Announcement

Collapse
No announcement yet.

Linux To Incorporate Intel CPU Hybrid Topology For Determining Vulnerabilities/Mitigations

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

  • Linux To Incorporate Intel CPU Hybrid Topology For Determining Vulnerabilities/Mitigations

    Phoronix: Linux To Incorporate Intel CPU Hybrid Topology For Determining Vulnerabilities/Mitigations

    Within the every increasingly complex world of CPU security mitigations, Intel engineers have submitted Linux kernel patches to begin taking into account the CPU core "hybrid" topology when determining relevant CPU security vulnerabilities and in turn the mitigations to apply...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    So let me get this straight. Even WITHIN Intel, iSomethingMeaningless is meaningless ?

    Comment


    • #3
      The immediate difference with these new Linux kernel patches is that the P-core-only processors will no longer see Register File Data Sampling (RFDS) mitigation applied with RFDS only being needed for E-core / Atom processor cores. Now the mitigation code is "smart" enough to account for the core type topology.​
      Thank you! This is a question I had, back when you benchmarked that mitigation - whether it applied only to the E-cores. I'm glad to see that's now being addressed!

      Also, I think we need something akin to the Family/Model information for the individual cores.
      Last edited by coder; 17 June 2024, 08:08 AM.

      Comment


      • #4
        Funny how their higher priced products are not deemed vulnerable.

        Comment


        • #5
          I hope it is just the beginning of a push from Intel to make Linux a great fit for P/E-cores in general. As the Thread Director patches haven't made it over the finish line yet, there is still lots of work ahead to get to that goal...

          Comment


          • #6
            Originally posted by emansom View Post
            Funny how their higher priced products are not deemed vulnerable.
            Huh? The example given by the article of a P-core -only i3-14100 is not a high-priced product! Current street price is $135, which is probably lower than any of their current-gen hybrid CPUs. Then again, it's only a 4-core CPU, so that makes sense.

            BTW, it uses a different die which contains 6 P-cores and 0 E-cores. Obviously, two of the P-cores have been disabled for that model.

            Comment


            • #7
              Originally posted by emansom View Post
              Funny how their higher priced products are not deemed vulnerable.
              INTEL Security - You Get What You Pay For

              Comment


              • #8
                Originally posted by ms178 View Post
                I hope it is just the beginning of a push from Intel to make Linux a great fit for P/E-cores in general. As the Thread Director patches haven't made it over the finish line yet, there is still lots of work ahead to get to that goal...
                there are rumors that the 9000 X3D cores will all be symmetric cores. means the 9950X3D will have 2 X3D chiplets with two 64mb cache.
                any sane person will switch to a symmetric cpu design with AVX512 on all cores and much 3D Cache on all cpu cores.
                and the mental illness of intel that the E-Cores do not have AVX512 and their asymmetric ideology will go down in history as failure.

                Phantom circuit Sequence Reducer Dyslexia

                Comment


                • #9
                  Originally posted by qarium View Post

                  there are rumors that the 9000 X3D cores will all be symmetric cores. means the 9950X3D will have 2 X3D chiplets with two 64mb cache.
                  any sane person will switch to a symmetric cpu design with AVX512 on all cores and much 3D Cache on all cpu cores.
                  and the mental illness of intel that the E-Cores do not have AVX512 and their asymmetric ideology will go down in history as failure.
                  I agree with you fully on Intel's design mistakes. But AMD doesn't have a well-rounded mid-range CPU that is capable of both gaming and compilation workloads that appeals to me. AM4 SKUs are too old by now and AM5 is still too expensive overall (CPU + RAM + Motherboard). The 12-core models on both platforms also suffer from CCX-to-CCX latency issues. The X3D beasts you talk about could cost well over 650 EUR and are out of my price range anyway. As my beloved Haswell-Xeon shows some signs of the end of its life, I needed to invest now. Thankfully I could get a hefty discount for a 14700KF for 339 EUR with tax and shipping included this week. Fingers crossed that my sample won't show any stability issues or I won't hesitate to RMA it and re-evaluate my options.

                  The thinking behind that LGA1700 build was to bridge three to four years and re-use some DDR4 RAM that I already own and minimize my investment. I got an acceptable Z690 board for cheap, too. All in all, I've spend 460 EUR for CPU and Motherboard, the new setup should be good enough for my needs as I am planning to undervolt the CPU anyway due to using air cooling. AMD cannot beat the value of that combo for gaming and compilation workloads under these specific conditions. And with CAMM2 and other innovations on the horizon, I didn't want to invest in DDR5 and AM5 at this point. If AMD offered me a 8 Zen 5 + 12 Zen 5 C core for 300 EUR, I would have considered it. But their current offerings simply didn't fit my profile very well.
                  Last edited by ms178; 19 June 2024, 02:10 PM.

                  Comment


                  • #10
                    Originally posted by ms178 View Post
                    The 12-core models on both platforms also suffer from CCX-to-CCX latency issues.
                    Where do you see examples of this, in Phoronix' benchmarks? IMO, this is mostly a theoretical problem and not relevant for most everyday workloads.

                    An advantage of the 12-core models is that the heat is distributed between two dies running at 75% occupancy and the amount of L3 cache per core is greater than you get with either the 8-core or 16-core models (if we leave aside the matter of 3D cache).

                    Originally posted by ms178 View Post
                    ​Thankfully I could get a hefty discount for a 14700KF for 339 EUR with tax and shipping included this week. Fingers crossed that my sample won't show any stability issues or I won't hesitate to RMA it and re-evaluate my options.
                    Good luck.

                    Originally posted by ms178 View Post
                    ​​​I got an acceptable Z690 board for cheap, too.
                    Too bad you didn't go with a W680 board. Then, you could've replaced a workstation with another workstation and equipped it with ECC RAM.

                    Originally posted by ms178 View Post
                    ​​And with CAMM2 and other innovations on the horizon, I didn't want to invest in DDR5 and AM5 at this point.
                    I'm not convinced CAMM2 will have a significant presence in the enthusiast desktop market. Also, doesn't it limit you to the equivalent capacity of 1 DIMM per channel?

                    I like the idea of being able to carry over RAM from one build to the next, but the reality of how infrequently I upgrade means that memory standards have already changed by then. If you're still on Haswell, then the same might be true for you.

                    Originally posted by ms178 View Post
                    ​If AMD offered me a 8 Zen 5 + 12 Zen 5 C core for 300 EUR, I would have considered it. But their current offerings simply didn't fit my profile very well.
                    IMO, it was bold to pull the trigger just a month before Zen 5 launches. I hope you don't regret it.
                    Last edited by coder; 19 June 2024, 09:12 PM.

                    Comment

                    Working...
                    X