Benchmark Results: Efficiency
Six cores are decidedly fastest at completing our efficiency workload.
Interestingly, average power consumption was lower on five cores than on four during our workload.
Total power used scales well. It's obvious that the many-core configurations require less total power to complete our workload.
As a result, the higher active core counts deliver better performance per watt.
11
Comments
Read more
Sponsored
Latest CPU News
Latest CPU reviews
- 24/05 – Overclocking Core i7-3770K: Learning To Live With Compromise
- 22/05 – Core i5-3570K, -3550, -3550S, And -3570T: Ivy Bridge Efficiency
- 15/05 – AMD A10-4600M Review: Mobile Trinity Gets Tested
- 14/05 – Core i7-3720QM: Ivy Bridge Makes Its Mark On Mobility
- 23/04 – Intel Core i7-3770K Review: A Small Step Up From Sandy Bridge
An ok article, but I would have been interested to see maximum overclock at each point, even if it was without a full set of benchmarks.
Now if only AMD can upgrade its L3 cache technology...
Why is the system idle power so much higher than the system peak power (until you get to 6 cores when it's a match)?
Why is the system idle power so much higher than the system peak power (until you get to 6 cores when it's a match)?
Ah, because they should be %age graphs but the legend is in Watts. Doh!
Is the diminishing performance improvement as cores are added perhaps due to the L3 cache memory bottleneck? This diminishing return on extra cores has been highlighted in many previous articles, I was just wondering how much of the story is explained by the memory bottleneck as opposed to other I/O resource bottlenecks.
Good to see that our article on Multi-Core efficiency on the 1090T here:
http://www.hardwarereview.net/Revi [...] -1090T.htm
has spurred others onto focusing on this crucial area given that raw clock speeds have hit a ceiling some time ago.
Keep up the good work (AT has the resources to run far more tests more thoroughly than we could)
It really shows that unless you are doing media encoding/transcoding, you will have little use for more than 2 cores.
But even with that application, adding cores beyond 3 or 4 starts to equal diminished returns. Clearly it is because it is hitting the memory bandwidth wall. Guess we need DDR5 RAM.
A triple/quad channel controller plus larger L3 cache should help a bit.
Nothing that we all here didn't already know about.
Is the diminishing performance improvement as cores are added perhaps due to the L3 cache memory bottleneck? This diminishing return on extra cores has been highlighted in many previous articles, I was just wondering how much of the story is explained by the memory bottleneck as opposed to other I/O resource bottlenecks.
In the Phenom II it clearly is a bottleneck (although a relatively minor one). But the Intel i7-980X saw its L3 cache increased by a third, just like its core count, and there still was a diminishing return. A smaller one, but still.
However, as for the Phenom II it's impossible. They could never have stayed within the 95W/125W power limit with additional cache (unless AMD would have opted for a die shrink, massively increasing development time). And that would have posed a problem as not all AM3 boards support 140W and none can go higher. A bad signal for a company that has backwards compatibility as one of its prime selling points...