Announcement

Collapse
No announcement yet.

With AMD Zen 4, It's Surprisingly Not Worthwhile Disabling CPU Security Mitigations

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

  • geearf
    replied

    Originally posted by Anux View Post


    We are already looking ahead over 10000 µOPs.

    Originally posted by coder View Post
    BTB is not the same as speculative execution. It's really a type of cache.

    The key stat you should look to is the Re-Order Buffer size. For Zen 4, it sits at 320. Golden Cove's is 512. Meanwhile, the one in Apple's A14 is 630!

    https://www.anandtech.com/show/16226...14-deep-dive/2
    Thank you both!

    Leave a comment:


  • Anux
    replied
    Originally posted by coder View Post
    BTB is not the same as speculative execution. It's really a type of cache.
    You are right, I read a little more, it sounds like BTB is a read ahead to fill the µOP cache for the branch predictor?

    Leave a comment:


  • coder
    replied
    Originally posted by Anux View Post
    We are already looking ahead over 10000 µOPs.
    BTB is not the same as speculative execution. It's really a type of cache.

    The key stat you should look to is the Re-Order Buffer size. For Zen 4, it sits at 320. Golden Cove's is 512. Meanwhile, the one in Apple's A14 is 630!

    https://www.anandtech.com/show/16226...14-deep-dive/2

    Leave a comment:


  • Anux
    replied
    Originally posted by geearf View Post

    Sorry with my limited knowledge on this I don't see what you're pointing to. I see there are improvements to prediction but I don't really understand it.

    Thank you!
    AMD has increased the L1 Branch Target Buffer (BTB) cache size by 50%, to 2 x 1.5k entries. And similarly, the L2 BTB has been increased to 2 x 7k entries
    We are already looking ahead over 10000 µOPs.

    Leave a comment:


  • geearf
    replied
    Originally posted by Anux View Post
    Sorry with my limited knowledge on this I don't see what you're pointing to. I see there are improvements to prediction but I don't really understand it.

    Thank you!

    Leave a comment:


  • Anux
    replied
    Originally posted by geearf View Post

    You can look that far ahead? Wow!

    Thank you!
    You will be in for a suprise: https://www.anandtech.com/show/17585...the-high-end/8

    Leave a comment:


  • geearf
    replied
    Originally posted by Developer12 View Post

    Sure. How many branches do you think you might encounter if you looked 50 or 100 instructions ahead? The other way to look at it, is look how many parallel integer and floating point engines modern CPU cores have.
    You can look that far ahead? Wow!

    Thank you!

    Leave a comment:


  • Developer12
    replied
    Originally posted by geearf View Post

    Thanks a lot for all this! I appreciate learning more.
    Since speculation happens in parallel (I assumed it'd be serial) does that mean there's more than 1 speculation done if the hardware is available?
    Sure. How many branches do you think you might encounter if you looked 50 or 100 instructions ahead? The other way to look at it, is look how many parallel integer and floating point engines modern CPU cores have.

    Leave a comment:


  • geearf
    replied
    Originally posted by NobodyXu View Post

    Yes, the CPU will keep executing as long the operands is available in registers and the unit for corresponding computation is available (integer and float point are in different units and there are multiple of int/float processing units).

    Sometimes it would even guess the result of a "if" and execute the corresponding branch.
    If the guess is wrong, that branch is discarded.

    And that is meltdown.

    Even with the branch discarded, some memory region is still accessed and bring into the cache.
    By measuring the time required to access certain memory region, the hackers can use that to do some fancy thing using that information.

    I'm not very familiar with the detailed exploit though, so I'm going to stop here.
    I see, thank you for the explanation, it makes me want to research this subject more.

    Leave a comment:


  • NobodyXu
    replied
    Originally posted by geearf View Post

    Thanks a lot for all this! I appreciate learning more.
    Since speculation happens in parallel (I assumed it'd be serial) does that mean there's more than 1 speculation done if the hardware is available?
    Yes, the CPU will keep executing as long the operands is available in registers and the unit for corresponding computation is available (integer and float point are in different units and there are multiple of int/float processing units).

    Sometimes it would even guess the result of a "if" and execute the corresponding branch.
    If the guess is wrong, that branch is discarded.

    And that is meltdown.

    Even with the branch discarded, some memory region is still accessed and bring into the cache.
    By measuring the time required to access certain memory region, the hackers can use that to do some fancy thing using that information.

    I'm not very familiar with the detailed exploit though, so I'm going to stop here.

    Leave a comment:

Working...
X