Turbo Boost Max 3.0 Patches Updated, Now On By Default For Multi-Socket Systems

[computerquip]

you are kidding right? the future is cpu and gpu working together using multitheatings

There seems to be some confusion. The intention of the message was to help understand that achieving efficient multi-threading is not easy at all. When multi-threading in a naive manner, it can even be slower than a single-threaded application. You *must* consider things like cache lines and boundaries, which most people don't. Scott Myers gives a good presentation on this: https://vimeo.com/97337258

[labyrinth153]

Starshipeleven is exactly right.

[starshipeleven]

unapproved post for andre30correia

[starshipeleven]

you are kidding right? the future is cpu and gpu working together using multitheatings

Uhhh, no, the future is CPU still doing most single-treaded/small-multithreaded jobs, while multithreaded jobs get offloaded to something that is designed for that like a GPU.

If all loads were massively multithreadable then why the fuck we still use CPUs at all? Even an APU's GPU has like a thousand of cores (stream processors)

[labyrinth153]

If by only certain jobs you mean almost all then yes.

Ugh. Go suck a log you nitwit.

[SaucyJack]

That is not how it works. Only certain jobs even benefit from parallelism.

If by only certain jobs you mean almost all then yes.

[labyrinth153]

you are kidding right? the future is cpu and gpu working together using multitheatings

That is not how it works. Only certain jobs even benefit from parallelism.

[andre30correia]

In addition to this, a lot of people overestimate multi-threading. They sometimes think that simply throwing multiple jobs onto multiple threads will equal a linear increase in the amount of work done. Believe me when I say that there are cases in the programming world where a single-threaded application can outperform a multi-threaded application due to the ability of the program to better manage cache, handle resource sharing (or lack thereof), and less complicated design.

you are kidding right? the future is cpu and gpu working together using multitheatings

[computerquip]

It isn't the CPU that needs to be made better at parallel tasks. It is entirely about the type of workload. Parallelisation requires that many things can be calculated at once without being dependant on the results of other things too much. It just happens to be the case that graphics rendering tasks can be done more or less independently of one another.

In addition to this, a lot of people overestimate multi-threading. They sometimes think that simply throwing multiple jobs onto multiple threads will equal a linear increase in the amount of work done. Believe me when I say that there are cases in the programming world where a single-threaded application can outperform a multi-threaded application due to the ability of the program to better manage cache, handle resource sharing (or lack thereof), and less complicated design.

[arbition]

Bad programmers? Bad programming languages? Bad compilers?...
at parallelization at least? Why?

How can CPUs be better at parallelize tasks? Why GPUs are such parallelization beasts?

It isn't the CPU that needs to be made better at parallel tasks. It is entirely about the type of workload. Parallelisation requires that many things can be calculated at once without being dependant on the results of other things too much. It just happens to be the case that graphics rendering tasks can be done more or less independently of one another.