPU.

    Considering the target market for Haswell-E, the question is whether or not Intel can charge more for it, and in this case I believe they can. Given the HEDT label on the platform and CPU, they'll most certainly feel justified in charging as much as they can for it. That being said, the price for the 6-core will definitely depend on the price of the 8-core(s) that Intel SKUs up. The crowding between the conventional $999 Extreme Edition price and the ~$600 price of the next chip down in the SKU stack (which will likely be a neutered 8-core) leads me to strongly believe that the new Haswell-E 8-core EE price will be in excess of $999 (I'm guessing $1399), with the neutered 8-core falling in and around the EE's old price point, but probably a bit less to avoid a direct comparision with the old $999 price...call it $899.

    Basically, I'm predicting a Haswell-E SKU stack something like this (prices + or - $100):

    $1399 -- i7 8-core Extreme Edition
    $899 -- i7 8-core Neutered Edition;) (16MB of L3 vs 20MB of the EE). I think this SKU will show up after the EE has been available for some time to "encourage" folks who want an 8-core to buck up for the EE...I'm guessing the end of the year-ish.
    $600 -- i7 6-core (my cache guess is 15MB to provide the illusion of an improvement over the i7-3930K/i7-4930K).

    Of course, the above is just my prognostication run amok...time will tell how accurate (or inaccurate it is)...;)
     
  40. Michaelius

    Michaelius [H]ardness Supreme

    Messages:
    4,684
    Joined:
    Sep 8, 2003
    Well it's up to them how much of my money they want :)

    If there's cheap 6 core sku i'm willing to buy socket 2011 mobo and cpu.
    If not i'll get haswell refresh or broadwell i5 together with $100-120 mobo :D