Previous Deceit in Technology *SPAWN*

Discussion in 'Politics & Ethics of Technology' started by DavidC, Jan 20, 2015.

  1. DavidC

    Regular

    Joined:
    Sep 26, 2006
    Messages:
    348
    Likes Received:
    27
    Few of you guys at Beyond3D will remember this slide(this is back when they had Sandy Bridge Ultrabooks): [​IMG]

    Footnote says: "Projections based on 2013 platforms". Obviously they did not meet that, or "2013" really meant "project finished in 2013" or Broadwell, which got delayed again. Nevertheless, they failed on the graphics improvement. Or, you have to take in terms of getting only few of the features promised rather than all, which is frankly, misleading. I remember about the graphics they were touting something equal to performance of a "high-end" Notebook GPU. But let's ignore that and give them slack. What about Core M?

    Look at a few of these links and decide for yourself.
    -http://www.windowscentral.com/lenovo-silently-reduces-yoga-3-pro-lowers-price

    -How does $1299 pricing make sense? Notice its not just one model failure: http://www.ultrabookreview.com/5868-asus-zenbook-ux305-review/
    1.66 points in Cinebench R11.5? Really? Starting at $700? Atom Z3795 gets 1.6 points. So the point of paying for a $200+ CPU is that you get lower battery life, thicker system, but better single threaded performance? Is that worth it to you? Why not get a Haswell system if even "thinness" isn't guaranteed?

    -Intel claims: http://www.tabtech.de/windows/intel-core-m-50-mehr-leistung-40-mehr-gpu-performance-fuer-4k-tablets
    Compared to what? Core i5 4302Y @ 4.5W. Ok, but what the heck is this then? Why is it only ~20%?
    http://www.chip.com.my/2014/10/07/intel-supercharges-mobile-devices-core-m-cpus/

    Oh I see, the 50% claim is with Specfp_rate_base 2006. The Pentium 4 1.5GHz was almost 50% faster than Pentium III 1GHz at Spec in those days too. Really convincing.

    -What about this slide?: http://www.androidcentral.com/sites...announcement_computex_slide.jpg?itok=ZB8c4D6c
    Tell me, how many people thought they were comparing against a Core i5 Y set at 4.5W? Didn't they keep drilling us the difference between SDP and TDP? If you artificially set the Haswell Y at 4.5W, what difference does it make whether its SDP or TDP?

    -Who thinks this article is talking against 22nm? Who thinks its comparing against STANDARD 14nm?: http://images.anandtech.com/doci/8355/BDW-14nm.png
    Look at the title: "Broadwell Y 14nm Design/Process Optimizations Delivered 2x Lower Power than Traditional Scaling"

    I'd have thought if it was true, 30% reduction from 14nm, and 2x reduction further would result in 15W performance at ~5W(15*0.7*0.5)Now looking at products, it seems that the slide's claims of "2x lower power" would be right if its comparing against the 22nm generation. Well... barely, since its only outperforming the 22nm Y by about 20%.

    -Not to mention the biggest of them all: http://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain
    http://www.legitreviews.com/intel-core-m-5y70-broadwell-y-benchmark-numbers-run-at-idf-2014_150217
    Tell me of one single device that gets close to those Cinebench scores. Real world 5Y70/5Y71 devices don't go much over 2 points. Funny thing is the 5Y70/5Y71 are faster than 5Y10/5Y10c, but only in single thread.

    Update:
    Because Intel's overall success is hinges on it so much, or at least perceived as on their "bleeding-edge" process technology, I'd think if they can't bring the advantages they claim it doesn't matter whether they have those fabs or not. 2x perf/watt should turn out 2x performance at same TDP. If not on Core M, at least on 15W chips. If you see their previous products, process tech always brought significant(70%+) gains on GPUs.

    I really think the truth is somewhere a mix of S|A's article about "broadwell getting castrated" and their hype about process tech and products not really living up to it.
     

    Attached Files:

    #1 DavidC, Jan 20, 2015
    Last edited: Jan 20, 2015
  2. Andrew Lauritzen

    Moderator Veteran

    Joined:
    May 21, 2004
    Messages:
    2,526
    Likes Received:
    454
    Location:
    British Columbia, Canada
    I'm not really sure it's worth responding here TBH, but let's go on the assumption that you're not just trolling for a moment.

    I think you're massively stretching and cherry-picking here. If you were actually confused by meaningless marketing slides (that don't even define the comparisons that are being made), you probably shouldn't read any tech news because you're going to run into the same thing - or worse - everywhere. Yeah marketing is stupid, move along.

    It doesn't even say what it's compared to or in what workload... pure useless marketing slide.

    Yoga 3 I already addressed... it has a lot of problems, none of which are really related to architecture. They made some bad design calls IMO and I hope they pay for it and learn better for next time. The existence of bad systems (or rather, ones with different design goals) isn't particularly interesting especially in the current race-to-the-bottom market.

    Comparing BDW-Y to HSW-Y performance, ISO-power is completely sensible... I'm not sure why you are confused there.

    The device they demoed it on...? Even articles about that demo state that Intel was very clear that OEM devices may or may not be configured similarly depending on how much cooling capacity they have. The reality of modern SoCs is that you have to compare per-device now, not per chip. The performance of a device in these form factors is determined as much or more by cooling as the chip itself.

    Hey, I'd love to see an OEM build a similarly powerful device - and maybe we'll get that, who knows - but that's totally up to the OEMs. Maybe tell your favorite OEM that you'd be interested in one :)

    First, I think you're still confused about the claims (why would there even be a "traditional" column in that chart if it wasn't comparing to another process node?), but more importantly, that reasoning is actually flawed. First - as I mentioned in the other thread - "efficiency" is not flat across the board, it's a function of power and workload. If things worked the way you thought there you could take a mobile chip, scale it up and annihilate any desktop GPU in raw performance. Hell you could probably do it with software rendering on a mW SoC. Sadly, it's not quite that simple :)

    Heh, I severely doubt that similar process shrinks brought 70%+ alone. You're confounding architecture changes here. You may not even be comparing parts that were power-limited to begin with as that is a fairly recent phenomenon in the PC space.
     
    #2 Andrew Lauritzen, Jan 20, 2015
    Last edited: Jan 20, 2015
  3. Blazkowicz

    Legend Veteran

    Joined:
    Dec 24, 2004
    Messages:
    5,607
    Likes Received:
    256
    Yes, single-threaded CPU performance still is the most important metric.
    Also that laptop is over-crippled but that's because it is a fanless, solid state laptop. They went a bit far but consider a fanless "laptop" I once used had a slow ARM with 128MB RAM and Windows CE, and was entirely useless. (so I didn't use it. a friend who knows nothing had bought it at very low price)
     
  4. DavidC

    Regular

    Joined:
    Sep 26, 2006
    Messages:
    348
    Likes Received:
    27
    No, you are right, but whether a product is good or not is relative. If you think in terms of absolutes, then I guess its not bad. But at the low end it competes against ARM and Atom chips, and at the "high-end", 15W Core versions.

    When the performance increases started levelling, the excuse was that it was "good enough" so we didn't need much more. If you think in terms of "good enough", we reached that level back in Core 2 days. But then we find we go from "good enough" to "we want a perfect experience". Most people can't afford the latter(or don't see a reason for it), so they save money and go for the low end. Actually for most users Atom is not that slow, its quite "good enough".

    So what does the Core M bring? Performance? No the 15W chips are significantly better. Fanless? Not guaranteed. Mobility? Atoms are way better, battery life and weight/thickness. Does it fit in the middle road for price? No it definitely doesn't, its right in the 15W Core territory. You are paying $300-400 over an Atom device just for single-threaded performance and nothing else.

    The "good-enough" crowd finds its not cheap enough and mobile enough. If you want single-thread for responsiveness, then is it worth paying $300 extra over the Atom parts? Not many will find that. The "performance" crowd finds out that it doesn't do much better than Atoms in single/multi-thread mixed workloads. And for Intel to hype that product to say it's the "Ultrabook vision set out in 2011", is borderline lying. Ok, lot of products fail to live up to expectations. But try to compare their marketing presentations when they had Core 2. It was pretty close to the real thing. To say marketing is seperate thing from the actual product is not true at all. It just gets bad when the product is mediocre and then they have to resort to shady tactics.

    Again, a very valid point. However, just a year ago though when the concept of "SDP" came out they were giving all sort of explanations to justify the new term, despite the fact that "SDP" is really no different than taking regular non-SDP CPU and running light-load applications. When you see companies that can achieve their goals, they don't have to resort to such tactics. But when you start setting "SDP" to a fixed-point, its no different than TDP. To say something like that just a YEAR ago, and changing the rheotric, is really questionable. 11.5W TDP but-really-4.5W-SDP vs 4.5W TDP is way different than set @ 4.5W vs 4.5W TDP. The latter is essentially equivalent, the former is not.

    So you think between the engineer quoted at Phoronix that said "Gen 8 brings greatest changes since Gen 3 to 4", and 50-70% improvement we got from Gen 6 to 7(SNB to IVB) on a process shrink with Gen 8 and not being able to do even 40%(claimed to be the gain from 22nm to 14nm for a performance increase, not perf/watt which was ~2x) somehow makes perfect sense? If process tech superiority is so important to Intel's success, and the products do not reflect that, then what's the point of having superior process?

    They failed to achieve their goals, or they are simply misleading people, plain and simple. Marketing isn't bad, until the product disappoints.
     
    #4 DavidC, Jan 21, 2015
    Last edited: Jan 21, 2015
  5. Davros

    Legend

    Joined:
    Jun 7, 2004
    Messages:
    16,480
    Likes Received:
    3,794
    What!!! I dont remember any 50-70% performance jump
     
  6. DavidC

    Regular

    Joined:
    Sep 26, 2006
    Messages:
    348
    Likes Received:
    27
    Really?

    http://www.anandtech.com/show/5772/mobile-ivy-bridge-and-asus-n56vm-preview/6

    Benchmark 1: 53.1%
    2: 68.5%
    3: 13%
    4: 43.5%
    5: 36.0%
    6: 25.6%
    7: 80.3%
    Avg of 15 titles: ~50%

    In 3DMark Vantage, we see 34.8% gain for Gen 8, Notebookcheck shows similar gains for 3DMark11. For Gen 7 Ivy Bridge we see 2x in Vantage, 49% in 3DMark06.

    Mobile chips are getting near 2x increases and chips like X1 are practically at the level of HD 5300, or better, yet we are supposed to think 30% gain on a "fantastic" 14nm process is fine.
     
    #6 DavidC, Jan 21, 2015
    Last edited: Jan 21, 2015
  7. Davros

    Legend

    Joined:
    Jun 7, 2004
    Messages:
    16,480
    Likes Received:
    3,794
    Oh your talking about mobile chips
    Hang on every one of those is running a different gpu you cant compare cpu performance in games under those conditions
     
    #7 Davros, Jan 21, 2015
    Last edited: Jan 21, 2015
  8. Paran

    Regular Newcomer

    Joined:
    Sep 15, 2011
    Messages:
    251
    Likes Received:
    14
    Gen7 improvement over Gen6 at the same clock was ~50% in games when it launched. With current drivers probably much more since SB driver are old (low clock bug etc.)
     
  9. Andrew Lauritzen

    Moderator Veteran

    Joined:
    May 21, 2004
    Messages:
    2,526
    Likes Received:
    454
    Location:
    British Columbia, Canada
    It's a design point between Atom and 15W... it slots exactly in that range in terms of power, performance, etc. And yeah, it slots in precisely where it should with respect to competitive designs too. Like I said, there's no magic here...

    And significantly higher graphics performance of course. Unless you're going to argue that Core M sucks but Baytrail is somehow okay on that front? (And really, the overall CPU performance is better as well, but I'm not even going to go down that road with you.)

    But hey if Atom works for your use cases, by all means go with that :)

    So I'm not completely confused as to what you think you're getting out of that one slide - it was basically devoid of real information but you've somehow assigned a lot to it. I don't even think it would be talking about Core M - Core M is not an ultrabook SKU, despite some OEMs using it as such. It is meant to go into fanless tablets and detachables as is completely and utterly clear from the specs of it.

    I can't even parse this man... does this even make sense to you when you read it back? Here's the few bits that I can sorta respond to...

    a) The random quote picked out of commit logs for the Linux open source driver (this is what we're at now...?) was referring to the frontend interface of Gen 8 (i.e. how the CPU submits commands to the GPU). A cursory glance at an architecture diagram would reveal to you that the actual GPU architecture is quite similar to Gen7/7.5. There have been improvements of course, but that comment was taken completely out of context.

    b) There is always a law of diminishing returns in architecture. Gen6->7 fixed a lot of glass jaws in the architecture but there are fewer and fewer of those as you go forward. If you test - for instance - 2x MSAA on Gen8 vs Gen7.5 I think you'll see massive gains too :)

    None of this has anything to do with process tech, something about which you seem be overly obsessed with and think has way more direct affect on performance than it actually does.

    I'm really losing the thread here to be honest. You've clearly got some sort of bone to pick but I honestly don't see how it relates to reality.
     
Loading...

Share This Page

  • About Us

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

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