Performance and Power Investigated

Given all the performance packed into the i7-2820QM, worst-case heat and noise levels should still be similar to what we encountered with Clarksfield. Idle power is good, but if you want to do some heavy processing or gaming what happens? We connected the Compal system to a Kill-A-Watt device to check power draw under various loads, as well as testing battery life while looping a graphics intensive application. We’ve seen NVIDIA and AMD GPUs really curtail performance on DC power, but has Intel done the same?

We’ve created a table of power draw at the outlet for several usage scenarios, as well as the calculated power requirements on DC based on the 71Wh battery. We’ve also included the performance figures for the tasks where applicable, to see if performance throttling is in effect when on battery power. We used the “Balanced” power profile for the AC tests, and the Power Saver profile (but still allowing the CPU to go to 100%) for battery tests. For the graphics test, we enabled the “Maximum Battery Life” setting as well as the “Balanced” setting—the graphics tests on AC were done using the “Maximum Performance” setting.

Power/Performance Under AC/DC
  Power at Outlet /
Calculated DC Power
Performance
Idle 12-13W N/A
Idle (DC) 9.04W N/A
Internet 14-31W N/A
Internet (DC) 10.24W N/A
3DMark06 48-70W 5285
3DMark06 (DC) MaxBat 23.61W 2800
3DMark06 (DC) Balanced 41.18W 5184
H.264 Playback 20-21W N/A
H.264 Playback (DC) 16.38W N/A
Cinebench 11.5 SMP 70-89W 5.72
Cinebench 11.5 SMP (DC) 59.17W 5.09

Watching power draw and CPU clocks (using CPU-Z) during the tests was rather interesting. There’s not much going on in the idle test; looking at the numbers, AC power use is about 36% higher than the DC calculated power use. Most likely extra power-saving features are in effect under DC power.

In the Internet test (under AC), while the web pages are loading the system used anywhere from 18-31W. Once all four pages have finished loading, however, power would settle down to 14W—just slightly higher than the idle power draw. That’s quite impressive, given the Flash content on the active page, and that’s reflected in the only slightly higher calculated power draw for Internet battery life vs. idle. Also of note is that the CPU clock speed never even hit 2.3GHz—let along the maximum 3.4GHz—during the Internet test, at least not that we could detect. We could see it reach 1.6GHz for a few seconds, and then it would settle back to 800MHz.

The H.264 playback test is another example of low CPU clocks and utilization through the test. The initial loading of the x264 movie would bump clock speeds up, but then the CPU would drop back to the minimum 800MHz and stay there. Power draw is definitely higher than the idle/Internet tests, but 20-21W isn’t too shabby for a 17.3” notebook. And then we get to the power hungry tests, simulating gaming and heavy CPU use.

3DMark06 power requirements are generally similar to gaming results, with the wide spread being typical. Tests 1, 3, and 4 averaged power draw closer to 53W, while test 2 (the Firefly Forest) was nearly 10W higher on average. Turbo Boost—on both the CPU and GPU—is very likely in play, but we didn’t have a good way of measuring real-time clock speeds during the tests. We tested battery graphics performance using two settings; first is the “Maximum Battery Life” setting, which results in roughly half the performance compared to running on AC. The second mode is labeled “Balanced”, which improves the score quite a bit—at the cost of power consumption.

Based on the 3DMark06 results, plugging in improves graphics performance by 2-82%, depending on what graphics power saving setting you select. You’ll definitely want to run the higher performance GPU mode if you actually want to play games, as otherwise frame rates will drop into the low 20s or upper teens on most titles. With the “Balanced” or “High Performance” GPU setting, gaming performance is reasonable even on battery power, but it puts enough of a load on the battery that you won’t be able to last more than around 90-100 minutes. If you happen to have a game where you only need the power saving performance mode, though, you should be able to get gaming battery life up to three or perhaps even four hours (depending on the game).

Finally, we’ll wrap up this discussion by looking at maximum CPU loads. In the Cinebench test, quad-core Turbo is interesting to watch; running the CB11.5 SMP benchmark, at first all of the cores start at the maximum 3.10GHz speed—blisteringly fast for a notebook! About 11 seconds in to the test, the core speed drops to 3.0GHz, where it remained until 39 seconds; then it dropped to 2.9GHz, and at around 54 seconds the speed dropped briefly (1-2 seconds) to 2.8GHz before settling in at 2.7GHz for the remainder of the test. If you happen to run heavily-threaded benchmarks continuously, the first run will usually show about 10% higher performance thanks to the initial thermal headroom, but the lowest Cinebench SMP and x264 encoding scores that we measured are still within 10% of the maximum score, which is very impressive for notebook hardware.

At the highest point in the test, power draw for the notebook peaked at 89W; once the speed settled at 2.7GHz (which it appears the notebook could sustain indefinitely in our 70F testing environment), power draw was steady at 70W. Switch to battery power and the Power Saver profile, and performance did drop slightly but not as much as you’d expect. We measured 5.09 PTS while running off the battery, so plugging in nets you up to 12% better performance. Like gaming, battery life under a heavy CPU load is going to be much lower than our other tests, and we measured just 72 minutes. Then again, compare that with some of the other high-end notebooks we’ve looked at in the past, which managed a similar 72 minutes with no load whatsoever.

One thing to keep in mind is that the effectiveness of Intel’s Turbo Boost technology does depend on the cooling equipment. While the Compal sample runs reasonably cool—we’ll check temperatures on the next page—we have definitely seen larger, more robust cooling solutions. The profile of the Compal chassis is generally flat, so that limits the size of the fan(s) and the amount of airflow. Something like the ASUS G73 chassis has proven quite effective at running high-end mobile components in the past, and we suspect that better cooling will result in the CPU running closer to the maximum Turbo limits more of the time. We’ll have to wait for sample notebooks to confirm our suspicions, but we’ve seen it in the past with Clarksfield and Arrandale, so there’s no reason Sandy Bridge would behave differently.

All the Performance, and Good Battery Life As Well! What About Heat, Noise, and the LCD?
POST A COMMENT

66 Comments

View All Comments

  • tipoo - Monday, January 03, 2011 - link

    Sorry if I missed this somewhere in the review, but does the graphics component support OpenCL? Reply
  • RyuDeshi - Monday, January 03, 2011 - link

    Second to last paragraph on the "Extended compatibility and performance results:"

    "Ultimately, Sandy Bridge’s IGP is far more capable than many would have expected. Sure, it doesn’t even try to support DX11 or OpenCL, but at least for gaming DX11 is typically too much for even midrange GPUs."
    Reply
  • CharonPDX - Monday, January 03, 2011 - link

    An Intel rep has said that Sandy Bridge will support OpenCL. (http://news.cnet.com/8301-13924_3-20024079-64.html ) The trick is that it may be a combo CPU+GPU to do it. So it may not be what you are thinking by OpenCL being solely GPU, but OpenCL code should be able to run.

    And in the end, what does it matter, really, as long as it runs? As the desktop Sandy Bridge review points out, video encoding is just as fast using solely the x86 codepaths as using nVidia's CUDA or ATI's Stream.
    Reply
  • Voldenuit - Monday, January 03, 2011 - link

    OpenCL was designed from the outset to run on heterogenous resources, including CPU.

    So intel claiming that they "support" OpenCL is nothing special - they just needed the right drivers/API.

    However, don't expect OpenCL code running solely on the CPU (my guess as to how SB will handle it) to be any faster than the x86 codepath running on the same CPU.

    Checkbox feature.
    Reply
  • jameskatt - Monday, January 03, 2011 - link

    What Intel wants to do is to have the CPU run OpenCL code.

    This totally defeats the purpose of OpenCL.

    OpenCL is suppose to allow both the GPU and the CPU to run code simultaneously. This is to allow significant acceleration in running OpenCL code compared to using just the CPU.

    Sure. OpenCL code will run. But it will run MORE SLOWLY than with a discrete GPU. And the 16 GPUs in Sandy Bridge will be wasted.

    Intel's Sandy Bridge has non-programmable GPUs. This is a serious limitation and deal killer when it comes to running OpenCL code.

    I expect Apple to continue use nVidia's or AMD's discrete GPUs with the MacBooks and Mac Book Pros.

    This is very disappointing. It shows that Intel still doesn't have the talent to produce decent GPUs.
    Reply
  • PlasmaBomb - Monday, January 03, 2011 - link

    And the 16 GPUs in Sandy Bridge will be wasted.


    *cough* I think you mean 12 EU *cough*
    Reply
  • Guspaz - Monday, January 03, 2011 - link

    <i>What Intel wants to do is to have the CPU run OpenCL code.

    This totally defeats the purpose of OpenCL.

    OpenCL is suppose to allow both the GPU and the CPU to run code simultaneously. This is to allow significant acceleration in running OpenCL code compared to using just the CPU.</i>

    No, this is the *primary* purpose of OpenCL. The goal of OpenCL is not to "allow the GPU and CPU to run code simultaneously", but to provide a single unified code path that can be used with any hardware, be it CPU or GPU. There are/were already code paths specific to each vendor/type (CUDA for nVIDIA GPUs, Stream for AMD/ATI GPUs, x86 for Intel/AMD CPUs). The problem is that fully supporting all three platforms requires three separate code paths.

    OpenCL unifies this, and allows a single codepath to be used regardless of the GPU's type or existence. You've completely misunderstood the purpose of OpenCL.
    Reply
  • Wiggy McShades - Tuesday, January 04, 2011 - link

    You need to ask what applications on a desktop actually use OpenCL in a meaningful way? Intel added hardware for media transcoding, which makes transcoding on something besides the cpu useless and that was roughly all openCL can be used for on the desktop, laptop, or cellphone.
    OpenCL is for vector calculations, AVX is for vector calculations. All four cores running AVX instructions would just be a faster choice than OpenCL on a low end gpu. Intel most likely could get sandybridge's gpu running OpenCL, but it would be pointless. OpenCL just is not a desktop feature.
    Reply
  • strikeback03 - Wednesday, January 05, 2011 - link

    Given how much money they have, I doubt Intel is lacking the "talent" to do anything they want. OpenCL execution on the GPU portion of the SNB chips was probably just not that big a deal to them, and given the number of other things (such as speed and battery life) SNB brings to the table they probably won't have trouble selling lots of these to the average consumer. Reply
  • 8steve8 - Monday, January 03, 2011 - link

    which mobile cpus on pg1 support TXT or VT-d or AES-NI or VT-x or Quick Sync? Reply

Log in

Don't have an account? Sign up now