Even with 4 way SLI ?...
An then, if only a few cores are at work, it s just more amazing
since BD make use of a single module for 2T and 2 modules for 3 or 4T
while a 990X make use of a full core for each thread in such case...
It makes me wonder... if the L2 cache really is disabled, yet it performs on par with an i7-990, that is very very good news for AMD. It makes me wonder if they did not disable it purposely, just to hide its awesomeness from intel.
No. Windows will not schedule two threads on the same module if it doesn't have to.
She also hinted that scheduling two threads on the same module and leaving the other three modules idle, so they cold drop into a low-power state, might be the best path to power-efficient performance.
But if two threads are to be scheduled, they will be sent in a single
module , not to two modules..
No. Windows will not schedule two threads on the same module if it doesn't have to.
Seems that BD is rather late :
For the high-end Scorpius platform, AMD will announce the AM3+-based FX-8150, FX8100, FX6100 and FX4100 by the end of September with FX8170, FX8120, FX6120 and FX4140 set for the first quarter of 2012.
http://www.digitimes.com/NewsShow/NewsSearch.asp?DocID=PD000000000000000000000000020423&query=AMD
What part of 'no' didn't you understand? Windows will schedule threads across modules before it puts two threads on the same module.
On an eight core BD, each module will get one running thread until the thread count gets to five, then modules will get multiple threads.
Putting more than one thread on a module when it's not necessary will result in a performance decrease, so why would it get scheduled that way?
It's exactly the same as Hyperhtreading, Windows will schedule threads so only one thread lands on a core, until you are out of physical cores.
It s not the same as hyperthreading , and that s why you misled yourself.
The OS will understand if your machine is setup for maximum performance or for maximum performance/watt which takes advantage of Core Performance Boost.
Didn't JFAMD state the exact opposite? That module usage will be will be kept as low as possible so they can be gated off, to get the most out of turbo?
Since AMD Phenom II X6 1100T can beat Intel Core I5 2500K in almost all multithreaded applications at the same price point, i would say BD will be much faster at the same apps even than SB 2600K.
I'm sure you all would like a link to support what i say,
http://www.anandtech.com/show/4310/amd-phenom-ii-x4-980-black-edition-review/1
No mater how fast your IPC will be, more cores are better in multithreaded applications and since BD will have both higher IPC and more Cores (than Phenom II) we really have to wait and see the real silicon and what is capable off.
So is AMD being a “big baby” for quitting BAPCo, or has Intel’s purported influence over SYSmark just reached the breaking point for other semiconductor companies? Either way, there will now be even more scrutiny over which benchmarks reviewers use to measure new processors, like AMD’s Bulldozers, and what those results are. Obviously, AMD needs its Llano and Bulldozer chips to test as competitively as possible to blunt Intel’s sales advantage with its Sandy Bridge processors. But what “competitively” means and how it’s measured will now draw more attention than ever.
Thanks, I looked it up, per my previous post.
JFAMD said:Yes, people are starting to come around. Everyone was getting all wrapped around "how do I spread my threads out across modules so that I have one thread per module. Yes, you get a performance increase with that, but it is marginal. However, running threads on the same module would allow for a.) sharing of L2 cache for apps that are utilizing the same data set and b.) the other modules to be shut down, reducing power and increasing the ability to boost.
Lots of people don't get it. You have a maximum amount of power that the processor can consume. You may be better off concentrating the power on fewer modules to achieve higher clocks than try to spread threads out to get 100% of the L2 resources.
What I want to know is will we be able to configure windows so that it schedules one thread per module, or two threads per module. I would like to be able to try both and see which works best. It is an interesting choice.
a) Two modules for two threads = more floating point resources per thread.
b) One module for two threads = more potential clock speed.
I still think option "a" would be faster.
What I want to know is will we be able to configure windows so that it schedules one thread per module, or two threads per module. I would like to be able to try both and see which works best. It is an interesting choice.
a) Two modules for two threads = more floating point resources per thread.
b) One module for two threads = more potential clock speed.
I still think option "a" would be faster.
What I want to know is will we be able to configure windows so that it schedules one thread per module, or two threads per module. I would like to be able to try both and see which works best. It is an interesting choice.
a) Two modules for two threads = more floating point resources per thread.
b) One module for two threads = more potential clock speed.
I still think option "a" would be faster.
Originally Posted by AtenRa
I don't remember where i read this but like Intel's HyperThreding, the second thread to be executed is going to the next free core and not to the HT core (when there is one available) does the same happens with BD modules ???
When we have one thread to be executed, then core 1 of Module 1 will be used, but when we have a second thread then Core 1 of the Module 2 will be used and so on.
So this 80% performance is only applicable when all cores are being used, for example if we have a dual Module BD and we have to execute 4 threads. ???
Originally Posted by Dresdenboy
This is a power efficiency problem and might be handled by a new scheduler in Windows 8 (I heard something like that). So far I'd say since AMD's cores are seen as physical cores, the OS currently wouldn't consider the module architecture. Power wise a further activated module (to run one additional thread) would instantly cause sth. like ~35% of max. power due to leakage and activity power levels (according ISSCC paper). This might be lower as long as CPU load is low and frequency+voltage could be lowered. But while this 35% (or 10-20% w/ lower voltage/freq) would mean a lot, the same thread running on the first module's second core might cause a lower power consumption increase. So Turbo might help a bit. But overall this is not easy to answer. It might depend on energy saving policies etc.
Is this for real, I hope so?
Great breakthrough: AMD FX @ 5,1 GHz on AIR!
http://obrovsky.blogspot.com/2011/07/great-breakthrough-amd-fx-51-ghz-on-air.html