Beror på vad man gör med.
har man 20 program som kan utnyttja t.ex 4 kärnor så har du nytta av 80 kärnor.
Bara OS har ett 100 tal processer igång som kan spridas ut över flera kärnor.
Webläsare skulle kunna skicka varsin öppen sida till egen kärna osv.
Och vore rena drömmen för saker som kompilering, virtualisernig och såna saker.
Tänk att kunna kompilera en kernel med make -j80
Sen ang problemen med prestanda med dom nya CPU/APU från AMD så förklarar den här texten det ganska bra.
The Bulldozer chips have got an acceleration built into them called FMA4... This is an acceleration that boosts floating point performance by a solid 20% and is not enabled by 99.9% of binaries for Windows because there are no compilers for Windows yet that can create binaries that use FMA4.. The only compilers out there that can enable FMA4 floating point accelerations is the one that AMD made called Open64, and also there have been some highly experimental patches to the Open Source GCC compiler for FMA4 as well.
In addition to that, the core structure of Bulldozer is not like any processor that has come before it.. Because of this,. the Windows kernel, in the presence of two threads that have work that need to be done, is incorrectly assigning both these threads to the same Bulldozer module. Since Bulldozer modules have some shared hardware components these threads will run slower when assigned to the same Bulldozer module than if they were assigned to different Bulldozer modules on the same CPU.. Again, this is Microsoft's thread scheduler incorrectly assigning work to be done to the wrong cores which is considered a failure in the operating system's thread scheduler since it's the operating system's duty to ensure that the hardware is used optimally. Microsoft is working on a hotfix for Windows 7 (they made one, but then recalled it because they screwed it up) and will have it fixed in Windows 8.. The Linux kernel already has it fixed and is already in the release candidate for the next stable Linux kernel release.
Also Microsoft will likely have an FMA4 compiler ready by the time Windows 8 is released.. Which means that people will start using that compiler to compile software and magically all the newer executables compiled will run 20% faster than they had before on Bulldozer chips.
AMD has explained this.. I've researched this and explained this.. The people who write compilers (GCC Team, Open64 Team, etc.) has explained this.. The Linux kernel developers have explained this... Microsoft has explained this in a hotfix bulleteon.. Yet people don't understand or read any of it, so I really don't care anymore.
So 20% boost of performance from FMA4 when binaries are compiled with a compiler that supports it and on top of that, an additional 10% boost in performance for apps that use less than 4 threads but more than 1 thread.