Announcement

Collapse
No announcement yet.

Vega M Graphics On Intel Kabylake G CPUs Are Beginning To Work Under Linux

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

  • Vega M Graphics On Intel Kabylake G CPUs Are Beginning To Work Under Linux

    Phoronix: Vega M Graphics On Intel Kabylake G CPUs Are Beginning To Work Under Linux

    We have been covering the Linux driver upbringing of "Vega M" for the Vega/Polaris graphics found in select newer Intel "Kabylake G" processors. The code is still in flight before it will work in all released versions of the Linux driver components, but for those willing to build the code or rely upon third party repositories, Vega M is now working on Linux...

    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
    Originally posted by debianxfce View Post
    That does run quite hot, almost 50C. That temperature is the same when gaming with Ryzen 5 1600 and Rx560. Idle temp for both is 28C when room temp is 22C.
    Apples vs oranges (notebook vs desktop cooling).

    Comment


    • #3
      Besides Spectre variant 1 and 2 and Meltdown there is also now Spectre variant 3 and 4.
      Seems like x86 is like Swiss cheese - full of holes.

      I will not be buying any new CPU in a while.
      Maybe Intel "Tiger Lake" will be interesting.

      Comment


      • #4
        Originally posted by uid313 View Post
        Besides Spectre variant 1 and 2 and Meltdown there is also now Spectre variant 3 and 4.
        Seems like x86 is like Swiss cheese - full of holes.

        I will not be buying any new CPU in a while.
        Maybe Intel "Tiger Lake" will be interesting.
        laaame trolling, ARM has ALL of those "holes" spreaded on different chips

        Comment


        • #5
          Originally posted by davidbepo View Post

          laaame trolling, ARM has ALL of those "holes" spreaded on different chips
          I don't know if they have all the holes, but you're right that is not only x86 since Spectre and Meltdown have impacted ARM, POWER, SPARC and S/390 too. So basically everything.

          So maybe it is wrong for me to say that x86 is like Swiss cheese. All processors are like Swiss cheese.
          Intel ME / Intel AMT doesn't help me to see x86 in a positive light though.

          Comment


          • #6
            Sad that AMD whored out their "Vega" name for a Polaris part.

            Comment


            • #7
              Originally posted by uid313 View Post
              I don't know if they have all the holes, but you're right that is not only x86 since Spectre and Meltdown have impacted ARM, POWER, SPARC and S/390 too. So basically everything.
              Huh, nope.

              In general speculative execution attacks only affect CPU that do speculative execution (some older Intel Atoms, Intel's Xeon Phi GPUs, and *LOTS* of RISCs including *lots* of ARM do not do any form of speculative execution and just cannot be affected in any way by this. Ever. Unpossible)

              Also, all the attacks depend on the depth of the pipeline : how much speculation will the CPU do in advance before throwing the work out ?
              - on one hand of the spectrum, you have some of the relatively older PowerPC. They do speculative execution, but only do a couple of instruction in advance, and thus you can't exploit much out of them.
              - on the other hand you have Intel CPUs with extremely long pipe-line that are able to speculates tons of instruction in advance. To the point that some of Google Project Zero attacks that relied on the Linux kernel's packet filter's bytecode... without even turn the JIT on. Even if each bytecode's opcode has to be interpreted (and results into several actual CPU instructions issued), these CPUs will "successfullly" speculate around such long chunks of code.


              Also, there are lots of different type of attacks. Some relying more on CPU manufacturer's weirdness than other.

              - Spectre v1 : is about speculative execution working as it is supposed to work. On the other hand it's about reading data to which a process has already access to. So proper practice can severly limit the impact (do not keep 3rd party provided JIT-ed script and sensitive information in the same proces. e.g.: a webbrowser's javascript engine and the password manager).

              - Spectre v2 : is about confusing and abusing the way some shorthand note-keeping of the CPU about execution flow and where jumps with as-of-yet-unkown destinations might land (e.g.: jump tables such as virtual functions in C++).
              Run some attacker code in one process (e.g.: a routine that you run on some VM instance to which you have normally access) and suddenly an entirely different process running on the CPUs start to speculatively jump at all the wrong places (e.g.: you end up controlling the execution flow of the hypervisor on the same machine as your VM), enabling you to do some "return-oriented programming"-like remote execution.

              Intel's CPUs have been successfully abused.
              AMD's CPUs, despite doing speculative branch target prediction, have not : nobody has yet figured if there's an actual way to abuse the system successfully).

              - Meltdown is about CPUs not even paying attention to memory protection during speculation, processes access data that they shouldn't have access to (e.g.: a
              Only Intel CPUs and a handful ARM CPUs influenced by intel design actually do something as stupid in the name of shaving cycles.
              Nobody else is affected by this.

              - Spectre v4 is about read after write, and CPUs attempting to speculate reads before the content of a memory location has been committed (e.g.: because pointer math is involved, and there's no guarantee that the specific cell you're reading from will be the over-written one).
              Again, you're still basically only reading memory that your process has already access to.
              On a decent OS like Linux, the amount of sensitive information that can leak is very limited (memory pages never contain 3rd party content, only zeroes). Basically only the stack (when calling and returning from system calls), and that should be sanitized/overwritten anyway.
              You'd need CPUs with horribly long speculation pipeline to be able to exfiltrate data past such barriers (hello, Intel CPUs).

              The demo I've seen (That are supposed to also work on AMD) have all been of the "reading my own data" kind of cases.
              (The kernel case once again abuses the kernel's packet filter's bytecode's JIT, and requires patches in additional instruction opcodes that the standard bytecode's opcode don't provide but which are required to pull the demo successfully. Yay~ )


              So basically :
              - No, not all CPUs are affected at all. Some even don't do speculative execution to begin with. (lots of ARM don't)
              - Some of these attack like Meltdown require the manufacturer to do some really stupid optimisation that nearly nobody does (except Intel)
              - Lots of the attacks highly depend on how the CPUs is designed and might not even be exploitable in any meaningful way (again, Intel CPUs seems slightly more affected)
              Last edited by DrYak; 28 May 2018, 06:18 AM.

              Comment


              • #8
                Originally posted by microcode View Post
                Sad that AMD whored out their "Vega" name for a Polaris part.
                Funny, but as far as I know, it should have more performance than raven ridge (Vega+Ryzen APU). I don't agree with marketing calling it VegaM, but at the end of the day, calling it a Polaris could be misleading as well. I'm told the memory changes alone puts it in a different category.

                Comment


                • #9
                  Originally posted by Mystro256 View Post

                  Funny, but as far as I know, it should have more performance than raven ridge (Vega+Ryzen APU). I don't agree with marketing calling it VegaM, but at the end of the day, calling it a Polaris could be misleading as well. I'm told the memory changes alone puts it in a different category.
                  Polaris 69?

                  Comment

                  Working...
                  X