Announcement

Collapse
No announcement yet.

Another reason for me not to buy AMD anymore

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by Vim_User View Post
    Since the Phenom II X6 has a turbo function it better knows about its power envelope. The point is not that CPUs do that, but that a CPU that is advertized as 3.8 GHz CPU is not able to sustain that speed under load. While Kano thinks that such a CPU is a joke I would go a step further and say that such a CPU is a fraud.
    I really don't understand the point you are trying to make. Ability to sustain clocks is always going to be a function of ambient temperature, cooling system effectiveness/cleanliness, case design re: airflow and turbulence etc... The only thing that has changed is that every year CPUs and GPUs are able to adjust clocks more finely and more quickly, and to do this with less (or zero) OS intervention and control.

    I think you're saying "I want to go back to the good old days when hardware overheated under certain stress tests unless you had a big-ass aftermarket heat sink", and I sympathise with that (I hate it when my hardware makes its own decisions without consulting me ) but I don't think that is going to happen.

    If you're just saying that processor specs should reflect 99.999999% of use cases rather than 99.99999% (or whatever the numbers are) I think that's fair, although I'm not sure it's a good idea. Again, look around a bit more. There have been complaints and questions about thermal throttling ever since the first implementations appeared, and for a while the most common question was "how do I turn this ^#$%$# off ?".

    The problem is that fine-grained thermal management makes things better for users, although it makes spec'ing and benchmarking *much* more complicated. I suspect it's one of the reasons that different sites are increasingly showing significantly different results for the same products in their benchmarks and reviews.
    Last edited by bridgman; 04-04-2013, 11:04 AM.

    Comment


    • #17
      The same reason you don't advertise the turbo clock as standard, you don't advertise the standard clock as standard in this case; it cannot sustain it in all workloads in manufacturer specified operating conditions. It is a lie. It is a 3.4ghz processor, not 3.8 ghz.
      Last edited by varikonniemi; 04-04-2013, 11:20 AM.

      Comment


      • #18
        OK, so you're saying "standard" should be defined as "minimum under any and all conditions, even with degraded cooling and running a synthetic stress test" rather than "normal/usual/whatever" ? That would mean you basically have minimum** and maximum clocks with no indication of where the processsor normally runs, which seems like a big step backwards to me.

        Server parts (GPU and CPU) do tend to have more conservative specs than client parts (for all vendors AFAICS), and I believe operation under sustained workloads is part of the reason for that.

        ** I know "minimum" isn't quite the right word here either because the processor can also be put into lower power states but it's tough even finding the right words given how aggressively modern hardware manages its own clocks.
        Last edited by bridgman; 04-04-2013, 12:09 PM.

        Comment


        • #19
          Originally posted by varikonniemi View Post
          The same reason you don't advertise the turbo clock as standard, you don't advertise the standard clock as standard in this case; it cannot sustain it in all workloads in manufacturer specified operating conditions. It is a lie. It is a 3.4ghz processor, not 3.8 ghz.
          That's what Intel does with its latest smartphones chips (Z25xx and Z27xx): they quote highest turbo speed which can't be sustained for long.

          Comment


          • #20
            Goodness, what is happening in the world that CPU manufacturers are pulling out any excuse they can to use some pretty number for marketing purposes? I remember the good old days, when... oh yeah, wait a minute, they have always done that. Nothing new here, moving right along.

            Comment


            • #21
              Originally posted by bridgman View Post
              OK, so you're saying "standard" should be defined as "minimum under any and all conditions, even with degraded cooling and running a synthetic stress test" rather than "normal/usual/whatever" ? That would mean you basically have minimum** and maximum clocks with no indication of where the processsor normally runs, which seems like a big step backwards to me.

              Server parts (GPU and CPU) do tend to have more conservative specs than client parts (for all vendors AFAICS), and I believe operation under sustained workloads is part of the reason for that.

              ** I know "minimum" isn't quite the right word here either because the processor can also be put into lower power states but it's tough even finding the right words given how aggressively modern hardware manages its own clocks.
              These tests were done with new systems, no degraded cooling, ... . What I am saying is that a CPU that is marketed as 3.8GHz CPU should be able to sustain that speed with any workload, of course under default conditions (boxed cooler, standard case without blocked airflow, normal room temperatures, ...). If the CPU is not able to do that than there is a problem, either with AMD's labeling or with the dimensioning of AMD's boxed cooler. In both cases the fault lies definitely on AMD's side and can only be fixed by AMD.

              Comment


              • #22
                The drop to 3.4 happens only in linpack which is an extreme case.
                They got it from intel.com which means that who knows what it does when it detects AMD cpu.

                Comment


                • #23
                  When I've got all cores blazing during a compile job, the last thing I would want is my CPU to drop its frequency. Don't tell me some benchmark stresses the CPU more than a "make -j10" where all cores are 100% utilized. (Try that for a Chromium build... it's relentless for the whole job, for a good 25 minutes. I chose that example because it has a lot of objects that can be compiled in parallel, unlike some builds that wait more on dependencies when you use more jobs). I compile software pretty much every day. If that's outside of my CPU's "market segment" then I would want my money back. Since when does the processor manufacturer decide what applications people are going to run?

                  I don't believe their PR bullshit for a second, that it's only Linpack. I too consider that akin to "fraud". Also disingenuous is the "it would be a problem and unfair if the tests ruined the lifespan of consumers' CPUs" statement. It's more of a problem, and unfair that AMD sells CPUs that can't even sustain running at their rated clock speeds. If that "ruins the lifespan" of the product, then it's a faulty product.

                  I do not give one tapered turd about power consumption... I don't care how clever it is. The bottom line is that the processors deliver lower performance than advertised.

                  Comment


                  • #24
                    Originally posted by Grogan View Post
                    Don't tell me some benchmark stresses the CPU more than a "make -j10" where all cores are 100% utilized. (Try that for a Chromium build... it's relentless for the whole job, for a good 25 minutes.
                    Compiles are pretty memory-intensive and *much* less demanding than compute-intensive tasks.

                    Originally posted by Grogan View Post
                    Since when does the processor manufacturer decide what applications people are going to run?
                    More precisely, processor manufacturers decide what applications people are *likely* to run during the product's market window and optimize for those applications and workloads. This includes power consumption as well as a lot of other aspects.

                    Originally posted by Grogan View Post
                    I do not give one tapered turd about power consumption... I don't care how clever it is. The bottom line is that the processors deliver lower performance than advertised.
                    Not sure if you had a chance to actually read the article but it seemed pretty clear that if you turn Turbo Core off then the processor runs at full speed.
                    Last edited by bridgman; 04-05-2013, 12:28 AM.

                    Comment


                    • #25
                      Originally posted by bridgman View Post
                      More precisely, processor manufacturers decide what applications people are *likely* to run during the product's market window and optimize for those applications and workloads. This includes power consumption as well as a lot of other aspects.
                      I can't see a "This product may possibly not be able to run at the specified clockspeed when you run workloads that AMD considers not to be mainstream!" warning on the CPU packages.

                      if you turn Turbo Core off then the processor runs at full speed.
                      So you have to disable an advertized feature of the CPU in order to make it running at nominal clockspeed. I don't know, but for me that doesn't magically make this less cheating from AMD's side.
                      In reality, it should be as simple as that: If the CPU is not able to sustain its nominal clockspeed at 100% load under default conditions (as described by me above) then don't sell it as a CPU with exactly that clockspeed, since it simply isn't such a CPU.

                      Comment


                      • #26
                        I haven't looked at the packages, but I suspect the spec sheets say something similar about Turbo Core. Read up on T-states for the Intel equivalent (clock frequency stays the same but the clock is stopped in short bursts to give the same effect as reducing frequency).

                        >>In reality, it should be as simple as that: If the CPU is not able to sustain its nominal clockspeed at 100% load under default conditions (as described by me above) then don't sell it as a CPU with exactly that clockspeed, since it simply isn't such a CPU.

                        I think it is that simple with Turbo Core disabled. That's what the article suggests anyways.

                        Let's see if our CPU folks respond any further to the article.

                        EDIT -- looks like the article was already updated and your response was to the AMD feedback. Sorry, I missed that.
                        Last edited by bridgman; 04-05-2013, 01:56 AM.

                        Comment


                        • #27
                          Originally posted by Vim_User View Post
                          Now they are cheating on their processor clockspeeds
                          Thermal throttling is a normal use case for ALL CPUs that overheating (including Intel). Just use a better case and cooling (enable HPC Mode in BIOS settings if availabe).

                          Comment

                          Working...
                          X