Bulldozer Innovations Target Energy Efficiency

To everyone harping about AMD concentrating on efficiency and "forgetting" performance...you do realize that is the entire purpose of these ISSCC presentations, right? This conference is for improvements in transistor and logic design, not how fast it runs PiMark Vantage 2012.

Beside that, if you extrapolate what is presented and think about what it can impact it does have huge implications for possible desktop/server performance. What AMD has done is designed a completely new power distribution infrastructure for the cores.

Sure, this results in idle power savings and better power usage under low load which doesn't matter to you hardcore folders or media junkies so much. However, it also means is that they have the ability to crank up the power to very specific parts of the chip when necessary. I think this is why AMD has been so gung-ho about their new turbo in BD and how they will really try to make use of any available headroom if the software needs it.

Think of how far along we've come since the days of warranty voiding BIOS hacking on GPUs to overvolt and overclock compared to now where it is available in the stock driver. More recently in the AMD 6xxx series there is further evolution of built in performance control vs. TDP with Powertune. That's is the type of thing AMD is going for on the CPU side as well here. We might not know how fast it will wind up benchmark wise still, but these blogs at least make you think they have a solid foundation to work with.
 
Silent.Sin
This is awsome! Great inovations! Green power savings!
But will it beat Intel? That is what will sell CPUs. Frames per second at the highest settings sells CPUs, infrastructure be damned if it won't beat Intel's I7 extreme in Batman Arkham City next year.
 
griff30 said:
Silent.Sin
This is awsome! Great inovations! Green power savings!
But will it beat Intel? That is what will sell CPUs. Frames per second at the highest settings sells CPUs, infrastructure be damned if it won't beat Intel's I7 extreme in Batman Arkham City next year.
Click to expand...

I don't think 'beating Intel' actually sells CPUs like you think, more chips sell in OEM boxes than anywhere else. If AMD sells a processor that: is cheaper than the Intel equivalent; consumes less power, therefore generating less heat; Allows them to use cheaper PSUs and motherboards; AMD will move volumes of chips. OEMs aboslutely do not care about whether or not AMD can beat Intel and vice-versa, they only care about total cost.

The main people who care about benchmarks are not the people who will buy processors in lots of thousands or millions, like OEMs, the kinds of customers that actually keep AMD around.
 
griff30 said:
Silent.Sin
This is awsome! Great inovations! Green power savings!
But will it beat Intel? That is what will sell CPUs. Frames per second at the highest settings sells CPUs, infrastructure be damned if it won't beat Intel's I7 extreme in Batman Arkham City next year.
Click to expand...

We'll have to wait and see, AMD have been extremely tight lipped about letting any performance numbers slip for any of their products (mostly GPU) lately. That has been the case for mediocre products as well as ones that have slapped the competition silly. You can't judge expected performance based on lack of leaks or info about other CPU features.

Something as far away as next year (Batman AC is still a 2011 title AFAIK) is even harder to tell. At that point Intel's EE might be based on Ivy Bridge which we have no numbers for, and BD will be ramping up clockspeeds and also will get core improvements, possibly even a bump to 10 cores, with Komodo at some point during 2012. [source]

+1 to JMccovery's post.
 
amdgamer said:
I dunno, because a lot of those guys run some crazy overclocks that operate the processor well outside of the window of efficiency.
Click to expand...

Serious/dedicated folders avoid running outside the efficiency window like the plague. Crunching 24/7 with bad settings can cost enough to buy two rigs to run at good settings. Overclocking != inefficient by itself. Most CPU's can run at much higher frequencies using the stock voltages, and/or can run at much lower voltages running the stock frequencies.

I personally find how low a voltage my CPU's can run at stock speed. Then I find the minimum voltage for each bump, measuring power consumption. If you plot it, there is a clear place where the consumption increase versus clock gains is nowhere near linear, so you clock it right before that point. The power increase is never linear, but there is a definite cut off where diminishing returns snowballs.

On topic: Even if BD isn't the best IPC, if it can complete the EXACT same task using LESS wattage, it will become the replacement for i7's which currently are untouched in performance/watt.
 
DedEmbryonicCe11 said:
What partial core? The "80%" that people are so confused about isn't a detriment... it's the lack of a benefit that BD can have when using only one core of a module.
Click to expand...

"The module, described as two cores, can be compared to a single Intel core with HyperThreading. The difference between the two approaches is that Bulldozer provides dedicated schedulers and integer units for each thread, whereas in Intel's core each thread can access all available resources, except for the individual thread state information."

So the second thread on a BD core should be able to give you about 80% of the speed that the first thread can.

First core = 100%, second core = 80%

On Intel's chips, HT only gives you around 20% boost.
So.. core = 100% , second thread (HT) 20%.

Is that a bit clearer now?

And of course you won't get a benifit at all if you aren't using more than one thread, but that is true of any CPU that has multiple cores and/or can run multiple threads per core.
 
A Bulldozer module and a hyper-threaded core are not similar enough to compare to each other. They don't work in the same way, in fact they are almost opposite of each other.

A HT core is a single core that can schedule a second thread. The goal is to take advantage of the fact that usually, a single thread isn't going to utilize all of the computational hardware that a core has, so you might as well try and schedule more work to use the idle parts of the core each cycle.

A Bulldozer module contains two completely separate cores. The "module" concept comes in when those cores don't have their own floating point hardware, and they use the "modules" FP units. If both threads are running at 100% they are going to each take up half the FP units and act like 2 normal cores. If one core is just hammering away at integer math, or it's just idle, the other core can use ALL of the modules FP abilities and do a 256-bit instruction or two 128-bit instructions. It's not 2 "pseudo" cores, its 2 fully functional cores that happen to have the ability to schedule even more FP work if the other core in its module isn't using all of that resource up.
 
cyclone3d said:
"The module, described as two cores, can be compared to a single Intel core with HyperThreading. The difference between the two approaches is that Bulldozer provides dedicated schedulers and integer units for each thread, whereas in Intel's core each thread can access all available resources, except for the individual thread state information."

So the second thread on a BD core should be able to give you about 80% of the speed that the first thread can.

First core = 100%, second core = 80%

On Intel's chips, HT only gives you around 20% boost.
So.. core = 100% , second thread (HT) 20%.

Is that a bit clearer now?

And of course you won't get a benifit at all if you aren't using more than one thread, but that is true of any CPU that has multiple cores and/or can run multiple threads per core.
Click to expand...
I understood what you were trying to say the first time, but you're still looking at it wrong. Your "80%" is really the baseline, AKA 100% in multithreaded code while a single thread operating on a single core of a module whose other core is idle can operate faster than that baseline.
 
I would love to see performanch benchmarks and see how they compare to Sandy Brindge. In addition I'd like to know when Bulldozer Launches? March? April? :confused:
 
adonn78 said:
I would love to see performanch benchmarks and see how they compare to Sandy Brindge. In addition I'd like to know when Bulldozer Launches? March? April? :confused:
Click to expand...

word, the waiting game is getting really old. Intel needs some competition to keep their prices in check. With this latest information of them focusing on efficiency I have a feeling (like many of you) that this won't even begin to compete in the high end segment.
 
You must log in or register to reply here.
Back
Top