Are the APUs more efficient than a well-chosen CPU + GPU combination?

The APUs are considered very efficient, because they combine a multi-core CPU and discrete-level GPU on a single power saver chip. These chips are produced by AMD and Intel for laptops, desktops and other devices;

Is that so?​

I have no doubt APUs suffice laptops very well, but how about an Ultrabook with low voltage Intel i5 processor and a super-efficient AMD GCN 2.0 discrete graphics? The Intel i5-3337U processor has only a 17W nominal TDP, the AMD Radeon 8750M GPU similarly low. The two together should not consume more power than the most modern AMD Richland A10-5750M APU (35W) itself. And how about the 17W Intel APU without discrete graphics? So let’s see which contender is the most efficient!


Test Systems:

HP Pavilion G6-2235us
- AMD Richland A10-5750M ES 35W APU,
- AMD Radeon HD 8650G,
- 2X4GB DDR3 1866MHz CL10 1.5V RAM,
- Samsung 830 Pro 128GB SSD,
- Windows 8 X64,
- AMD Catalyst 13.6 beta 2 Driver

Samsung Series 5 Ultra 530U4E-S02
- Intel Ivy Bridge i5-3337U 17W APU,
- Intel HD Graphics 4000,
- AMD Radeon HD8750M 2GB DDR3 discrete GPU,
- 2X4GB DDR3 1600MHz CL11 1.35V RAM,
- Seagate 500GB HDD +24GB ExpressCache SSD,
- Windows 8 X64,
- Intel HD 15.31.9.64.3165 Driver,
- AMD Catalyst 13.6 beta 2 Driver


As you can see I maxed out the systems much as possible and the good news is; this time the low voltage Core i5 3337U will benefit from 1600MHz DDR3 memory speed. The faster system memory improving APU’s graphics performance! Important to mention I have not overclocked any of these systems, so everything at factory clocks.


Tests;

In the following tests I’m going to compare performance and power consumption numbers to get an idea of energy efficiency. I’m using the same settings and resolutions for all configurations, divide the results by the power consumed and we have the performance per watt for each racer in the applications. In the games I selected a very demanding scenario to give plenty of work for both CPU and GPU and I measured performance only after 5 minutes gameplay. I did because APUs can trick the benchmarks by performing well in the first minute than underperorming (throttling) in the rest of the gameplay.


First let’s see how these competitors are performing;

In games the 35W AMD A10-5750M is the fastest APU, but the i5+8750M combination taking quite lenghty lead over both APU. If we count only CPU performance (Handbrake), than everybody looks equally fast.​

Now let’s see how those distant FPS numbers influence the power consumption;

As we expected the 17W Intel APU is the most energy saver, more interesting the CPU + GPU combination does not consumes more power than the 35W AMD APU. Wow, much faster and consumes less power! I think it is already indicating the winner of this review...​

Lastly the performance per watt (energy efficiency) calculations;

And the absolute winner of this test is the Intel Core i5 3337U + AMD Radeon 8750M aggregate. In games producing 83% more FPS / Watt than AMD APU and 60% more than Intel ULV APU. At CPU front (Handbrake) if the Radeon discrete GPU is inactive the Ultrabook is 77% more efficient than a laptop with standard voltage (35W) AMD APU.​

Final words:

The APUs are utterly destroyed in this review, but keep in mind not every CPU + GPU combination as efficient as a low voltage Intel Ivy Bridge i5-3337U processor and a Graphics Core Next 2.0 architecture AMD Radeon 8750M discrete graphics. An APU still can be more efficient than some other badly choosen CPU + GPU combination or I have not mentioned yet the new Intel Haswell APU family. That seems promising and even more encouraging the next year’s AMD Kaveri APU with GCN architecture integrated graphics.

So, what was wrong with the APUs in this review? The AMD APUs near weak CPU cores employing two generation older GPU architecture as currently available in the discrete GPU market. For example the A10-5750M’s Radeon 8650G is based on the very old Radeon HD 6000 Series (Cayman 4-way VLIW) architecture, which never was too efficient! The Intel Ivy Bridge APUs are at least have very efficient CPU part, but their integrated graphics sluggish as well.

After all these tests I’d like to examine some new Haswell APUs from Intel, especially the 28W with GT3 and the 47, 57W versions with GT3e graphics! Anandtech were impressed of these APUs, but they’ve forgot tell us the power consumption figures!

 

AMD CPUs do not match up to the core "i" architecture. That's what's wrong. You cannot "choose" a bad GPU CPU combination in laptops. Manufacturers usually have the best possible combination they can find either for the price or for the application. Bottomline is, any mid to high tier GPU with an Intel processor is going to be much better than the APUs. Also, I have no idea what an Intel APU is.

 

Ah... but this review is also comparing an old igp (and CPU architecture) to a modern discrete GPU and CPU (not to mention manuf. process)... which is not exactly 'fair' - though it does provide us with an insight into how powerful they are in relation to each other.

I agree that AMD's cpu performance is lacking in single-threaded tasks (especially in an age where most software is still dependent on that bit - even though a shift to multi-threaded started happening as well) - but Kaveri (which is Steamroller based), if it lives up to what AMD is saying, should see a very noticeable change on both CPU and iGP.

I'm thinking that for APU's at least, AMD probably should have sticked to improving Llano based architecture (was it K10 something?) after seeing problems with Bulldozer (at least until they addressed those issues with Steamroller like they are doing now).

However, this review also shows another thing:
That comparing APU to APU directly (without discrete GPU), the AMD solution is faster (and that's saying something for AMD) ... at least in graphics.

 

This laptop used to be priced at $1,400 at launch: Amazon.com: ASUS A53SD-TS71 Core i7 4GB/750GB NVIDIA GT 610M Laptop: Computers & Accessories

Intel Core i7-2670QM 2.2GHz Quad Core
4GB DDR3 10600, 750GB Hard Drive
NVIDIA GeForce GT 610M 2GB Graphics

FYI, Notebookcheck placed the GT 610M between an Intel HD 3000 and 4000.

 

Yes, like I said, USUALLY. There is probably a close relation between that config and the fact that there's only two reviews for a two year old machine.

After that I also stated what the point I was tryign to get across was -

That is not a mid tier GPU. You don't get as much of a choice as with a desktop. With a desktop, you can choose any CPU you want and pretty much any GPU. With laptops, it is usually manufacturers that make those choices. You have to choose what CPU you want and usually you only get three or four choices of GPUs. I hope this made my first post more clear.

 

Absolutely wrong statement manufacturers usually have the best possible combination, for example look at Dell's last generation laptops; standard voltage 35W i5 CPUs with Geforce GT610M, 620M, 630M and quadro craps in Latitudes. What is the sense of those? 50-70 TDP and barely faster gaming performance than a 35W Trinity...
You might also should know Intel's current processors (since Ivy Bridge) considered as APUs, even if officially they do not call it.

Well, I've tested what currently is available. Unfortunately cannot possible to find any Haswell laptop with GT3 or GT3e graphics yet and Kaveri is far away...

 

Then everything can be termed an APU because that definition is very broad. If you are going to give it this broad definition then you can call ALL Intel processors as APUs because since GMA, Intel had included OpenGL into their iGPUs. Where is your source for stating that since IVB, Intel CPUs have been considered APUs? If they don't officially call it that, do not include it because an "Intel APU" sounds completely out of place.

Secondly, the reason Latitudes contain lower eng GPUs is because not everyone wants to play games, unlike what you may think. You know how people drive to those tall buildings (or the short ones in Silicon Valley)? Yeah they work in there. A good number of them simply want to use Office Suite or managerial software. Not everyone needs to run CAD simulations (which APUs btw won't perform too well on either). So what's the difference? The i5. The i5 from last year was much better than the Trinity APUs that AMD managed to design. Here is a source. Latitudes are used a LOT at workplaces. Right now from where I am seated, I can see four Dell Latitudes and one of them looks like it is 4 years old. The user is happily typing away on his LaTex software. You don't need an APU for that. You need a good, fast CPU, something AMD's A10 cannot provide when compared to an IVB i5.

 

APU is AMD's marketing term, just like Ultrabook is Intel's marketing term (enforced with more lawyers).


Anyways, I came across this: AnandTech | AMD's A10-5750M Review, Part 1: The APU and Radeon HD 8650G Performance

 

APU is not brand of AMD, they use it and Intel also could use it! It is like both AMD and Intel used "CPU" or Nvidia, ATI, 3dfx "GPU" term. This site also call intel products as APU since Sandy Brirdge. Basically all processor what include a graphics processor in the same die is an APU;

​

Is there any sign they retested the Trinity APU with new drivers, or same 14 months old results?

 

My friend, you would be hard pressed to find drivers that increase performance by over 50% across all tests while keeping the hardware the same. Even if drivers were newer, you aren' going to see the A10 beating the i7 any time soon.

 

Pff... I do not know what are you still talking about here and why are you calling me friend?
you mean CPU performance? Where did I mentioned A10 beating i7 in any CPU term??? I talked about Anandtech has been using the same very old results for A10-4600M for over a year and not fair comparison with Richland A10 APU and others. Plus their A10-4600M's white book were equipped only with 4GB RAM.

 

Nevermind. You are correct. Everyone else is wrong. Intel sucks. AMD FT dub. Please purchase an A10 for playing video games.

 

Back in 2011, Tom's hardware did a comparison of drivers for a Radeon HD 4850 at launch, later drivers, and the most recent ones. They ran into some difficulties when some games refused to run on the older drivers, especially the launch version.

They concluded that an 4850 with the latest 2011 driver can match a 4870 with launch driver.

 

This test was all about how efficient a low voltage Intel ivy bridge together with a GCN 2.0 AMD Radeon, which probably the best combination for Ultrabooks. I just cannot see what is wrong with you or where and what are you reading?

 

And I agreed that the APU is the best combination and all. So go that route?

 

:hi2: next please... !

 

Very useful review, something that nowhere else possible to read. Bad news for APU users, though. The combination is amazing performer, hope Kaveri will be similarly good with GCN cores

 

May I ask where did you buy the laptop? I can only find 13 and 14" Ativbook 5 without the Radeon GPU. Any other ultrabook with similar specs as yours?

 

Atom you just made me more proud for my Samsung ultra, performance / watt numbers really amazing.
Jknows,
I think Atom also purchased in Germany, maybe your country doesn' have yet.

 

APU advantages are more in cost and pcb space mostly.

 

Which gets negated when the APU is paired with another GPU.

 

Yepp, this year AMD APU is no more than that.

Reading trough the whole test, they very right with the end conclusion;

Thanks for the good words and bringing my attention for the 13" version! My model is from the German Amazon and it is not yet available in the USA. Not easy to find similarly good configured ultrabooks, if you can hold I would wait for ultrabooks with 28W Haswell APU. The upcoming Dell Latitudes can be also interesting with Radeon 8790M, they finally drops their quadro craps...

 

At least AMD have everything to put a good GPU into Kaveri, but how about CPU front? Piledriver seems really sucks compared to i5.
I guess I'll wait for the entire Latitude lineup, need something thin as my current laptop.

 

Piledriver suffers from low single-threaded performance - which is increasingly becoming less important (but should be addressed by Steamroller/Kaveri).
In multi-threaded performance however, AMD can be comparable to higher end Intel parts (especially with OpenCL activated).
But AMD's problems are not just in that area.
There are big issues with manufacturers intentionally putting AMD's solutions into poorly designed products... such as mainly 15" and 17" on the mobile front.
Latest examples include putting a low power Kabini chip into a 17" notebook.
That is downright insane.
Kabini is Intel Atom 'competitor' and should be be in tablets, smartphones and hybrid solutions of smaller size.

AMD's APU is excellent for gaming and for tasks that support OpenCL.
Its also more than enough for majority of people out there who do basic tasks, but the problem is Intel which has permeated the market so much it doesn't allow AMD room to breathe at all.

I think we might be starting to see a turnaround in this what with AMD's 'penetration into consoles, and software going multithreaded on an increasing level.

 

Performance is the smaller problem, but it consumes almost twice as much power during load as the similarly fast or faster Core i5 ULV. Is the steamroller can bring 80% performance / watt improvement? I'd be happier if Kaveri coming with their new Jaguar cores around 3GHz...

 

This is an awesome review, never seen this type of watt / performance test before. Impressive the processor and GPU combo are so much faster than AMD's fastest APU in the same power envelope. It surprised me , you busted the APU and power efficiency myth:thumbsup:.

 

Yeah, the myth is busted; AMD's APU is far from efficient, at least near this year's Hardware. Thanks for the comment!