Try running the tests with Throttlestop set so the Intel CPU run at the highest frequency that's not in turbo mode. If it's the same as Sandy Bridge. Then the iGPU clocks will not be throttled if the CPU is not in turbo mode.
I don't think it would help a lot. Turbo mode is basically nonexistent in Diablo III and for Dirt 3 its only 10% higher clock. Considering average power of the iGPU is using 2-3x of the CPU, lowering the CPU clock by 10% wouldn't give a lot of headroom to the iGPU.
The graphs from Excel actually are a bit deceiving, which is why I explicitly listed average CPU clock. At least in the two games tested while logging with HWiNFO, the IVB ULV chip never has a Max CPU clock under the rated 1.8GHz of i5-3427U; it might look like it's at 1.7GHz, but the Excel file shows 1796.4MHz (give or take 1MHz).
Also, I should note that the graphs from HWiNFO include loading the games and interacting with the menu, so that flat line in DiRT 3 after all the jagged CPU clocks is the post-race report from a two stage rally. After the flat line I ran the built-in benchmark, which looks like it hits higher CPU clocks on average at the cost of iGPU clocks. In Diablo III, other than the first 20-30 seconds I'm running around the desert in Act 2; at the end I town portal and then exit the game.
That would defeat the entire focus of this article, Real-Time iGPU Clocks on ULV vs. Quad-Core as stated in the title. The point is to see how the GPU performs with said thermal restraints.... better yet, the purpose is to see the clock speeds of both CPU and GPU in said thermal restraints, under application loads.
So, how much faster is an Ivy ULV compared to Sandy ULV in terms of graphic performance? I would love to see a comparision between Ivy-Sandy-Trinity in the 17W TDP market, because those are basically the only choices you have if you go for a "thin and light" notebook.
We only have five results from our seven games for the Dell XPS 13, but it works out to around 30% faster on average. Of course, that's an i7 SNB and an i5 IVB, so potentially the gap would be wider if you compared i5 to i5, or i7 to i7. http://www.anandtech.com/bench/Product/608?vs=553
Finally someone tested the frequency on the ULV used in games. Have you tested what happens with the GPU clocks when you disable the CPU Turbo or SMT? Maybe this improves the iGPU clocks in games.
"After performing all of the above tests, we end up with an average iGPU clock on the i5-3427U around 1050MHz compared to 1250 on the larger notebook. That represents a nearly 20% performance advantage on the GPU clocks, but in the actual games we’re still seeing a much greater advantage for the quad-core chip. "
It's the faster CPU that can make an additional difference in some games like Skyrim or Dirt but also the bigger L3 cache that can be used by the iGPU through the ringbus. 3720QM has 6MB while i5-3427U 3MB.
Configurable TDP is available on all IVB chips I believe, but it's up to the laptop OEMs to enable it. Some might only do TDP Down (I think that's what ASUS has done on the UX21A at least), others will do up and down. Most likely, TDP Up will only be available on larger laptops and/or laptops with docking stations/cooling stations. It's pretty obvious (based on the CPU/GPU Turbo modes) that the prototype Ultrabook is already out of room as far as cooling goes, and it would be difficult given the position of the intake/exhaust ports to increase the cooling without redesigning the chassis. Other Ultrabooks and laptops could change the design and thus improve cooling and performance.
The way that the Quad-Core iGPU traces are pegged and the ULV is not makes me wonder if there are some driver issues keeping the ULV chip from running at a fixed frequency for a long time. By the looks of it, that frequency would still be lower than the quad core iGPU.
The ULV chip is supposed to max out lower than the quad-core (1150 vs. 1250). The difference is in the TDP limit, as well as temperatures. The Ultrabook is hitting a pretty constant 80C during gaming, compared to 68-70C on the ASUS N56VM (using only the IGP). Everything is working as designed, and if I were to put the i7-3720QM under a torture test (e.g. block off the exhaust so it heats up more, or run a heavily threaded CPU test concurrent with gaming), I expect we'd see similar variance in GPU clocks. In fact, I'm going to test that right now....
And the result: nope, I can't get the N56VM to throttle down on the iGPU. It stays at 1250MHz, but with a heavy CPU load running (6-threaded x264 encode, leaving one core/two threads for DiRT 3) the CPU clocks drop to 2700-3000MHz instead of 3400-3600MHz. Package temperatures are hitting up to 91C though.
If you were to take the ULV laptop and run your benchmark while holding it over an air conditioning vent (or something similar) would you approach the 35W results, or are we purely power limited here (since we're hitting about 17W already)?
I did do a test at night next to an open window (probably mid-50s or lower outside) and the results didn't change that I could tell. Temperatures actually don't look too bad on the CPU/package (80C or so), so I think it's mostly the TDP limit. If the laptop supported configurable TDP with the option to set a 20W TDP, I bet performance would improve at least 10% in several games.
Awesome article. Looking at the HWiNFO64 results, it appears that any downward spikes in GPU clock are accompanied by upward spikes in CPU clock. Wouldn't this seem to be consistent with the thought that the limit is TDP and not thermal?
I'm also curious if you have any insight on how the ULV would perform with non-graphics applications (such as say crunching numbers in MATLAB). Presumably the majority of the wattage would then be available to the CPU and we'd see more sustained turbo CPU speeds? Any thoughts/speculation?
- if you're looking for steady state performance (sustained longer than for a few seconds) and not for burst-mode performance (less than a few seconds), feel free to skip the higher clocked 17 W CPUs - they won't reach their higher turbo modes anyway (in games) - 25 W TDP seems the sweet spot for dual core IVB. Look for laptops slightly larger than Ultrabooks if you're looking for performance - lowering CPU & iGPU voltage might increase performance (question: how can it be done? software?)
You won't hit the higher Turbo Boost modes on an i7, but you should hit the guaranteed base clock (e.g. 2.0GHz on the i7-3667U vs. 1.8GHz on the i5-3427U). In some cases, that would make the i7 faster, but if you're comparing 11.6" i7 to 13.3" i5, e.g. the ASUS UX21A, all bets are off.
As for the last item, AFAIK there are very few utilities that even try to work with laptops for over/under volting/clocking. The OEMs that make the laptops might have the knowledge to release such a utility, but they don't want to do so as it will more likely than not cause problems.
For mobile Llano, "k10stat" is commonly used for over/under volting/clocking of the CPU. (I think there is another alternate utility but I forgot the name.) It is surprising that there is no equivalent in Intel land (I guess stock mobile Intel CPUs still beat overclocked mobile Llano).
Not sure if "k10stat" works with Trinity. I can't check it but someone can :-)
For Brazos there's Brazos Tweaker. However, there's not much juice left in this CPU. For Intel we had RMClock back in the C2D days. I used it to lower the real world full CPU load power consumption of 2 of my CPUs by 10 W each, from ~35 W to 25 W (for the entire machine). That was pretty significant, especially since it drove the fan down from speed 3 to speed 1 (lowest).
I guess Intel is setting the voltage a bit more tight on current products, since they're now completely power consumption constrained.. but I'm sure there'll be some reserve left.
Oh, I think Throttle Stop /or so) was trying to implement customizable voltages on core i.. not sure how successful, though.
ThrottleStop does not work for setting voltages, at least not on the laptops where I've installed it -- that goes for Sandy Bridge as well as Ivy Bridge; I never tried with Arrandale. The problem is I think many laptops use either non-programmable voltage controllers, or they're programmable but completely undocumented so unless you know exactly how to talk to each and every laptop it won't work.
Excellent article. I'm wondering what the results would be with an i7 which has the same 650-1250 MHz frequency range for graphics as the 3720QM, however its TDP is 35W and it is not a quad core processor. I'm talking about the i7 3520M. I agree that most games are perfectly happy with two cores, so how the 3520M would score against the tested ULV and quadcore chip?
Given the results with the i5-2410M and i5-2520M vs. i7-2820QM, performance should be very similar. There are a few instances where the larger L3 cache will help on the quad-core to the tune of 5-10%, but with 35W TDP and only two CPU cores we rarely saw issues on Sandy Bridge and I expect Ivy Bridge will be the same. We'll know for certain when we get one of the 35W dual-core chips in a review laptop. :-)
Considering that on Linux that ENABLING the L3 cache use for Sandy Bridge iGPU resulted in only 15-20% improvement, I doubt you'd see that much of a difference due to L3 cache differences. 5-10% would require significant difference in hit rates, like going from a 100% hit rate down to a 50%.
Plus, Ivy Bridge's iGPU has its own L3 cache, further reducing the impact.
Does anyone else recall Intel speaking about the potential of mobile IVB, specifically, how, when combined with a cooling dock, the package would be able to stay at Turbo for longer periods? Was that Intel making hypothetical statements, or have some oems expressed interest in doing this?
Silly question: what does the current MacBook Air have, a 17W or a 35W? I'm just curious, I'm due to replace my five year old Mac with a new one. If they're currently 17W, then I'll probably hold off until next year for Haswell.
MagSafe has to be able to do more than 45W, as the high-end MacBook Pro 15/17 still use MagSafe and I'm sure they can pull more than 45W under load (35W for the CPU and at least 30W for the discrete GPU, if not 45W). Other than that, yes, MacBook Air has always been a 17W CPU to my knowledge.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
35 Comments
Back to Article
Mayuyu - Friday, June 1, 2012 - link
Try running the tests with Throttlestop set so the Intel CPU run at the highest frequency that's not in turbo mode. If it's the same as Sandy Bridge. Then the iGPU clocks will not be throttled if the CPU is not in turbo mode.IntelUser2000 - Friday, June 1, 2012 - link
I don't think it would help a lot. Turbo mode is basically nonexistent in Diablo III and for Dirt 3 its only 10% higher clock. Considering average power of the iGPU is using 2-3x of the CPU, lowering the CPU clock by 10% wouldn't give a lot of headroom to the iGPU.JarredWalton - Friday, June 1, 2012 - link
The graphs from Excel actually are a bit deceiving, which is why I explicitly listed average CPU clock. At least in the two games tested while logging with HWiNFO, the IVB ULV chip never has a Max CPU clock under the rated 1.8GHz of i5-3427U; it might look like it's at 1.7GHz, but the Excel file shows 1796.4MHz (give or take 1MHz).Also, I should note that the graphs from HWiNFO include loading the games and interacting with the menu, so that flat line in DiRT 3 after all the jagged CPU clocks is the post-race report from a two stage rally. After the flat line I ran the built-in benchmark, which looks like it hits higher CPU clocks on average at the cost of iGPU clocks. In Diablo III, other than the first 20-30 seconds I'm running around the desert in Act 2; at the end I town portal and then exit the game.
hsew - Friday, June 1, 2012 - link
That would defeat the entire focus of this article, Real-Time iGPU Clocks on ULV vs. Quad-Core as stated in the title. The point is to see how the GPU performs with said thermal restraints.... better yet, the purpose is to see the clock speeds of both CPU and GPU in said thermal restraints, under application loads.MatthiasP - Friday, June 1, 2012 - link
So, how much faster is an Ivy ULV compared to Sandy ULV in terms of graphic performance? I would love to see a comparision between Ivy-Sandy-Trinity in the 17W TDP market, because those are basically the only choices you have if you go for a "thin and light" notebook.JarredWalton - Friday, June 1, 2012 - link
We only have five results from our seven games for the Dell XPS 13, but it works out to around 30% faster on average. Of course, that's an i7 SNB and an i5 IVB, so potentially the gap would be wider if you compared i5 to i5, or i7 to i7.http://www.anandtech.com/bench/Product/608?vs=553
mikk - Friday, June 1, 2012 - link
Finally someone tested the frequency on the ULV used in games. Have you tested what happens with the GPU clocks when you disable the CPU Turbo or SMT? Maybe this improves the iGPU clocks in games.mikk - Friday, June 1, 2012 - link
"After performing all of the above tests, we end up with an average iGPU clock on the i5-3427U around 1050MHz compared to 1250 on the larger notebook. That represents a nearly 20% performance advantage on the GPU clocks, but in the actual games we’re still seeing a much greater advantage for the quad-core chip. "It's the faster CPU that can make an additional difference in some games like Skyrim or Dirt but also the bigger L3 cache that can be used by the iGPU through the ringbus. 3720QM has 6MB while i5-3427U 3MB.
Ytterbium - Friday, June 1, 2012 - link
Are any of these chips cTDP enable?if you could up the up the 17W to 25W then you could start to see soem quad like results.
spcarr - Friday, June 1, 2012 - link
Great article. Is there a way to allow a ULV to exceed the 17w TDP? Adding some extra form of cooling and maybe software to let it go further?JarredWalton - Friday, June 1, 2012 - link
Configurable TDP is available on all IVB chips I believe, but it's up to the laptop OEMs to enable it. Some might only do TDP Down (I think that's what ASUS has done on the UX21A at least), others will do up and down. Most likely, TDP Up will only be available on larger laptops and/or laptops with docking stations/cooling stations. It's pretty obvious (based on the CPU/GPU Turbo modes) that the prototype Ultrabook is already out of room as far as cooling goes, and it would be difficult given the position of the intake/exhaust ports to increase the cooling without redesigning the chassis. Other Ultrabooks and laptops could change the design and thus improve cooling and performance.stanwood - Friday, June 1, 2012 - link
The way that the Quad-Core iGPU traces are pegged and the ULV is not makes me wonder if there are some driver issues keeping the ULV chip from running at a fixed frequency for a long time. By the looks of it, that frequency would still be lower than the quad core iGPU.JarredWalton - Friday, June 1, 2012 - link
The ULV chip is supposed to max out lower than the quad-core (1150 vs. 1250). The difference is in the TDP limit, as well as temperatures. The Ultrabook is hitting a pretty constant 80C during gaming, compared to 68-70C on the ASUS N56VM (using only the IGP). Everything is working as designed, and if I were to put the i7-3720QM under a torture test (e.g. block off the exhaust so it heats up more, or run a heavily threaded CPU test concurrent with gaming), I expect we'd see similar variance in GPU clocks. In fact, I'm going to test that right now....JarredWalton - Friday, June 1, 2012 - link
And the result: nope, I can't get the N56VM to throttle down on the iGPU. It stays at 1250MHz, but with a heavy CPU load running (6-threaded x264 encode, leaving one core/two threads for DiRT 3) the CPU clocks drop to 2700-3000MHz instead of 3400-3600MHz. Package temperatures are hitting up to 91C though.seapeople - Sunday, June 3, 2012 - link
If you were to take the ULV laptop and run your benchmark while holding it over an air conditioning vent (or something similar) would you approach the 35W results, or are we purely power limited here (since we're hitting about 17W already)?JarredWalton - Sunday, June 3, 2012 - link
I did do a test at night next to an open window (probably mid-50s or lower outside) and the results didn't change that I could tell. Temperatures actually don't look too bad on the CPU/package (80C or so), so I think it's mostly the TDP limit. If the laptop supported configurable TDP with the option to set a 20W TDP, I bet performance would improve at least 10% in several games.cwcwfpfp - Monday, June 4, 2012 - link
Awesome article. Looking at the HWiNFO64 results, it appears that any downward spikes in GPU clock are accompanied by upward spikes in CPU clock. Wouldn't this seem to be consistent with the thought that the limit is TDP and not thermal?I'm also curious if you have any insight on how the ULV would perform with non-graphics applications (such as say crunching numbers in MATLAB). Presumably the majority of the wattage would then be available to the CPU and we'd see more sustained turbo CPU speeds? Any thoughts/speculation?
MrSpadge - Friday, June 1, 2012 - link
- if you're looking for steady state performance (sustained longer than for a few seconds) and not for burst-mode performance (less than a few seconds), feel free to skip the higher clocked 17 W CPUs - they won't reach their higher turbo modes anyway (in games)- 25 W TDP seems the sweet spot for dual core IVB. Look for laptops slightly larger than Ultrabooks if you're looking for performance
- lowering CPU & iGPU voltage might increase performance (question: how can it be done? software?)
JarredWalton - Friday, June 1, 2012 - link
You won't hit the higher Turbo Boost modes on an i7, but you should hit the guaranteed base clock (e.g. 2.0GHz on the i7-3667U vs. 1.8GHz on the i5-3427U). In some cases, that would make the i7 faster, but if you're comparing 11.6" i7 to 13.3" i5, e.g. the ASUS UX21A, all bets are off.As for the last item, AFAIK there are very few utilities that even try to work with laptops for over/under volting/clocking. The OEMs that make the laptops might have the knowledge to release such a utility, but they don't want to do so as it will more likely than not cause problems.
ssiu - Friday, June 1, 2012 - link
For mobile Llano, "k10stat" is commonly used for over/under volting/clocking of the CPU. (I think there is another alternate utility but I forgot the name.) It is surprising that there is no equivalent in Intel land (I guess stock mobile Intel CPUs still beat overclocked mobile Llano).Not sure if "k10stat" works with Trinity. I can't check it but someone can :-)
MrSpadge - Saturday, June 2, 2012 - link
For Brazos there's Brazos Tweaker. However, there's not much juice left in this CPU. For Intel we had RMClock back in the C2D days. I used it to lower the real world full CPU load power consumption of 2 of my CPUs by 10 W each, from ~35 W to 25 W (for the entire machine). That was pretty significant, especially since it drove the fan down from speed 3 to speed 1 (lowest).I guess Intel is setting the voltage a bit more tight on current products, since they're now completely power consumption constrained.. but I'm sure there'll be some reserve left.
Oh, I think Throttle Stop /or so) was trying to implement customizable voltages on core i.. not sure how successful, though.
JarredWalton - Saturday, June 2, 2012 - link
ThrottleStop does not work for setting voltages, at least not on the laptops where I've installed it -- that goes for Sandy Bridge as well as Ivy Bridge; I never tried with Arrandale. The problem is I think many laptops use either non-programmable voltage controllers, or they're programmable but completely undocumented so unless you know exactly how to talk to each and every laptop it won't work.manguszta - Friday, June 1, 2012 - link
Excellent article.I'm wondering what the results would be with an i7 which has the same 650-1250 MHz frequency range for graphics as the 3720QM, however its TDP is 35W and it is not a quad core processor. I'm talking about the i7 3520M.
I agree that most games are perfectly happy with two cores, so how the 3520M would score against the tested ULV and quadcore chip?
JarredWalton - Friday, June 1, 2012 - link
Given the results with the i5-2410M and i5-2520M vs. i7-2820QM, performance should be very similar. There are a few instances where the larger L3 cache will help on the quad-core to the tune of 5-10%, but with 35W TDP and only two CPU cores we rarely saw issues on Sandy Bridge and I expect Ivy Bridge will be the same. We'll know for certain when we get one of the 35W dual-core chips in a review laptop. :-)IntelUser2000 - Friday, June 1, 2012 - link
Considering that on Linux that ENABLING the L3 cache use for Sandy Bridge iGPU resulted in only 15-20% improvement, I doubt you'd see that much of a difference due to L3 cache differences. 5-10% would require significant difference in hit rates, like going from a 100% hit rate down to a 50%.Plus, Ivy Bridge's iGPU has its own L3 cache, further reducing the impact.
Jamahl - Friday, June 1, 2012 - link
Almost like an old AT article back when it was worth reading. Well done...and more of this in future please.tuxRoller - Friday, June 1, 2012 - link
Does anyone else recall Intel speaking about the potential of mobile IVB, specifically, how, when combined with a cooling dock, the package would be able to stay at Turbo for longer periods?Was that Intel making hypothetical statements, or have some oems expressed interest in doing this?
MrSpadge - Saturday, June 2, 2012 - link
Configurable TDP is there, but it's up to the OEMs to enable this feature.secretmanofagent - Friday, June 1, 2012 - link
Silly question: what does the current MacBook Air have, a 17W or a 35W? I'm just curious, I'm due to replace my five year old Mac with a new one. If they're currently 17W, then I'll probably hold off until next year for Haswell.noeldillabough - Saturday, June 2, 2012 - link
Pretty sure 17w, my air is much slower than my x220 (35w)secretmanofagent - Saturday, June 2, 2012 - link
Thanks for the info. I found a site which says it's this processor:http://ark.intel.com/products/54617
Since the MagSafe is only 45W if memory serves, it would point to it only being the 17W model as the likely possibility.
JarredWalton - Saturday, June 2, 2012 - link
MagSafe has to be able to do more than 45W, as the high-end MacBook Pro 15/17 still use MagSafe and I'm sure they can pull more than 45W under load (35W for the CPU and at least 30W for the discrete GPU, if not 45W). Other than that, yes, MacBook Air has always been a 17W CPU to my knowledge.extide - Monday, June 4, 2012 - link
I seem to remember mag safe being 60 or 80w. It's been a while, but thats what I remember from having a 17" mbp.MrSpadge - Saturday, June 2, 2012 - link
You can't reasonably cool 35 W in such a form factor.. so yes, pretty sure 17 W.hosaka_heavy_industries - Friday, January 4, 2013 - link
You're really lazy that you wouldn't do the tests, and you should be ashamed.