Settle down everyone at least AMD is trying.

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Aug 11, 2008
10,451
642
126
it's not rocket science, when intel keeps a 5-10% increase pr tick tock and amd does a 10-15%. While slower atm (and not counting power), amd will catch up .. even on dozer arch.

As someone else stated in another thread, it is easier to improve from an inefficient state (to put it charitably) than to improve from a highly efficient state. Overall, I still feel intel is the superior CPU. It is more well rounded when you consider single threaded, multithreaded, gaming and power use. Besides, there is no assurance that AMD will continue 10 to 15 percent increases with each new processor.

Intels focus now (unfortunately IMO) is increasing efficiency per watt rather than raw processing power. In this metric amd still trails badly, and will probably be even further behind when Haswell comes out. It would be nice if Piledriver's improvements in multi-threaded apps would force Intel to bring out a mainstream priced hex core haswell, but I doubt that will happen. In fact intel may even be ceding some segment of the market to AMD partially because they do not want them to go out of business.
 

cytg111

Lifer
Mar 17, 2008
25,663
15,164
136
As someone else stated in another thread, it is easier to improve from an inefficient state (to put it charitably) than to improve from a highly efficient state. Overall, I still feel intel is the superior CPU. It is more well rounded when you consider single threaded, multithreaded, gaming and power use. Besides, there is no assurance that AMD will continue 10 to 15 percent increases with each new processor.

Intels focus now (unfortunately IMO) is increasing efficiency per watt rather than raw processing power. In this metric amd still trails badly, and will probably be even further behind when Haswell comes out. It would be nice if Piledriver's improvements in multi-threaded apps would force Intel to bring out a mainstream priced hex core haswell, but I doubt that will happen. In fact intel may even be ceding some segment of the market to AMD partially because they do not want them to go out of business.

- Why yes it is! For now. Intel will fight a battle downhill to conquor the tablet and smart phone niche. And most likely fail. While failing at that -making x86 power effecient competetive with arm- they leave a niche for amd to catch up in single threaded AND provide the core count to boot.
 

Ben90

Platinum Member
Jun 14, 2009
2,866
3
0
The entire reason why Intel has such good IPC is because they have focused on power efficiency. Race to idle is the name of the game in the growing ultraportable market. Intel understands power efficiency revolves around keeping your execution units well fed, not just adding more cores. Pretty much every increase of IPC from Intel since Prescott has been in the name of power efficiency.
 
Last edited:

cytg111

Lifer
Mar 17, 2008
25,663
15,164
136
The entire reason why Intel has such good IPC is because they have focused on power efficiency. Race to idle is the name of the game in the growing ultraportable market. Intel understands power efficiency revolves around keeping your execution units well fed, not just adding more cores. Pretty much every increase of IPC from Intel since Prescott has been in the name of power efficiency.

Summed down; the reason intel has good ipc is because they dont suck much wattage.
Sorry. Still dont get it.
 

Ben90

Platinum Member
Jun 14, 2009
2,866
3
0
The two are linked more closely than you realize. By eliminating stalls in the pipeline you gain both efficiency executing instructions and power efficiency by extension.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Summed down; the reason intel has good ipc is because they dont suck much wattage.
Sorry. Still dont get it.

Focus on improving power consumption results in a natural progression towards improving the IPC. Why bother having the clock (uses ~25% of total package power) if you aren't doing instructions in the clock?

You either reduce the clock, to reduce its power footprint, or you increase the work that gets done within the clock, increasing the IPC.

Only one of those two options results in improved performance and improved performance/watt. The other option only results in improved performance/watt. Why settle for one if you can have both?
 

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
833
136
- Why yes it is! For now. Intel will fight a battle downhill to conquor the tablet and smart phone niche. And most likely fail. While failing at that -making x86 power effecient competetive with arm- they leave a niche for amd to catch up in single threaded AND provide the core count to boot.
Do you think that Intel would have any trouble releasing a 6 core version of their CPU's on 14nm and beyond?

The only reason they don't at the moment, is because they make more profits on this current node, using Quads as their spearhead.

The notion of AMD having any kind of meaningful advantage on core counts going forward, is crazy. AMD here is at the whim of Intel's desire for profits.
 

anongineer

Member
Oct 16, 2012
25
0
0
Focus on improving power consumption results in a natural progression towards improving the IPC. Why bother having the clock (uses ~25% of total package power) if you aren't doing instructions in the clock?

You either reduce the clock, to reduce its power footprint, or you increase the work that gets done within the clock, increasing the IPC.

Only one of those two options results in improved performance and improved performance/watt. The other option only results in improved performance/watt. Why settle for one if you can have both?

From what I have seen, the clock network accounts for at least half of the total power consumed in a block. Much of that power is consumed by flip-flops in pipeline stages. Doing more work per cycle means less pipeline stages, means less flops, means less clock power consumption.

Of course if it were so easy then every single register to register instruction would take only one cycle.
 
Mar 10, 2006
11,715
2,012
126
Some simple math folks...

Improving IVB by 10% is not the same thing as improving Bulldozer by 10%. One of these is much higher performing to begin with, so that "10%" improvement on the faster chip is much more in absolute performance terms than on the slower chip.

Also, Intel isn't really gunning for brute, raw performance. It is aggressively and relentlessly trying to maximize performance/watt. And that's the right thing to do.

Go to Dell, Lenovo, or HP's websites. See all the new Windows 8 products? All running Intel. Why? Well, Intel's just a lot better at sipping power than AMD's low power parts are.

Yeah, FX-8350 can sorta match the i7 3770K...with 2x the power and no IGP.

That being said, AMD's FX stuff seems great for a gaming PC on a budget. Also, the Llano/Trinity chips rock for non-ultrathin style laptops (I own one).