Looking for a heavier-duty task, I also threw in our standard encoding test, where we take a 30GB, p MKV file and use Handbrake to transcode, using the Android for Tablet profile.
On dual-core machines, it takes in excess of two hours to complete. This test tells us two really useful data points. The first is how well a particular CPU performs in this heavily multi-threaded test.
The second data point you get from this test is how much the design of a laptop suffers from thermal throttling. If you look at the result from the HP Spectre x and compare it to the Dell XPS 13 that has the same CPU, you also see the performance drop-off to aid and abet the cooling capability of the laptop or the decision by the laptop maker not to heat it up too much.
Both use the exact same cooling system, from what I could tell looking inside both. The Broadwell technically loses here, by less than 1 percent.
That shows up in 3DMark, where the Broadwell-chipped Dell offers up about 10 percent more graphics performance in Cloudgate. I even saw a percent difference between the two when the graphics load is scaled back further to 3DMark Ice Storm Extreme.
My last benchmark chart for the mobile dual-cores is the most important one for the applications in an ultra-portable laptop. All within the performance envelope of a dual-core processor. What people care about most is the battery life. The test even lets the laptop go to sleep for a few minutes, when the typical office drone might be checking their phone or chatting to the person in the next airplane seat.
Right from the outset, the Haswell microarchitecture was designed from the ground up to be a mobile-focused chip. One of the biggest additions in Haswell is the full integration of the voltage regulator onto the CPU die. Haswell idle power consumption [Image credit: Anandtech]. The fully-integrated voltage regulator and higher power consumption under load mean that Haswell is actually worse than Ivy Bridge at overclocking.
In reality, the picture is even muddier than that: Early reports suggest that some Haswell chips can only reach 4. If we look at the bigger picture, though, overclocking has been slowly dying for a decade. Each one consists of 10 EUs, an instruction memory and a radically redesigned texture sampler with four times the throughput of Ivy Bridge's.
Thanks to an extremely wide bit interface — RAM is usually only connected via two 64 bit channels - the bandwidth of the GPU is very large. We also want to briefly mention another feature of the eDRAM here. So, you could also call it a very big "off-die" L4 cache.
This could bring same performance advantages in special tasks which depend heavily on bandwidth. There were also several changes in the feature-set of the GPUs. The graphics chip can still drive up to three displays in different configurations with a maximum resolution of 4k x 2k see the following figures for details.
The available interfaces are HDMI 1. The new Wireless Display version 4. Finally, the Quick-Sync transcoder has also been redesigned and its performance improved. The mobile chips that include the Core iMX the fastest chip in the roster directly replace all the processors from the Ivy Bridge generation with similar clocks and L3 cache.
Meanwhile, the HQ models are completely new. Although the model number would suggest higher performance e. Core iHQ vs. Only the Core iHQ will offer the fastest graphics solution. Schenker kindly made a barebone available to us for our CPU tests.
Detailed information about the notebook, benchmarks with dedicated graphics card, etc. Before we come to the results, we have to mention a limitation of our test system. The Clevo PSM and other Haswell barebones from the Taiwanese manufacturer still had problems allowing the maximum Turbo frequency to be reached during multi-threaded operations at the time of testing.
As a result the performance is 3 to 9 percent lower and we hope that this problem will be solved in the final devices. We have still mentioned the results even with this problem. At least, they allow for a rough classification of the Haswell CPUs. We had no issues with the single-threaded tests with maximum Turbo frequencies possible with the device. Because of the aforementioned turbo problem, our results are not a clear representation of the capabilities of the Haswell platform.
We were very impressed by the three fastest chips which matched the performance figures of the Core iXM which is clocked about MHz higher. The performance-per-MHz increased by about 10 percent on average. In the single threaded test of Cinebench R If you take the different clock rates into account, the performance gain is about the same in most other applications.
But we need to mention that we did not see any significant improvements in the Hyper-Threading technology. However, performance gain smaller than 10 percent is still very small. But, you cannot expect more from this architecture without significantly changing the process or using much higher frequencies.
We will probably have to accept that the performance per core will only marginally increase in the coming years. At least the buyer gets the performance of a significantly more expensive Ivy Bridge CPU for the same price - and we did not consider the new instruction set extension of the Haswell models here.
The difference might be significantly bigger in newer software which supportsAVX2 etc. Due to the new socket of the Haswell CPUs, we had to use a different test system and cannot directly compare the power consumption with Ivy Bridge. In addition, we have to mention the turbo problem again.
It does not only have an impact on performance, but also on power consumption. Therefore, we also mention the used frequencies in all following measurements. Like in the previous tests, we have been able to measure the total power draw only. We will not be able to specifically evaluate the relative power-draws in absolute values but we will try to give you a picture of the extent of energy consumption improvements in the platform by eliminating as many variables as possible.
For the best possible comparison, all measurements of the energy demand were done with maximum display brightness, active WLAN, and the energy profile "maximum performance". While idle, the power consumption is rather high at about 26 W, but the chips do not differ much.
0コメント