Volodymyr Zubov #: The first hypertrading Intel made in the days of Pentium4, it was unsuccessful, so it is unsuccessful now, there is just a branching on the conveyor of tasks - the conveyor is even slower. And the first dual-core is the Intel Core 2 Duo.
What do you know about it? It's not efficient on Intel,
Each processor has a pre-convector - a handler. In hyper-trading it tries to guess the next task and to make the kernel execute it, the percentage of guessing is 0-5% then the convector is cleared and the programme is executed.
I'm talking about the Managers, they talk about hyper trading like it's two cores, in fact it's two convolutions and if you make a mistake, you lose performance.
Volodymyr Zubov #: That's why Agents don't take this hyper-trading into account.
it's actually a memory problem, they buy a bunch of cores and 16gb of memory.
I have a 16 thread robot using about 3gb per thread.
7 characters on each tick. in normal work CPU load 1-3% only, but memory in the tester really consumes a lot, 32 for 16 threads was not enough, and it depended on the period, in some months it was impossible to make a test, everything stood up at a standstill.
The first hypertrading Intel made in the days of Pentium4, it was unsuccessful, so it is unsuccessful now, there is just a branching on the conveyor of tasks - the conveyor is even slower. And the first dual-core is the Intel Core 2 Duo.
What do you know about it? It's not efficient on Intel,
in AMD the threads work 50/50 almost.
What do you know about it? It's ineffective at Intel,
AMD threads are 50/50 almost.
And I'm not advertising here, just technical aspects.
And I'm not advertising here, just the technical aspects.
Of course you do, you were kicked a long time ago.
hypertrading is considered unsuccessful at intel, amd has no complaints.
actually maybe something changed in the new models.
That's why Agents don't take this hyper-trading into account.
it's actually a memory problem, they buy a bunch of cores and 16gb of memory.
I have a 16 thread robot using about 3gb per thread.
7 characters on each tick. in normal work CPU load 1-3% only, but memory in the tester really consumes a lot, 32 for 16 threads was not enough, and it depended on the period, in some months it was impossible to make a test, everything stood up at a standstill.
no problems with 64gb.