PassMark Software


No announcement yet.

RX Vega and Thread Ripper Sabotage

  • Filter
  • Time
  • Show
Clear All
new posts

  • RX Vega and Thread Ripper Sabotage

    Ok..maybe I'm just paranoid. I've seen low benches before with most CPU's/GPU's from either of the big three, some people are really bad at building computers, some people lose the silicon lottery really badly.

    But out of the few Threadripper and Vega benches that have gone up, some are INSANELY low, like... RX Vega is worse than my RX 480 BAD. I've never seen anything like this before. Many of the recent benchmark results need to be purged, Somebody is sabotaging the rankings (Paid Intel shills maybe?) I simply find it too hard to believe that an Intel/Nvidia fanboy would BUY Vega or Threadripper just to drop the clock to 1/3rd its normal level and sabotage the rankings...but it definitely looks like some serious sabotage from SOMEBODY going on here. The number of scores being sent in by benchmarkers that score around 1/3rd what the hardware can actually do is beyond staggering as far as percentages of actual entries go.

    Passmark please look into this, there is no way that this is the result of hordes of people building Vega and Threadripper rigs with IQ's levels that should render them completely incapable of building something that even posts much less actually working but scoring 1/3rd of what it should be.

  • #2
    maybe I'm just paranoid
    Yes, likely you are.

    There aren't hordes of people as you claim. Vega and Threadripper 1920X are both still very rare. So we don't have many samples to create a stable average as yet. Even the 1950X isn't very popular as yet.

    It also isn't helping that AMD has internally given all the Vega models the same ID. So we can't tell them apart as yet (we are working on it).

    Yes, there are a couple of low results (and also some high ones), but the ratio seems about the same as for other models of CPU and GPUs.

    It is actually pretty easy to get low results. Bad device drivers, monitors locking frame rates to 60Hz, some junk software running in the background or overheating are just a few examples of things that go wrong. We see it every day. I did delete a couple of duplicate results today however.

    We had a look at where the low results came from, how many there are and why the overall result was low. We couldn't see any evidence of the deliberate fraud / sabotage.

    Having enough money to buy these parts doesn't automatically mean you're an expert PC builder. And even the experts don't get it right first time every time.


    • #3
      To be true, I fully agree (and many people will agree) with neilio. I don't think that there is sabotage or something, but results are definitely ruined and in favor of Intel and nVidia.
      Yes it can be because of small numbers of samples, maybe cause of bad drivers, but that is quite hard to believe that it can happen from 30 samples of Threadripper 1950X.

      How it is possible to 7900X be even near 1950X in synthetic benchmarks? Even an Liquid Nitrogen cooled 7900X can't match a 1950X in Cinebech and GeekBench. I don't think that all 7900X samples here are Nitrogen cooled. Non-overclocked 1950X gets ~3100 Cb in Cinebench, 7900X gets ~2200 Cb. In GeekBench scores are with even bigger margin. How can they have similar scores here? Maybe it's because this benchmark testing different instruction sets, I don't know. But that's very bad and super misleading. Exact same situation is with some AMD GPU's, like R9 295X2. This card scores lower than GTX 780? Than R9 390? GTX 970? Are you kidding me? This card was long time in Top #3 of 3DMark cards 2016 years, now it is #6. In Unigine there is same situation like in 3D Mark. How can it have so low scores here? I had this card, it was much much more faster than all 3 cards I mentioned. Not even close. Almost 2x faster. In games ~40-50% faster. But this is not game, this is synthetic benchmark. It should show fullest possible SLI/CFX potential.

      And I'm not being paranoid, I like passmark tests very much, but some results are misleading at least for now. I hope they will get back on track in some time
      Last edited by IFeelYou; 09-01-2017, 12:54 PM.


      • #4
        Different benchmarks produce different results. While our CPU benchmark is mostly multithreaded, it also includes a single threaded task. This hurts the AMD CPUs (when compare to the top end Intel CPUs).

        I should point also point out that we are ranking Threadripper as the number 1 CPU at the moment (if you ignore a couple of $2000 Xeons). I'm not sure how this can be an Intel bias?

        Quoting for Toms hardware,
        "It's clear that Intel's Core i9-7900X offers better average frame rates during purely gaming workloads..."

        "Threadripper outpaces the similarly-priced -7900X in rendering, encoding, and compression. As expected, it isn't quite as nimble in lightly-threaded applications, such as decompression and portions of the Adobe suite. Those applications continue to favor Intel's IPC throughput and frequency."

        Quote from Hot hardware,
        "In lightly-threaded workloads, Threadripper trails Intel's latest processors, however, which translates to lower performance in applications and games that can't leverage all of the additional compute resources Threadripper has at its disposal."

        Quote from Anandtech,
        "Threadripper's gaming performance is middling at best: very few games can use all those threads and the variable DRAM latency means that the cores are sometimes metaphorically tripping over themselves trying to talk to each other and predict when work will be done."

        Of course if you insist on cherry picking results, you could come to a different opinion than all the reviewers.

        The R9 295X2 is a two GPUs in Crossfire mode. Only a few games really make good use of Crossfire. (In short it only really works well if AMD optimize their driver for your game). So again, this is cherry picking. One fast GPU is better than two slow ones.


        • #5
          Well ThreadRipper is not #1 CPU in Your list even ignoring Xenons, cause:
          It got beaten badly from 7900X, which is absolute nonsense. There are not a single benchmark (except single threaded) in which 1950X is beaten by 7900X.

          You wrote few quotings from reviewers, but that's just few quotes from context. I can tell some more quotes:
          - "AMD Threadripper 1950X review: Better than Intel in almost every way" -
          - " The key difference being that the 1950X often delivers 20-30% more performance while consuming almost 10% less power under full load." -
          - "For now, that means AMDís $999 processor offers significantly more performance in multithreaded workloads than Intelís similarly priced 10-core Core i9-7900X. There were even a couple of instances where AMDís $799 processor was faster than Intelís $999 part." - hothardware,com

          And many more. If You read those reviews, you can't deny the FACT that 1950X is much more stronger than 7900X in almost any multi threaded test.

          Passmark CPU test doesn't consist only from Single Threaded test (single thread test occupies only small part as far as I know), so... the results are definitely wrong.

          Now, I think I might found a problem. My friend has AMD Ryzen 1700 OC to 4.0 Ghz. When benchmarking with Performance test he run CPU-Z, AI-Suite, Task-Manager - all programs shows that CPU is running only at 2.3 Ghz! Maybe that is the case?


          • #6
            And again, I'm not trying to bash PerformanceTest or something. I want to help as much as I can to make this test best of the best.


            • #7
              The reviews you quoted all reference the value of the CPU (performance for the cost). Which we agree is great. Or was before Intel started dropping prices. But this isn't what our high end chart is graphing. We are just graphing performance (regardless of cost).

              Your overclocked CPU might be overheating and throttling.


              • #8
                Performance for the cost? Man, they both cost 999 dollars. They're totally similar in cost, so it's only the performance that counts. I don't understand why you trying to neglect that 1950X is much more faster? You didn't read reviews? Maybe you need links?

                In like 9 of 10 synthetic multi-threaded test 1950x is better, and by big margin most of the cases. The only place where 1950x is slower - it's games and single threaded programs. How many percent of Passmark CPU score is multi-thread and how many is single-thread?

                That CPU isn't overheating and isn't throttling. If that was the case - I wouldn't wrote that here. It has dual 420mm EKWB rads. It is just the CPU clock detection problem in Passmark Test.


                • #9
                  In like 9 of 10 synthetic multi-threaded test
                  Yes if you cherry pick results this might be true. But what about the real world?
                  Real world software is rarely fully multi-threaded.

                  Our formula for mixing the single threaded test and multithreaded tests to get an overall result is here,

                  It is just the CPU clock detection problem in Passmark Test.
                  If you are overclocking, but you are getting low benchmarks compared to the average, and also reading low clock speeds, then likely it is the system that has the problem, not the benchmark. Clock detection doesn't impact the benchmark in any case, but it can be an indicator that something isn't right.


                  • #10
                    There was this score I saw
                    Which is run inside a virtual machine, has ridiculously low score and probably isn't an objective test result and as publicly displayed results are an average of submitted results and as such this will skew the results.

                    I'm all for virtual benches but they should be separate from the normal benches.


                    • #11
                      Results from baseline 890871 were automatically excluded from the average. The CPU was only running on half of it's normal number of cores in a VM. We auto exclude any result where the number of threads running doesn't match the number of available cores.

                      Benchmarking VMs is a valid thing to do however, as they are very widely used and people have a valid reason to want to compare their performance. So we don't want to delete all results from VMs.