Ryzen fiasco

By: juanrga (nospam.delete@this.juanrga.com), March 16, 2017 4:30 am
Room: Moderated Discussions
majord (majord.delete@this.gmail.com) on March 15, 2017 6:09 pm wrote:
> juanrga (nos.delete@this.juanrga.com) on March 15, 2017 3:21 am wrote:
> > majord (majord.delete@this.gmail.com) on March 14, 2017 8:25 pm wrote:
> > > juanrga (no.delete@this.juanrga.com) on March 14, 2017 5:05 am wrote:
> > > > Gabriele Svelto (gabriele.svelto.delete@this.gmail.com) on March 13, 2017 2:54 pm wrote:
> > > > > juanrga (no.delete@this.juanrga.com) on March 13, 2017 10:38 am wrote:
> > > > > > That real TDP of RyZen is the reason why its dissipation
> > > > > > is much closer to a (125W) FX-8370 than to (91W) i7-6700k.
> > > > >
> > > > > There is no such thing as a "real TDP". Intel and AMD have different definitions. What you can
> > > > > claim is that a 1800X consumes more than an i7-6700k or a FX-8370 at a given workload (and for
> > > > > a given performance). But that has nothing to do with the TDP whose definition changes from vendor
> > > > > to vendor and it does not represent an expectation of the actual power consumption of a processor
> > > > > (and is often meaningless even when comparing processors of the same family).
> > > > >
> > > >
> > > > No. The existence of different conventions/definitions doesn't imply measurements
> > > > aren't "real". And once again you selectively quote me ignoring the other issues,
> > > > including the fact AMD is violating its own definition of TDP:
> > > >
> > > > 1) The "95W" on RyZen chips is a marketing label that fooled certain people, which believed
> > > > that RyZen was super-efficient design. People in this same forum pretended that 1800X
> > > > RyZen was 60% more efficient than Broadwell-E 6900k... nothing more far from reality,
> > > > because the 1800X is about 15% less efficient than the 6900k: 16.78 vs 19.40.
> > > >
> > > > 2) AMD sent two "95W" QS for validation, because one of them wasn't a real 95W chip. The chip
> > > > identified itself as 95W, but even disabling turbo/XFR the real TDP of the sample was noticeable
> > > > higher. I estimated it would be about 106W and CanardPC confirmed real TDP was 105W
> > > >
> > > > https://mobile.twitter.com/CPCHardware/status/828666744432652290/actions
> > > > https://mobile.twitter.com/CPCHardware/status/832575767490023428?p=p
> > > >
> > > > 3) AMD "95W" retail chip at stock settings consumes about same amount of power
> > > > (more in several cases) than AMD 125W chips like the FX-8350 and 8370
> > > >
> > > > AMD "95W" --> 128.9W
> > > > AMD "125W" --> 125W
> > > >
> > > > 4) TDP numbers is not the only element where AMD has tried
> > > > to fool us: IPC claims, XFR claims, and general gaming
> > > > performance are other aspects where AMD has shared marketing numbers that don't correspond to reality.
> > >
> > > Indeed they did fool us on IPC.. they said 40% for years, then gave us ~50.. bastards!
> >
> > They said 40% over Excavator and many people believed that 40% over Excavator was an average
> > value that did mean ~70% over Piledriver. However, that 40% was on a specific bench that is
> > very dependent on cache. They sandbagged the Excavator number, because the final score was
> > 64% for Excavator in that same bench. But the Piledriver percentage was much smaller in that
> > bench: the official 52% over Piledriver that Lisa Su gave us at the RyZen tech day.
> >
> > Also you omit all those slides where AMD pretended that RyZen IPC was
> > at the Broadwell-level and reviews unable to reproduce AMD claims,
> >
> > PCWorld: "My own tests don’t quite match AMD’s results."
> >
> > PCPER: "Still, the 8% gap between the 6900K and the Ryzen 7 1800X at 3.5 GHz
> > tells me that AMD’s claims of equal IPC appear to have been overstated."
> >
> > Same thoughts about total throughput. AMD demos suggested RyZen was 2--10%
> > above Broadwell-E. Reality is RyZen is 10% below Broadwell-E.
> >
> > And better we omit the fiasco on gaming or how AMD tried to manipulate reviews.
>
>
> > I honestly am not sure what you're getting at regarding IPC claims. Simply meeting the 40% Excavator
> > goal makes Zen over 50% faster than Piledriver by default (since Excavator has close to 10%
> > IPC adv over PD.. do the maths). Yet you've already conceeded AMD beat their claim..

Funny, you have ignored my point entirely.

> And No i'm not omitting any slides, I am unaware of any claiming equal ST IPC to Broadwell outside
> of cinebench (though not even sure that was a clk/clk claim.. pretty sure it was just SKU vs SKU)

Funny, you tacitly admit you have been living under a rock all this time and missed all demos, official slides, and leaks suggesting that the IPC was Broadwell or higher than Broadwell. Too bad for you that reviewers did pay attention all this time

PCWorld: "My own tests don’t quite match AMD’s results."

PCPER: "Still, the 8% gap between the 6900K and the Ryzen 7 1800X at 3.5 GHz tells me that AMD’s claims of equal IPC appear to have been overstated."

> Throughput wise, AMD claimed Ryzen matches Broadwell in specific benchmarks, all of which
> were Rendering and Encoding, and review after review has proven that it indeed does.. Cinebench,
> Handbrake, Blender, all popular MT workloads easily matching or beating BW-E. but these
> aren't even cherry picked.. Many many other workloads also showing this.

AMD slides said that Ryzen matches Broadwell on CB15 clock for clock. Reviews show a different picture:



AMD demoed Ryzen being 10% faster than Broadwell-E on Handbrake. This demo was the origin for chip-architect's nonsensical claim about RyZen was 60% more efficient. The performance gap is less than 2%



AMD blender demo put Ryzen @3GHz ahead of Broadwell @3GHz using a custom workload. Sites as Anandtech wrote an informative article on why one has to take those demos with a grain of salt

http://www.anandtech.com/show/10585/unpacking-amds-zen-benchmark-is-zen-actually-2-faster-than-broadwell

Again reviews show a different picture, clock for clock Broadwell is faster than Ryzen on the "Ryzen" workload



> regardless though, if you actually look at the individual tests,
> you'll see Ryzen trading blows in throughput.

> It's not all rosy of course.. no one's denying BW and SL cores walk away once AVX2 is involved,
> nor when Ryzen's sub-optimal memory subsystem shows its weakeness (Read - archiving, games
> etc) but your claim of trailing by 10% across the board is pure fiction.

You confound fiction and reality. Ryzen is a good 10% behind Broadwell in throughput workloads on average and a good 20% behind in games in average





And if we consider AVX workloads then RyZen gets destroyed by the pair of 256bit units

< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Intel's underwhelming core innovationsAshraf Eassa03/04/17 11:17 PM
  Intel's underwhelming core innovationsAdrian03/05/17 12:17 AM
    Intel's underwhelming core innovationsanon03/05/17 02:25 AM
      Intel's underwhelming core innovationsAdrian03/05/17 11:37 AM
        Intel's underwhelming core innovationsanon03/05/17 11:51 AM
        Intel's underwhelming core innovationsSeni03/05/17 01:44 PM
          Intel's underwhelming core innovationsanonymou503/05/17 07:07 PM
          Intel's underwhelming core innovationsAdrian03/06/17 01:20 AM
        Intel's underwhelming core innovationsDoug S03/05/17 03:45 PM
          Intel's underwhelming core innovationsAdrian03/06/17 01:30 AM
            Intel's underwhelming core innovationsJukka Larja03/06/17 08:51 AM
              Intel's underwhelming core innovationsTravis03/06/17 11:35 AM
                Intel's underwhelming core innovationsBigos03/06/17 11:44 AM
                  Intel's underwhelming core innovationsTravis03/06/17 03:49 PM
                Intel's underwhelming core innovationsGabriele Svelto03/06/17 02:38 PM
                  Intel's underwhelming core innovationsTravis03/06/17 03:46 PM
                    Intel's underwhelming core innovationsMaynard Handley03/06/17 05:10 PM
                      Intel's underwhelming core innovationsTravis03/06/17 09:07 PM
                        Intel's underwhelming core innovationsMaynard Handley03/07/17 10:23 AM
                      Intel's underwhelming core innovationsJukka Larja03/07/17 10:26 AM
                    Intel's underwhelming core innovationsGabriele Svelto03/07/17 02:09 AM
                      Intel's underwhelming core innovationsMichael S03/07/17 03:05 AM
                        Intel's underwhelming core innovationsGabriele Svelto03/07/17 05:18 AM
                          Intel's underwhelming core innovationsVertexMaster03/07/17 07:28 AM
                            Intel's underwhelming core innovationsJukka Larja03/07/17 10:09 AM
                              Intel's underwhelming core innovationsanon03/07/17 11:20 AM
                              Intel's underwhelming core innovationsVertexMaster03/07/17 11:31 AM
                                Intel's underwhelming core innovationsGabriele Svelto03/07/17 02:37 PM
                            Intel's underwhelming core innovationsGabriele Svelto03/07/17 02:32 PM
                              Intel's underwhelming core innovationsVertexMaster03/07/17 03:06 PM
                                Intel's underwhelming core innovationsMaynard Handley03/07/17 04:41 PM
                                Intel's underwhelming core innovationsJukka Larja03/08/17 08:33 AM
                                  Intel's underwhelming core innovationsVertexMaster03/08/17 05:44 PM
                                    Intel's underwhelming core innovationswumpus03/09/17 07:28 AM
                                    Intel's underwhelming core innovationsJukka Larja03/09/17 08:04 AM
                Intel's underwhelming core innovationsJukka Larja03/07/17 10:35 AM
              Intel's underwhelming core innovationsMichael_S03/06/17 01:43 PM
                Intel's underwhelming core innovationsJukka Larja03/07/17 08:47 AM
          Intel's underwhelming core innovationsGabriele Svelto03/06/17 01:57 AM
    Intel's underwhelming core innovationsGabriele Svelto03/05/17 02:30 AM
      Intel's underwhelming core innovationsJouni Osmala03/05/17 10:30 AM
        Intel's underwhelming core innovationsTravis03/05/17 12:15 PM
        Intel's underwhelming core innovationsGabriele Svelto03/05/17 12:41 PM
          Intel's underwhelming core innovationsJouni Osmala03/07/17 12:24 AM
  Back to futureMichael S03/05/17 03:30 AM
    Back to futureTravis03/05/17 12:21 PM
  Intel's underwhelming core innovationsAlberto03/05/17 02:36 PM
    Intel's underwhelming core innovationsanon03/05/17 03:21 PM
      Intel's underwhelming core innovationsTravis03/05/17 04:07 PM
        Intel's underwhelming core innovationsanon03/05/17 05:07 PM
          Ryzen vs BroadwellE PowerVertexMaster03/05/17 05:45 PM
            Ryzen vs BroadwellE PowerDavid Kanter03/05/17 06:53 PM
              Ryzen vs BroadwellE Poweritsmydamnation03/05/17 09:30 PM
                My Firefox shows blank frames in place of results (NT)Michael S03/06/17 03:20 AM
                Ryzen vs BroadwellE Power - More DataVertexMaster03/06/17 07:25 AM
                  Ryzen vs BroadwellE Power - More DataAlberto03/06/17 02:23 PM
                    Ryzen vs BroadwellE Power - More DataKalelovil03/06/17 02:40 PM
                    Ryzen vs BroadwellE Power - More Datablue03/06/17 03:55 PM
              Ryzen vs BroadwellE PowerDoug S03/05/17 11:42 PM
                Ryzen vs BroadwellE PowerBrett03/06/17 01:21 AM
                  You still are. (NT)Megol03/06/17 08:53 AM
              Ryzen vs BroadwellE PowerHans de Vries03/06/17 02:51 AM
                Ryzen vs BroadwellE Powerjuanrga03/06/17 12:47 PM
              Ryzen vs BroadwellE Poweranon03/06/17 06:09 AM
                Ryzen vs BroadwellE PowerMichael S03/06/17 07:50 AM
                  Ryzen vs BroadwellE Poweranon03/06/17 08:09 AM
                    Ryzen vs BroadwellE PowerMichael S03/06/17 08:31 AM
                      Ryzen vs BroadwellE Poweranon03/06/17 08:58 AM
                        Ryzen vs BroadwellE PowerMichael S03/06/17 10:09 AM
                          Ryzen vs BroadwellE PowerBlue03/06/17 10:22 AM
                            Ryzen vs BroadwellE PowerMichael S03/06/17 10:42 AM
                              Ryzen vs BroadwellE PowerAdrian03/07/17 02:44 AM
                                Ryzen vs BroadwellE PowerMichael S03/07/17 03:11 AM
                                  Ryzen vs BroadwellE PowerAdrian03/07/17 04:05 AM
                                    Naples presentation published at servethehome.com (NT)Adrian03/07/17 07:47 AM
                                      Naples presentation published at servethehome.comAdrian03/07/17 08:07 AM
                          Ryzen vs BroadwellE Powergallier203/06/17 10:31 AM
                          Ryzen vs BroadwellE Poweranon03/06/17 12:37 PM
              Ryzen vs BroadwellE PowerFoyle03/06/17 06:17 AM
              Ryzen vs BroadwellE Powerjuanrga03/06/17 12:25 PM
                Ryzen vs BroadwellE Powerhobel03/08/17 02:13 AM
                  Ryzen vs BroadwellE PowerMichael S03/08/17 03:11 AM
                  Ryzen vs BroadwellE PowerSimon Farnsworth03/08/17 04:50 AM
                  Ryzen vs BroadwellE PowerAdrian03/08/17 06:11 AM
                    Ryzen vs BroadwellE Powerhobel03/08/17 06:47 AM
                      Ryzen vs BroadwellE PowerTravis03/08/17 01:47 PM
                    Ryzen vs BroadwellE PowerWilco03/08/17 04:08 PM
                      Ryzen vs BroadwellE PowerAdrian03/09/17 03:29 AM
                      Ryzen vs BroadwellE PowerAdrian03/09/17 03:58 AM
                    Ryzen vs BroadwellE PowerRicardo B03/12/17 11:54 AM
                  Ryzen vs BroadwellE Powerjuanrga03/09/17 12:52 PM
                  Ryzen vs BroadwellE PowerDavid Hess03/09/17 07:14 PM
                  Ryzen vs BroadwellE PowerRicardo B03/12/17 11:42 AM
                AMD admits 128W and 90W real TDP for RyZenjuanrga03/25/17 06:34 AM
                  Coud you please stop your bullshit?Heikki Kultala03/25/17 07:25 AM
                    X399 official TDP 180W (>200W real TDP)juanrga03/25/17 09:17 PM
                      wow that's a lot for a chipset (NT)Eric Bron03/28/17 02:49 AM
                  I already wrote that, why can't you read?anon03/26/17 01:48 AM
                    AMD admits 128W and 90W real TDP for RyZenjuanrga03/26/17 04:44 AM
                      I love how you went to the effort of changing the topic back, you child. (NT)Anon03/26/17 06:40 AM
                      You don't have any idea about thermodynamics eitheranon03/26/17 09:32 AM
                        You don't have any idea about thermodynamics eitherAlberto03/27/17 05:39 AM
                          You don't have any idea about thermodynamics eitherjuanrga04/01/17 05:19 AM
                        You don't have any ideajuanrga04/01/17 05:45 AM
                          You don't have any ideaanon04/01/17 08:39 AM
                      AMD admits 128W and 90W real TDP for RyZenhobel03/27/17 08:20 AM
                        AMD admits 128W and 90W real TDP for RyZenMegol03/30/17 09:23 AM
            Ryzen vs BroadwellE Powerjuanrga03/06/17 12:35 PM
              Ryzen vs BroadwellE Poweranon03/06/17 12:45 PM
                Ryzen vs BroadwellE Powerjuanrga03/06/17 12:59 PM
                  Ryzen vs BroadwellE Poweranon03/06/17 01:10 PM
                    Ryzen vs BroadwellE Powerjuanrga03/06/17 06:20 PM
                      Ryzen vs BroadwellE Poweranon03/07/17 02:38 AM
                        Ryzen vs BroadwellE Powerjuanrga03/07/17 05:35 AM
                          Ryzen vs BroadwellE Poweranon03/07/17 06:03 AM
                            Ryzen vs BroadwellE Powerjuanrga03/07/17 11:28 AM
                              Ryzen vs BroadwellE Poweranon03/07/17 12:07 PM
                                Ryzen vs BroadwellE Powerjuanrga03/07/17 04:34 PM
                                  Ryzen vs BroadwellE Poweranon03/07/17 04:59 PM
                                    Ryzen vs BroadwellE PowerAlberto03/08/17 04:59 PM
                                      Ryzen vs BroadwellE Poweranon03/09/17 03:33 AM
                                        Ryzen vs BroadwellE PowerAlberto03/09/17 07:50 AM
                                          bullshit on ryzen apu release timeheikki kultala03/09/17 09:18 PM
                                            Problems in mobile line.....shhhhhAlberto03/11/17 12:54 AM
                                          Ryzen vs BroadwellE PowerMatthias Waldhauer03/11/17 03:21 PM
                                            Timelines? What timelines?juanrga03/12/17 05:40 AM
                                              Timelines? What timelines?Matthias Waldhauer03/26/17 09:31 AM
                                                Timelines? What timelines?juanrga04/01/17 05:23 AM
                                      Ryzen vs BroadwellE Powermuziqaz03/09/17 04:12 AM
                                        Ryzen vs BroadwellE Powerjuanrga03/09/17 12:47 PM
                                          Ryzen vs BroadwellE Powermuziqaz03/10/17 06:52 AM
                                            Ryzen fiascojuanrga03/11/17 04:01 AM
                                              Ryzen fiascoblue03/11/17 08:57 AM
                                                Ryzen fiascojuanrga03/12/17 05:51 AM
                                                  Ryzen fiascoanon03/12/17 08:18 AM
                                                    Ryzen fiascojuanrga03/12/17 09:14 AM
                                                      Ryzen fiascoAlberto03/12/17 10:33 AM
                                                      Ryzen fiascomajord03/12/17 10:26 PM
                                                        Ryzen fiascoMegol03/13/17 04:50 AM
                                                        Ryzen fiascojuanrga03/13/17 10:14 AM
                                                          Ryzen fiascomajord03/14/17 07:51 PM
                                                            Ryzen fiascojuanrga03/15/17 02:51 AM
                                                              Power matters servers, not consumerblue03/16/17 09:22 PM
                                                                Power matters servers, not consumerjuanrga03/18/17 12:32 PM
                                                                  <105W over 13 resultsblue03/18/17 08:02 PM
                                                                    <105W over 13 results (only in your imagination))juanrga03/19/17 06:07 AM
                                                                      you confirm 105W *doh*Blue03/19/17 04:00 PM
                                                                        you confirm 105W *doh* (only in your imagination) (NT)juanrga03/20/17 03:35 AM
                                                  Ryzen fiascoGabriele Svelto03/12/17 01:25 PM
                                                    Ryzen fiascojuanrga03/13/17 10:38 AM
                                                      Ryzen fiascoGabriele Svelto03/13/17 02:54 PM
                                                        Ryzen fiascojuanrga03/14/17 05:05 AM
                                                          Ryzen brings back the fight to PC marketmuziqaz03/14/17 08:09 AM
                                                            Ryzen fiascojuanrga03/15/17 03:56 AM
                                                              Ryzen fiascoBrendan03/15/17 09:59 AM
                                                          Ryzen fiascomajord03/14/17 08:25 PM
                                                            Ryzen fiascojuanrga03/15/17 03:21 AM
                                                              Ryzen fiascomajord03/15/17 06:09 PM
                                                                Ryzen fiascojuanrga03/16/17 04:30 AM
                                                                  Ryzen fiascomajord03/16/17 03:34 PM
                                                                    Ryzen fiascoblue03/16/17 09:28 PM
                                                                    Ryzen fiascojuanrga03/18/17 01:10 PM
                                                                      HPC relative to consumers, not professional/HPCblue03/20/17 12:41 PM
                                                                        HPC relative to consumers, not professional/HPCGabriele Svelto03/20/17 03:08 PM
                                                                          HPC relative to consumers, not professional/HPCMichael S03/20/17 03:26 PM
                                                                            HPC relative to consumers, not professional/HPCGabriele Svelto03/20/17 04:39 PM
                                                                            HPC relative to consumers, not professional/HPCJukka Larja03/21/17 07:53 AM
                                                                        HPC relative to consumers, not professional/HPCKlimax03/21/17 12:33 AM
                                                                          HPC relative to consumers, not professional/HPCDomaldel03/21/17 01:41 AM
                                                                            HPC relative to consumers, not professional/HPCanon03/21/17 11:13 AM
                                                                          HPC relative to consumers, not professional/HPCMichael S03/21/17 02:33 AM
                                                                        HPC relative to consumers, not professional/HPCEmil Briggs03/21/17 11:03 AM
                                                                          HPC relative to consumers, not professional/HPCMichael S03/21/17 11:23 AM
                                                                            HPC relative to consumers, not professional/HPCEmil Briggs03/21/17 11:39 AM
                                                                              HPC relative to consumers, not professional/HPCGabriele Svelto03/23/17 08:10 AM
                                                                                HPC relative to consumers, not professional/HPCEmil Briggs03/23/17 06:30 PM
                                                                                  HPC relative to consumers, not professional/HPCGabriele Svelto03/24/17 03:13 AM
                                                                                    HPC relative to consumers, not professional/HPCIreland03/24/17 03:06 PM
                                                                                  HPC relative to consumers, not professional/HPCAdrian03/24/17 07:16 AM
                                                                                    HPC relative to consumers, not professional/HPCEmil Briggs03/24/17 01:53 PM
                                                                                      HPC relative to consumers, not professional/HPCAdrian03/25/17 03:04 AM
                  Ryzen vs BroadwellE Powerblue03/06/17 04:29 PM
      LinkAlberto03/06/17 06:30 AM
        Link.....sorryAlberto03/06/17 06:32 AM
          Link.....sorryanon03/06/17 08:21 AM
            Link.....sorryAlberto03/06/17 02:11 PM
            Link.....sorryAlberto03/06/17 02:42 PM
              Link.....sorryanon03/06/17 04:06 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell green?