Intel Skylake / Kaby Lake / Coffee Lake Thread - Skylake-X reviews out (page 501)

Discussion in 'CPUs and Overclocking' started by Sweepr, Apr 16, 2015.

  1. nvgpu

    nvgpu Senior member

    Joined:
    Sep 12, 2014
    Messages:
    553
    Likes Received:
    130
    frozentundra123456 likes this.
  2. IEC

    IEC Lifer

    Joined:
    Jun 10, 2004
    Messages:
    12,497
    Likes Received:
    1,274
    Would be nice if people were more clear about what they are talking about, or at least move the server part discussion to its own thread. Talking about Skylake, Skylake-X, Skylake-SP, Kaby Lake, Kaby Lake-X, and Coffee Lake all in one thread can lead to a lack of clarity. I don't blame people for mixing them up, at all.
     
    Drazick and Gikaseixas like this.
  3. IEC

    IEC Lifer

    Joined:
    Jun 10, 2004
    Messages:
    12,497
    Likes Received:
    1,274
    Some back of the napkin calculations:

    Kaby Lake-S and Kaby Lake-X:
    7700K = 4c/8t, 4.2 base/4.5 turbo, iGPU, 91W TDP
    7740K = 4c/8t, 4.3 base/4.5 turbo, no iGPU, 112W TDP

    Theoretical 6c/12t Coffee Lake (best case assumptions):
    +50% vs 7700K = 136.5W at 4.2/4.5 assuming linear scaling
    +50% vs 7740K = 168.0W at 4.3/4.5 assuming linear scaling

    Split the difference = ~152.3W

    If we assume 14++ essentially improves efficiency enough to lead to a perfectly linear increase in TDP, then a target TDP of 140W-150W at 4.2/4.5 or even 4.3/4.5 clocks may realistically be achievable.

    5GHz is a different story.

    Stock voltages are typically 1.24V-1.25V on 7700Ks:
    1. https://www.bit-tech.net/hardware/2017/01/03/intel-core-i7-7700k-i5-7600k-review/8
    2. http://www.anandtech.com/show/10968...-review-the-new-stock-performance-champion/11

    Let's take an average of three different sites (AnandTech, Bit Tech, Silicon Lottery) for 5GHz clocks:
    1. (1.425V) http://www.anandtech.com/show/10968...-review-the-new-stock-performance-champion/11
    2. (1.36V) https://www.bit-tech.net/hardware/2017/01/03/intel-core-i7-7700k-i5-7600k-review/8
    3. (1.376V) https://siliconlottery.com/collections/all/products/7700k50g
    = 1.387V for 5GHz (will likely require more for 6c/12t, but again, best case assumption)

    Using a overly-simplified formula (reality is a bit worse than this), we can calculate the TDP of a hypothetical Coffee Lake 6c/12t processor at the typical voltages required for 5GHz for a 7700K.
    OC TDP = TDP * (OC MHz / Stock MHz) * (OC vCore / Stock vCore )^2

    This yields an estimate at the low end (using 140W base TDP) of:
    =140W * (5000/4300) * (1.387V/1.24V)^2
    =140W * 1.1628x from frequency * 1.2512x from voltage
    =203.7W*

    *These figures pretty much assume perfect execution and real gains from process improvement. Reality may be worse.
     
    Drazick likes this.
  4. krumme

    krumme Diamond Member

    Joined:
    Oct 9, 2009
    Messages:
    4,699
    Likes Received:
    650
    Good proposal. Its different segments. Segments not brands should be the determinant.

    The idea that intel or amd thread should stay "clean" is a disaster of thinking.
    Eg a 7700 and 1700 cost the same so comparing them is straightforward.
    The thought that eg 7700 or 1700 must not be mention in a discussion is imo straightforward fascist thinking. Its a very dangerous road to take and it takes away excactly what is the value of discussing.
    I am not going to bend for "clean" threads with a brand name on a box. A name.
    Never.
     
  5. xdfg

    xdfg Member

    Joined:
    Mar 6, 2017
    Messages:
    25
    Likes Received:
    5
    .



    Calling enthusiasts, "fanboys", is not allowed.


    esquared
    Anandtech Forum Director
     
    #10255 xdfg, Apr 29, 2017
    Last edited: May 31, 2017
  6. mikk

    mikk Platinum Member

    Joined:
    May 15, 2012
    Messages:
    2,017
    Likes Received:
    237

    You are the best example of a disaster. This guy is trolling around and a day or two later he will remove his nonsense. It's a typical AMD hardcore fanboy, a disaster.


    [​IMG]




    Calling enthusiasts, "fanboys", is not allowed.


    esquared
    Anandtech Forum Director
     
    #10256 mikk, Apr 29, 2017
    Last edited by a moderator: Apr 29, 2017
    jwapk, psolord, CHADBOGA and 2 others like this.
  7. xdfg

    xdfg Member

    Joined:
    Mar 6, 2017
    Messages:
    25
    Likes Received:
    5
    Instead of shooting the messenger, why not refute my facts if you think they are incorrect? The numbers were posted just two pages ago by Markfw. It's not like any of this is based on speculation - you can rent a Skylake server today and discover the IPC for yourself!

    Look, I understand where you guys are coming from. You burned yourselves hard putting all that cash in overpriced Intel CPUs and AMD comes out of nowhere and blows them out of the water. Rationalization and denialism are natural instincts, but you have to look at the facts if you want to advance the conversation. There's just no evidence that Skylake-X and Coffee Lake will meet these ridiculous expectations that keep getting thrown around in this thread.
     
    #10257 xdfg, Apr 29, 2017
    Last edited: Apr 29, 2017
  8. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,472
    Likes Received:
    707
    Its in the article, and even in the URL, duhhhh

    Everything I posted was about that article as the source.
     
    #10258 Markfw, Apr 29, 2017
    Last edited: Apr 29, 2017
  9. Shivansps

    Shivansps Golden Member

    Joined:
    Sep 11, 2013
    Messages:
    1,790
    Likes Received:
    113
    hang on, the 6950X is 140W @ 3.4Ghz ACT on the crappy 14nm process, and that one has the full L2 8-way, quad channel and 10 cores.

    Not sure how much higher they are willing to push stock TDP on 1151, if they want to stick to 91W well there is going to be a problem with stock freq, if they are going to 125W its possible there gonna be compatibilities issues with 100 and 200 mbs. New 300 mbs are going to be non-issue.
     
  10. Ajay

    Ajay Diamond Member

    Joined:
    Jan 8, 2001
    Messages:
    3,526
    Likes Received:
    491
    One more Jonny come lately for the ignore bin. Crapy lake - seriously o_O
     
  11. LTC8K6

    LTC8K6 Lifer

    Joined:
    Mar 10, 2004
    Messages:
    25,073
    Likes Received:
    443
    It's amazing.

    AMD finally produces a reasonably good core after 5 or so years, and somehow this means that Intel doesn't know how to make and sell chips anymore?
     
    Ajay, psolord, Drazick and 4 others like this.
  12. DrMrLordX

    DrMrLordX Diamond Member

    Joined:
    Apr 27, 2000
    Messages:
    8,327
    Likes Received:
    481
    Possibly nothing, from a technical point of view. From a market perspective, it means that Intel had Kabylake out as its flagship consumer CPU for only . . . 6 months? Kaby had no time in the spotlight. It may cause inventory/supply problems. Even Devil's Canyon had more time to shine than that, and it was less of an update than Kaby.

    Um, well I don't think it's logical to conclude that Intel has forgotten anything. What we're seeing is Intel laboring to steer a ship with a wide turn radius. It'll take time for them to fix things. They may not have everything sorted out properly on the desktop and/or HEDT side until Icelake hits the scene.

    Also, pushing Coffeelake and Skylake-X out to market sooner does not necessarily mean that they'll be any quicker to launch Icelake. If we're lucky, maybe Icelake will move up by 6 months; otherwise, Intel will have a fairly stagnant desktop/HEDT product stack until Icelake finally hits the streets.
     
    #10262 DrMrLordX, Apr 29, 2017
    Last edited: Apr 29, 2017
    Drazick likes this.
  13. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,472
    Likes Received:
    707
    So, what do you think of tweaktowns comment that Intel is scared of the new naples processors ? And that the launches its doing early are trying to save their bacon ?
     
  14. LTC8K6

    LTC8K6 Lifer

    Joined:
    Mar 10, 2004
    Messages:
    25,073
    Likes Received:
    443
    Exactly. It's silly to think Intel won't respond strongly if and when they need to.
     
    Drazick likes this.
  15. LTC8K6

    LTC8K6 Lifer

    Joined:
    Mar 10, 2004
    Messages:
    25,073
    Likes Received:
    443
    I think that's silly, too.

    It's weird. For a long time we have had a lack of competition with CPUs and much complaining about that lack.

    Now we have some competition and people are acting like they hope Intel can't compete anymore, thus ending the competition they wanted so bad.

    Personally, I hope Intel crushes RyZen soon, forcing AMD to bring out RyZen2 much faster, and so on.
     
    thepaleobiker, Drazick and ozzy702 like this.
  16. xdfg

    xdfg Member

    Joined:
    Mar 6, 2017
    Messages:
    25
    Likes Received:
    5
    Intel is definitely scared of Zen and Naples, but SKL-X and CFL have nothing to do with it, since they aren't targeting the datacenter market. The problem for Intel is that they already launched SKL-SP last year and they have nothing left they can bring out until Ice Lake in 2019 or later. Meanwhile, all the businesses that got fucked over by Intel and their anti-competitive deal with Facebook to delay SKL-SP for the general market, are going to be looking very closely at any possible alternative vendor.

    What we're going to see is AMD take over the client market with their superior value and slowly encroach on Intel's Xeon market. Even with Ryzen eating their lunch twice over, Intel is still playing their segmentation game, trying to drop gimped crap like SKL-X in HEDT. Customers can tell when they're being ripped off, and that's why they're pretty much all going with AMD, since they don't gimp their chips to protect margins.
     
    #10266 xdfg, Apr 30, 2017
    Last edited: Apr 30, 2017
    misuspita and Azuma Hazuki like this.
  17. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,472
    Likes Received:
    707
    Well, if Ryzen continues to be a success, it does not mean Intel failed, it just means they may have to lower prices, and maybe make a little more progress than they have in the last 10 years while AMD was trying to catch up.

    Wouldn't you like to see them do something like when conroe came out and crushed AMD 10 years ago ? I see it as a see-saw, and AMD is on top for a little while now, THATS what sparks innovation. How long they are on top depends on both sides, but for the moment, its Intels turn to innovate.
     
    Gikaseixas, Drazick and Azuma Hazuki like this.
  18. itsmydamnation

    itsmydamnation Golden Member

    Joined:
    Feb 6, 2011
    Messages:
    1,516
    Likes Received:
    460
    Its silly to think Intel can respond strongly instantly, they can't without giving something up and that something is revenue.

    Intel's issue is that within the space of 6 months, (march 17 to sept 17) a company with 1/10th its operating cost and 1/20th it R&D and admin cost will go from completely noncompetitive to completely competitive from 15 watt ultrabook all the way to 200watt 32core 2P servers. Thats probably minimum 50 billion of intels 2016 revenue number thats at threat.

    Given the lead times it takes to get things done in the CPU space you cant react instantly and bring out all concurring tech, which means Intel has a fine rope to walk as they want to maximize revenue.


    I think a lot of people want AMD to have a strong financial period so they can consolidate. Your idea of intel/amd responding to each other doesn't allign with the real world ( outside a clock speed race). If you are waiting for your competitor to have the advantage before you react then its a long road before your response is complete. If intel has something that can smash Zen then it as already been in the works for 3-4 years, if AMD have something that can smash the chip that smashed Zen it has already been in the works for 3-4 years..........
     
    Gikaseixas, Markfw and .vodka like this.
  19. xdfg

    xdfg Member

    Joined:
    Mar 6, 2017
    Messages:
    25
    Likes Received:
    5
    The thing about Intel's "response" with the "rushed" Skylake-X is that it's only possible because Intel has been sitting on SKL-SP for almost a year (anti-competitive agreement with Facebook to hold back the market!). Sure, with 12 cores, it might be able to outperform AMD R7 (won't be cheap), but it's a play they can only make once. After that, they have no more tricks until Ice Lake-SP, and that's not going to be until 2019 at the earliest. Meanwhile, AMD using more scalable and efficient MCM technology can make their chips as big as they want, forcing Intel to keep trickling down their $$$ Xeon line and discounting it.
     
    #10269 xdfg, Apr 30, 2017
    Last edited: Apr 30, 2017
  20. Shivansps

    Shivansps Golden Member

    Joined:
    Sep 11, 2013
    Messages:
    1,790
    Likes Received:
    113
    No, is not that, i lived though Athlon XP, Athlon 64, Conroe, Nehalem, Barcelona, Sandy Bridge, Bulldozer eras and ive never seem anything like this, you cant even say anything remotely negative about ANY AMD product, and is not only here, same happen with GPUs. Maybe is because im getting old but i dont remember this to have happen before. Well actually BD launch was a little bit like this, but it was way too poor to support it.

    Anyway, going back to the silicon, this is actually a less damage than A64 did back in its day, Intel can counter with the stuff in the pipeline THIS YEAR, it may be not enoght to fully beat Ryzen across the board at price/performance, but im sure it will be close enoght so nothing will change, and that is bad for us.
     
    #10270 Shivansps, Apr 30, 2017
    Last edited: Apr 30, 2017
  21. DrMrLordX

    DrMrLordX Diamond Member

    Joined:
    Apr 27, 2000
    Messages:
    8,327
    Likes Received:
    481
    Naples is a bit of an unknown to me. In HPC applications (whether dGPUs/compute cards are involved or not) I think Intel will still have the stronger product stack, but in everything else I think Naples is a potential threat. We haven't even seen Purley yet and AMD threatens to release 32c/64t CPUs? Based on what we've seen from Summit Ridge, AMD can make some hay anywhere that AVX-512 is a non-issue. There are lots of people looking for non-Intel alternatives at least as a leverage point to talk Intel down on price.

    Zen in general is strong enough that I do not think server buyers will be upset with its IPC, which is/was an issue with anything Con-based.

    Intel's best counter may be to hit back with low clockspeed, low power Cannonlake-based products, but that would require shifting the product stack and moving up launch. There's the possibility that they could throw together a Xeon-D product based on Cannonlake that will hit Naples from below while Purley hits it from above. It'll be hard to do that since I expect Naples to be clocked in the 2.5-3.2 GHz range where it will be much more efficient than something like an 1800x.

    I think it's more that Intel will offer a strong response when they're able. The next "killer" products I expect from Intel will be a Cannonlake-based Xeon-D product or . . . Icelake?
     
    Drazick likes this.
  22. Shivansps

    Shivansps Golden Member

    Joined:
    Sep 11, 2013
    Messages:
    1,790
    Likes Received:
    113
    Is really Naples that much of a problem to Intel when Naples is 2P, and LGA3647 has 2S, 4S and 8S configurations? Intel biggest server could be a 224C/448T one.

    Idont know i think we need to wait and see the Skylake-X/SP core performance.
     
  23. itsmydamnation

    itsmydamnation Golden Member

    Joined:
    Feb 6, 2011
    Messages:
    1,516
    Likes Received:
    460
    I've been a PC enthusistis since the family got its first PC a 486 DX2-66 and i dont see it anywhere near how you are seeing it. The actual problem is that outside of 7700K neither broadwell or kaby lake good good against zen when you consider all metrics. Then with the 7700K your fighting the current that game development is only heading in one direction and that is to use more threads, so its winning battles but its going to lose the war.

    In a way what you have is almost the complete opposite of the BD release where only the 8150 looked in anyway like a 1/2 ok product against sandy bridge, Do we need to go and dig up posts from anandtech from that era?

    What are you talking about, It has already changed for us! 6 and 8 core high IPC high clock processors for under 320USD, Also remember X399 is to come as well with upto 16cores.

    You also completely ignored the market situation and then just handwaved "intel will be better".......

    You are also aware that cannonlake is still just skylake cores so until 2019 what exactly are intel going to do other then clock speed? when icelake comes out it will be fighting against a 2nd or 3rd iteration of Zen both of which will have uarch improvements.


    Most of the market is 1P and 2P, there are very few system thats are 4P and 8P are in the really high end boutique SGI world. The cost for those 4P and 8P systems are extreme and the overheads for memory performance are huge, you will only find ultra extensive, ultra highend applications running on them.

    Naples has two or three massive advantages,

    1. the 4 small die approach is good form a yield perspective but is even better from a transistor variance perspective, this means AMD can bin power efficient/high performance/balanced zeppelins really well and then package them together.
    2. The amount of PCIe lanes, 1P has 128 PCIe v3 lanes, 2P has 64 PCIe lanes a proc ( 128 total) , will be great platform for AMD GPGPU, NV(even better then power) as well as knights-*. All those PCIe lanes are also really great for all flash hyper converged systems.
    3. On SOC crypto accelerators, we will have to wait and see if skylake-X/SP has them but Zepplin does and if skylake-SP doesn't when exposed this will be a massive processing and power advantage for anything TLS.

    Other interesting features are:
    8x 3200mhz DDR4 support ( skylake-SP is said to be 2666 and has only 6 channels).
    AMD SVE which allows for each virtual machine in a hypervisor to have its memory encrypted in hardware and transparently to the Guest and the keys aren't exposed to the hypervisor.
    Zen will also have more cores and will be way easier for amd to make 32 core chips then it will be for intel to make 30 core chips.

    Rome wasn't built in a day, but AMD have done everything they needed to, to take significant market share form intel in all markets ( do we really need to compare what Vega IP will look like against intle GFX?).

    If intel choose not to fight on price you can't cry that nothings changed because you only want to buy intel.
     
    .vodka likes this.
  24. imported_ats

    imported_ats Senior member

    Joined:
    Mar 21, 2008
    Messages:
    399
    Likes Received:
    52
    You really don't understand what actually goes into server silicon post validation. Intel was able to roll out chips early for a small number of set fixed configurations. This has always been true, and has happened with numerous other server chips from numerous vendors, usually in the HPC space. Getting the chip from 1 or 2 stable set configurations to supporting the full set of possible configurations takes a significant amount of work and time.

    And no, AMD is not using a more salable or efficient MCM technology. MCMs inherently are less scalable and less efficient than monolithic dies. And no, AMD cannot make their chips as big as they want, there are logical and architectural restrictions on how many dies you can put on a MCM. MCM is a solution AMD went down because they couldn't afford to do large monolithic designs, not because anything about MCM is inherently better.
     
    CHADBOGA likes this.
  25. itsmydamnation

    itsmydamnation Golden Member

    Joined:
    Feb 6, 2011
    Messages:
    1,516
    Likes Received:
    460
    There are plenty of things that are inherently better

    1. Transistor variance across a smaller die will be better then a monolithic die
    2. Yeilds on small dies are better then big dies
    2. if you look at zepplin you will see how small the GMI phy's are, there is no way you can make a 32core, 8 channel memory , 128 PCIe, 8x crypto coprocessor, 16x 10Gb/e, 16x USB3.1 monolithic processor in 600mm sq.
     
    Ajay likes this.