Page 337 of 354 FirstFirst ... 237287327335336337338339347 ... LastLast
Results 3,361 to 3,370 of 3537

Thread: Kaveri 2.0 discussion.

  1. #3361
    640k who needs more?
    Join Date
    Mar 2010
    Posts
    995
    1st of all, Haswell doesn't have 2x more ST performance than SR core, that is just nonsense. The difference varies between 30% and 60% depending on the test.
    Second, i5 Haswell has 4 full fledged(FP-wise) cores while 7850K has 2 FP coprocessors each featuring 2 FP pipelines. Basically 2 module Kaveri is at 2:1 disadvantage in sheer number of FP pipelines when compared to basic QC Haswell. That's why FX8xxx competes great with i5/i7 in MT benchmarks as it is finally on parity FP units-wise.

    If, and a big IF, Carrizo gets double FP resources then it will maybe come close to FX8xxx and i5s IB/Haswell parts in SIMD workloads. It also will depend on the way AMD designs the new FP unit and whether it will be able to fully utilize whole FP coprocessor outside of 256b AVX2/FMA code.

  2. #3362

  3. #3363
    No idea why one would want to drag SuperPi into this as an argument, as it's totally useless to make any assessment between different architectures.

    Anyhow, it was posted before, but I'll do it again:

    http://www.extremetech.com/wp-conten...ibreOffice.png

    This shows, perfectly great, why the pure raw power of the CPU cores is no longer the one and only thing to look at. Where Excell may still be favouring Intel, LibreOffice's Calc supporting HSA, makes it possible for Kaveri to totally destroy Intel's i5.

    Looking only at the CPU cores for performance, it's much 2013.
    Gigabyte GA-A75M-D2H||AMD A8-3850||Corsair XMS3 PC3-16000(2000MHz)||Sapphire Radeon HD6670||Fractal Design Define Mini
    Life, it all of a sudden comes back to you and you have no clue who it is.

  4. #3364
    640k who needs more?
    Join Date
    Mar 2010
    Posts
    995
    Now I think he is trolling us here. Sure, let me cherry pick a few benchmarks to "prove" Haswell is 2x faster in ST tasks. You cannot be serious, or can you?

  5. #3365
    Banned
    Join Date
    Jan 2013
    Posts
    86
    wtf, I tell the reality to juanrgra= infraction. I post realworld benchmarks is now trolling yet the people who troll with nothing to back it up don't. The reason why intel cpu is so fast compared to amd is amd modular design has basically made single threaded performance of piledriver steamroller and bulldozer half. I expect my warning to be revoked unless there is a new definition of trolling which is posting the truth is trolling now.

  6. #3366
    640k who needs more?
    Join Date
    Nov 2009
    Posts
    753

    Talking

    Quote Originally Posted by nissangtr786 View Post
    The reason why intel cpu is so fast compared to amd is amd modular design has basically made single threaded performance of piledriver steamroller and bulldozer half.
    or maybe audit benchmarks to ensure no ICC was used in compiling.
    it still probably uses sse1 only for amd cpu.

    single thread performance can be improved by using newer instructions sets to overcome the fewer resources per INT core too.

    but you already know that!

    i remember francois has a nissan gtr. is that you??

  7. #3367
    Banned
    Join Date
    Jan 2013
    Posts
    86
    http://www.anandtech.com/bench/product/837?vs=700
    This shows haswell 2.1x ipc over piledriver to as with same amount of cores but at a lower clock i5 4670k doubles the fx4300. Steamroller hasn't imporved much in ipc even comparing to an a10-5800k.
    http://www.anandtech.com/bench/product/837?vs=675

    Steamroller lack of fpu performance and ipc improvement shows even against older amd apu:
    http://www.anandtech.com/show/7677/a...0-a10-7850k/10
    This shows why comparing an i5 2500k to an a10-7850k is a mismatch.
    http://images.anandtech.com/graphs/graph7677/60947.png

  8. #3368
    Senior Member
    Join Date
    May 2010
    Posts
    185
    Quote Originally Posted by inf64 View Post
    1st of all, Haswell doesn't have 2x more ST performance than SR core, that is just nonsense. The difference varies between 30% and 60% depending on the test.
    That's not entirely accurate (semi-accurate, you could say).

    In their Kaveri-Review, computerbase.de tested - amongst other stuff - BF4 and Crysis 3 running at 1280x720 with a 290X to remove any GPU-limitation, and an i5-4670K was 86% (BF4) to 100% (Crysis3) faster than the 7850K.

    So at least in games, Haswell DOES have up to 2x the ST performance of SR.

  9. #3369
    Banned
    Join Date
    Jan 2013
    Posts
    86
    Quote Originally Posted by reaperrr View Post
    That's not entirely accurate (semi-accurate, you could say).

    In their Kaveri-Review, computerbase.de tested - amongst other stuff - BF4 and Crysis 3 running at 1280x720 with a 290X to remove any GPU-limitation, and an i5-4670K was 86% (BF4) to 100% (Crysis3) faster than the 7850K.

    So at least in games, Haswell DOES have up to 2x the ST performance of SR.
    yep crysis 3 and f1 2013 with a high end gpu shows basicaly that the i5 more brute performance required to push high fps at 720p gaming to load the gpu, 4 cores each and haswell ipc shown is around 1.8x-2x that of steamroller.

  10. #3370
    Senior Member
    Join Date
    Jun 2010
    Posts
    4,229
    Pointless benchmarks from bad reviews, those games are playable without a discrete GPU with Kaveri at that resolution.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux