i5 (6300HQ) vs i7 (6700HQ) worth the $100 difference for people who use VM's?

Discussion in 'Hardware Components and Aftermarket Upgrades' started by death_relic0, Nov 12, 2015.

Thread Status:
Not open for further replies.
  1. octiceps

    octiceps Nimrod

    Reputations:
    3,146
    Messages:
    9,956
    Likes Received:
    4,193
    Trophy Points:
    431
    Until Skylake, mobile i5 always came with HT. They were basically desktop i3--dual-core w/HT--but with Turbo Boost. On desktop there were some dual-core w/HT i5 back in Westmere and Sandy Bridge days, but other than that, all desktop i5 are quad-core w/o HT. Now with Skylake, mobile finally gets what desktop had for years, a quad-core w/o HT.
     
  2. nipsen

    nipsen Notebook Ditty

    Reputations:
    694
    Messages:
    1,685
    Likes Received:
    130
    Trophy Points:
    81
    So.. they're differentiating the product so a larger amount of normal users should want the i7 over the i5?
     
  3. octiceps

    octiceps Nimrod

    Reputations:
    3,146
    Messages:
    9,956
    Likes Received:
    4,193
    Trophy Points:
    431
    On the contrary. The mobile i5 is much more attractive to most users now that it's finally quad-core and cheaper than the i7. Hyper-Threading is still something with dubious real-world benefit at best.
     
  4. nipsen

    nipsen Notebook Ditty

    Reputations:
    694
    Messages:
    1,685
    Likes Received:
    130
    Trophy Points:
    81
    Mm. I don't know. It's kind of essential to context-switching if there's no shared cache between the cores. But.. haven't read anything about these i5s performing any worse than you'd expect, so.. just curious about why they'd set it up like that.
     
  5. octiceps

    octiceps Nimrod

    Reputations:
    3,146
    Messages:
    9,956
    Likes Received:
    4,193
    Trophy Points:
    431
    Intel CPUs have shared L3 cache.

    Because why pay $130 more for a feature (HT) that you might not need? There's a reason the i5 -K is consistently the bestselling model every gen. It's $100 cheaper and performs identically to the i7 in applications that don't utilize HT, e.g. majority of games. Laptop users have complained about the lack of true quad-core i5 CPUs on mobile that desktop has enjoyed for years, now we finally have parity. Good on you, Intel. About time.
     
    Last edited: Nov 21, 2015
  6. nipsen

    nipsen Notebook Ditty

    Reputations:
    694
    Messages:
    1,685
    Likes Received:
    130
    Trophy Points:
    81
    Yeah, but it's always coherent.. copied and maintained towards the l2 cache. So... as far as I can tell... it's technically impossible to use this for microcode optimisation, or to benefit from shortened or collapsed assembly expressions, and so on.

    ...so it would happen something like this: you could imagine that a complex registry operation was collapsed into one line of assembly code (this actually happens very often), and then the result of this final operation is stored in l2 cache. All the processed content is copied to l3 cache, which then is available to main memory. So you can fetch the relevant computed result. And until the next context-change (new thread is serviced for a microsecond or so), the l1-cache contains the shortened expression, and can reuse it. But the shortened assembly code is unavailable to the other cores. And the moment you make any changes to the l3-cache, also to the current core.

    But if it actually turns out that four cores with shared cache is faster in practice than two w/hyperthreading, that's kind of interesting. Because I've been wondering about exactly how specialized the operations have to be to cause HT to actually save processor cycles. In theory, it's easy to see a situation like that. But in practice, I kind of wonder if you need something like 50% instruction cache hits to break even on the processor diagram. Obviously it's useful on fewer cores. But with more free cores, does it make any difference at all.. you know...? It's... difficult for me to figure this out from code-examples, and intels software/hardware manual doesn't really go into it in any detail.

    So if that i5 actually doesn't fall off the scale on performance in general, and perhaps uses less power since the cores are less active, things of that sort -- that is actually very interesting.
     
  7. octiceps

    octiceps Nimrod

    Reputations:
    3,146
    Messages:
    9,956
    Likes Received:
    4,193
    Trophy Points:
    431
    Where you been man. That's always been the case. More physical cores are better than HT. HT just improves utilization of the existing cores in optimized applications, it doesn't give you more cores.

    It does not. Like I said, it performs indentically to i7 (at same clock speed) in applications that don't take advantage of HT. The heat, power, and die space savings from not having HT are minimal though.
     
    Last edited: Nov 22, 2015
    Starlight5 and i_pk_pjers_i like this.
  8. i_pk_pjers_i

    i_pk_pjers_i Even the ppl who never frown eventually break down

    Reputations:
    205
    Messages:
    1,033
    Likes Received:
    592
    Trophy Points:
    131
    For gaming the i5 is absolutely fine but for stuff like VMs or rendering I would always go with the i7. Then again, I pretty much always go with i7s.
     
    Starlight5 and alexhawker like this.
  9. nipsen

    nipsen Notebook Ditty

    Reputations:
    694
    Messages:
    1,685
    Likes Received:
    130
    Trophy Points:
    81
    Mm, no, I meant with a workload that would normally occupy two cores, distributed in some automatic way to four cores. If you have a way to fill four cores with the same finite amount of work as before, then obviously no amount of magic is going to compensate for that.

    What I've been wondering about, but haven't been able to figure out, is what sort of code you would have to create to get above that 15% "expected" bonus HT seems to give you. Or, well, that it maybe turns out in the end that the benefit is actually extremely small on anything that isn't specially coded and compiled in a specific way as well..
     
  10. i_pk_pjers_i

    i_pk_pjers_i Even the ppl who never frown eventually break down

    Reputations:
    205
    Messages:
    1,033
    Likes Received:
    592
    Trophy Points:
    131
    Like I said, you can notice a benefit in anything that really taxes your CPU such as VMs, video rendering, etc.
     
Thread Status:
Not open for further replies.

Share This Page