Whats the deal with the Piledriver cores?

[ryanwyvill1]

So I'm thinking of getting the FX-6300 in the christmas sales but i heard that the new amd cpu's are just in this case 3 'real' cores with 3 threads. Will these threads act as 'real' cores or take any performance hit? And if it is like this then couldn't intel market their hexacore i7's as a 12-core processor?

 

[loneninja]

You should read one of the review articles on Piledriver, they explain what your asking for well.

Basically Piledriver is based up of modules, the 6300 has 3 modules each containing 2 integer cores that share a variety of resources. It has 6 physical cores but they take a performance hit because they all share resources within their module with another core.

 

[butremor]

Hi, check this thread, asked same question
http://www.tomshardware.co.uk/forum/352425-10-what-cores

 

[-Fran-]

Long story short:

AMD's "modules" approach is better than Intel's "Hyper Threading", but worse than another "real" core.

So, for the FX6300 you get 3 modules, which together form "almost" a 6 core CPU. Put it in another way: "It usually behaves like a 6 core, so it must be a 6 core", haha.

Cheers!

 

[majestic1805]

Each module has only one FPU so that may be the 3 cores you're talking about. However, as others have said, each module also has 2 integer cores so there are actually 6 cores and 6 threads.

 

[blazorthon]

AMD's modules in the upcoming Steamroller micro-architecture are supposed to have roughly 100% scaling like distinct cores do because it'll fix the front-end bottle-necks such as the insufficient x86 decoders. Until then, we have Piledriver and Bulldozer that technically are four, six, and eight core parts, but each module has a bottle-neck in that it has trouble feeding both cores with instructions, so it doesn't perform as well as it should.

Each module also only has one FPU instead of one per core, but that's not the same as an integer core of which there are two per module.

 

[richardgal]

If it's cheap enough with an AM3+ mobo, then go for it.
(You better have an AM3 cpu at home to flash the board with,these are much older than piledriver,but then again, if you already own a Phenom2 955/ 1090T I wouldn't recommend updating at all.)

If you can pick up a Core i5 3470/3570k cheap enough, (maybe with an M-ATX board) then it is definitely the way to go.

 

[jdwii]

AMD's modules in the upcoming Steamroller micro-architecture are supposed to have roughly 100% scaling like distinct cores do because it'll fix the front-end bottle-necks such as the insufficient x86 decoders. Until then, we have Piledriver and Bulldozer that technically are four, six, and eight core parts, but each module has a bottle-neck in that it has trouble feeding both cores with instructions, so it doesn't perform as well as it should.

Each module also only has one FPU instead of one per core, but that's not the same as an integer core of which there are two per module.

doubtful the Phenom II only had 92-93% scaling(with my processor) and this is with less sharing then even Steamroller. Right now CMT on average is about 75-80%(based on Piledriver benchmarks).

 

[blazorthon]

doubtful the Phenom II only had 92-93% scaling(with my processor) and this is with less sharing then even Steamroller. Right now CMT on average is about 75-80%(based on Piledriver benchmarks).

It depends on what you're doing, but scaling on Phenom II most certainly can be about 100%. Some things can more effectively make use of multi-core CPUs and I was referring to proper scaling as such. For example, programs that don't have much dependency between threads can scale very effectively even across multiple CPUs (that's CPUs, not cores in a single CPU) despite the fact that work with more dependency might be lucky to get over 50% scaling with two CPUs.

Point is that scaling would be more like individual cores than Bulldozer and Piledriver. Just because what software you run isn't better at scaling on your system (which might also be caused by issues other than just the program(s) being tested, such as too much background work being run) doesn't mean that Phenom II as a whole scales like that. The multi-CPU example above is an excellent example of how scaling depends on software factors much more than on the micro-architecture itself, at least with the micro-architectures that we've mentioned here.

 

[jdwii]

It depends on what you're doing, but scaling on Phenom II most certainly can be about 100%. Some things can more effectively make use of multi-core CPUs and I was referring to proper scaling as such. For example, programs that don't have much dependency between threads can scale very effectively even across multiple CPUs (that's CPUs, not cores in a single CPU) despite the fact that work with more dependency might be lucky to get over 50% scaling with two CPUs.

Point is that scaling would be more like individual cores than Bulldozer and Piledriver. Just because what software you run isn't better at scaling on your system (which might also be caused by issues other than just the program(s) being tested, such as too much background work being run) doesn't mean that Phenom II as a whole scales like that. The multi-CPU example above is an excellent example of how scaling depends on software factors much more than on the micro-architecture itself, at least with the micro-architectures that we've mentioned here.

This is true i am going by Wprime(using 1 core and then 6) and i also used Fritz Chess benchmark, Cincebench, And handbrake(encoding the same movie with the same settings with 1 core and then all 6 and then find the scaling factor.)

Under all times i only had 92-93% scaling but that number might even be higher with less cores being used instead of all 6. Using those same benchmarks Piledriver has around 75% scaling which is actually pretty good if you ask me i think that number will go up 10% but its still going to be lower then the Phenom's sharing which it should be since Steamroller is still a CMT design.

P.S
Like talking to you about other things besides Windows 8 since i usually always agree with you

 

[Hemanth Immanuel]

Long story short:

AMD's "modules" approach is better than Intel's "Hyper Threading", but worse than another "real" core.

So, for the FX6300 you get 3 modules, which together form "almost" a 6 core CPU. Put it in another way: "It usually behaves like a 6 core, so it must be a 6 core", haha.

Cheers!

So, if you consider an Intel Quad core with 8 threads and an FX series processor with 8 modules, and 8 threads, would FX perform better ?