Announcement

Collapse
No announcement yet.

Google Engineers Are Becoming Concerned Over Some Arm Platforms Lacking Spectre V2 Mitigations

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

  • zyxxel
    replied
    Originally posted by vladpetric View Post

    Nowhere did I claim that IPC is a fixed value - no architect worth their salt would say that IPC is a fixed value. The equation is trivial, not the stuff behind it. The point of that equation, particularly in current times, is to highlight how important IPC is.

    Just read H&P and maybe take an architecture class. Again, you think you know stuff, but you don't. Based on the discussion I have some pretty good idea about how little you know about the topic, actually (but there's Dunning-Kruger, of course).
    Yes. And you need to work up on your Dunning-Kruger. With your ability to debate here, there is no way in hell you can defend a thesis.

    Leave a comment:


  • vladpetric
    replied
    Originally posted by zyxxel View Post

    What I do know is that you post bull. You have zero information about what I do - or do not - know. But that doesn't stop you from crapping in the thread.

    And you still failed to understand that if I add one point that is important, that doesn't invalidate the list of other important points that must be fulfilled to develop a great chip. That's on you to work on.

    Another thing - your "trival equation" isn't that trivial unless we are talking about trivial processors. Because non-trivial processors doesn't process a fixed number of instructions per second. So IPC isn't a fixed value.

    Id'd say you have a PhD in failing to debate, which means your posts unproductive. Something to work on - maybe it's an IPC problem?
    Nowhere did I claim that IPC is a fixed value - no architect worth their salt would say that IPC is a fixed value. The equation is trivial, not the stuff behind it. The point of that equation, particularly in current times, is to highlight how important IPC is.

    Just read H&P and maybe take an architecture class. Again, you think you know stuff, but you don't. Based on the discussion I have some pretty good idea about how little you know about the topic, actually (but there's Dunning-Kruger, of course).

    Leave a comment:


  • zyxxel
    replied
    Originally posted by vladpetric View Post

    I honestly couldn't care less about convincing someone like you.

    What's happening is that you don't understand microarchitecture (maybe you think you do, but you really don't). So you pull the discussion in a direction that you do understand. But, but, but, what about this? It really doesn't make your points valid.

    First chapter of H&P tells you about what determines the runtime of a simple process - a trivial equation involving instruction count, IPC, and clock cycle of the processor.

    Well, we live in a time where clock frequencies don't really change that much, and even instruction counts between different architectures don't vary that much. What's left is the IPC, which is the big elephant in the room. And yes, over there the difference between something like a Raspberry Pi and a lowly Pentium Gold is remarkable. Because of the microarchitecture, even a Pentium Gold can sustain much higher instruction throughputs than a RPi.

    The thing is - it's difficult to understand microarchitecture, so all the BSers focus on details that they can understand, things that are described with simple numbers.

    While I don't think you will, you really should read H&P.

    And, FWIW, I have a PhD in the topic with research work cited in microprocessor patents.
    What I do know is that you post bull. You have zero information about what I do - or do not - know. But that doesn't stop you from crapping in the thread.

    And you still failed to understand that if I add one point that is important, that doesn't invalidate the list of other important points that must be fulfilled to develop a great chip. That's on you to work on.

    Another thing - your "trival equation" isn't that trivial unless we are talking about trivial processors. Because non-trivial processors doesn't process a fixed number of instructions per second. So IPC isn't a fixed value.

    Id'd say you have a PhD in failing to debate, which means your posts unproductive. Something to work on - maybe it's an IPC problem?

    Leave a comment:


  • vladpetric
    replied
    Originally posted by zyxxel View Post

    I'd say you seem to be completely failing at keeping up a debate. Now try a better response where you actually argue.

    Hint here - there are more than one factor needed to make a good processor. So a good chief designer is one important factor. But that isn't mutually exclusive from having a very large walled needed to finance the work. And the current market situation affects the ability to be allowed to invest that huge amount of money.
    I honestly couldn't care less about convincing someone like you.

    What's happening is that you don't understand microarchitecture (maybe you think you do, but you really don't). So you pull the discussion in a direction that you do understand. But, but, but, what about this? It really doesn't make your points valid.

    First chapter of H&P tells you about what determines the runtime of a simple process - a trivial equation involving instruction count, IPC, and clock cycle of the processor.

    Well, we live in a time where clock frequencies don't really change that much, and even instruction counts between different architectures don't vary that much. What's left is the IPC, which is the big elephant in the room. And yes, over there the difference between something like a Raspberry Pi and a lowly Pentium Gold is remarkable. Because of the microarchitecture, even a Pentium Gold can sustain much higher instruction throughputs than a RPi.

    The thing is - it's difficult to understand microarchitecture, so all the BSers focus on details that they can understand, things that are described with simple numbers.

    While I don't think you will, you really should read H&P.

    And, FWIW, I have a PhD in the topic with research work cited in microprocessor patents.

    Leave a comment:


  • zyxxel
    replied
    Originally posted by vladpetric View Post

    Well, you seem to be completely ignoring what I'm telling you.
    I'd say you seem to be completely failing at keeping up a debate. Now try a better response where you actually argue.

    Hint here - there are more than one factor needed to make a good processor. So a good chief designer is one important factor. But that isn't mutually exclusive from having a very large walled needed to finance the work. And the current market situation affects the ability to be allowed to invest that huge amount of money.

    Leave a comment:


  • vladpetric
    replied
    Originally posted by zyxxel View Post

    There are other factors involved here. It costs huge amounts of money to develop a chip. AMD got robbed of receiving the incomes they should have from the Opteron processors because of illegal Intel marketing actions. That was money AMD should have used to develop their next line of chips.

    Right now, Intel have managed to run their company into a wall. Which means it has been much easier for AMD management to show that there is an open market to compete for. Which makes it much easier for AMD to get a go-ahead to go all-in developing newer implementations. As long as the competition has much better products and you have to settle for low-priced chips in the low-end segments, it's hard to motivate the owners to invest - how can you make them believe that you can not only match but beat the huge competitor?
    Well, you seem to be completely ignoring what I'm telling you.

    Leave a comment:


  • zyxxel
    replied
    Originally posted by vladpetric View Post

    I generally agree about management being an issue.

    But I think you're significantly underestimating the contributions of an engineering superstar, like Jim Keller. Why don't general ARM chips (not Apple's implementation) work as well as Intel/AMD? Because their superscalar microarchitecture is rather mediocre. And I'm sure they have good engineers.

    Between the Athlon/Opteron era and Ryzen, AMD sucked ... massively. Who was the lead with the first Athlon? Jim Keller. Who was the lead with the Ryzen? Oh, Jim Keller again. What happened in between? He was elsewhere ...

    The main issue with Intel right now is that they're a roughly a technology node behind AMD. Intel 10 nm is comparable to AMD 7 nm, in terms of things that actually matter (density of transistor/logic/memory - nanometers are not the size of a transistor, FWIW). But AMD's had it running at high volume for more than a year now, Intel's barely getting up to speed with it.

    P4 was an issue 18 years ago. It isn't now.
    There are other factors involved here. It costs huge amounts of money to develop a chip. AMD got robbed of receiving the incomes they should have from the Opteron processors because of illegal Intel marketing actions. That was money AMD should have used to develop their next line of chips.

    Right now, Intel have managed to run their company into a wall. Which means it has been much easier for AMD management to show that there is an open market to compete for. Which makes it much easier for AMD to get a go-ahead to go all-in developing newer implementations. As long as the competition has much better products and you have to settle for low-priced chips in the low-end segments, it's hard to motivate the owners to invest - how can you make them believe that you can not only match but beat the huge competitor?

    Leave a comment:


  • vladpetric
    replied
    Originally posted by zyxxel View Post

    Intel already has lots of great engineers - their problems isn't too bad engineers. Intel suffers a management issue.

    It was management - not engineers - who wanted about 30 steps pipeline in the P4 just hoping to have the highest GHz value when selling instead of having the fastest processor.
    It was management - not engineers - who wanted Intel chipsets to only support RAMBUS memory.
    It was management - not engineers - ...

    Intel engineers can do wonders - when they are allowed to fly. Most of the time, they are relegated to cost as little as possible while Intel dishes out the same processor with new coating to maximize the profit.
    I generally agree about management being an issue.

    But I think you're significantly underestimating the contributions of an engineering superstar, like Jim Keller. Why don't general ARM chips (not Apple's implementation) work as well as Intel/AMD? Because their superscalar microarchitecture is rather mediocre. And I'm sure they have good engineers.

    Between the Athlon/Opteron era and Ryzen, AMD sucked ... massively. Who was the lead with the first Athlon? Jim Keller. Who was the lead with the Ryzen? Oh, Jim Keller again. What happened in between? He was elsewhere ...

    The main issue with Intel right now is that they're a roughly a technology node behind AMD. Intel 10 nm is comparable to AMD 7 nm, in terms of things that actually matter (density of transistor/logic/memory - nanometers are not the size of a transistor, FWIW). But AMD's had it running at high volume for more than a year now, Intel's barely getting up to speed with it.

    P4 was an issue 18 years ago. It isn't now.
    Last edited by vladpetric; 06-01-2020, 03:42 PM.

    Leave a comment:


  • zyxxel
    replied
    Originally posted by vladpetric View Post

    Sure, but take a look at his entire track record. Intel should have hired him way earlier.
    Intel already has lots of great engineers - their problems isn't too bad engineers. Intel suffers a management issue.

    It was management - not engineers - who wanted about 30 steps pipeline in the P4 just hoping to have the highest GHz value when selling instead of having the fastest processor.
    It was management - not engineers - who wanted Intel chipsets to only support RAMBUS memory.
    It was management - not engineers - ...

    Intel engineers can do wonders - when they are allowed to fly. Most of the time, they are relegated to cost as little as possible while Intel dishes out the same processor with new coating to maximize the profit.

    Leave a comment:


  • zyxxel
    replied
    Originally posted by eydee View Post
    Proper approach would still be disabling all mitigations on all platforms and imprisoning those who exploit the vulnerabilities. It's a crime after all, in most countries.

    You don't stop selling knives because some people use them to kill. You don't lock cars to 10 km/h because some people cause accidents.
    You don't leave bank vaults unlocked just because the "right way" is to just imprison all thieves.

    What you are failing to understand is that Internet spans the whole world. And it's very complicated to try to catch and convict someone living in the "wrong" country. Next thing - what does it help that the person gets imprisoned, when the person has already created damages for billions of dollars?

    Security can only be proactive. That means the need for technical solutions. Going for the guilty party is a completely secondary step, but the only way to maintain security is to make systems as robust as is reasonably possible without making it too hard or inefficient to use the systems.

    Leave a comment:

Working...
X