Xbitlabs: Comparison of current APUs

[frozentundra123456]

How come people never look past their own situation...

http://store.steampowered.com/hwsurvey/videocard/
Look at that survey. It's not representative, but I covered a lot of cards and systems which are still in use. I'm not saying that they should replace their rigs with a Trinity system. But I am saying that there's quite the large group of gamers out there who don't live in your performance expectation. Heck, I had a HD4670 up until recently, it games just fine.

Btw a HD7770 alone is 70~90W at load, just saying.

The top 5 discrete desktop cards are more powerful than Trinity.
I have no way to know for sure, but I expect many of the igps in the most popular levels are laptop cards. We also could be talking about older systems which the users have not been able to afford to replace. I am talking about desktops in the present time frame. I see a place for igps in laptops where you have a low res screen and cannot upgrade easily. But in the desktop, too much sacrifice in performance except for a very isolated case such as an HTPC.

As far as power use goes, I am willing to use more power, within reason, to achieve better performance. Even then a 65 watt pentium or i3 plus a 55 watt HD7750 is only 20 watts more than a 100w A10, and will probably give 50% better gaming performance.

 

[SlowSpyder]

Considering the poor performance of the Intel iGPU, you won't be playing many games anyway, if any at all, and that will save even more electricity! :awe:

This...! The Intel chips are using less power, but they're also doing less (at least in the gaming benches). Not much good saving 45 watts of electric if you're only pulling 13.4 FPS.

 

[SPBHM]

5700 looks good, almost the same performance with much lower power usage than the 5800k,

5700 + some $50 MB + 2x2GB DDR3 could be a great cheap PC for basic gaming,

 

[Dark Shroud]

Why would you use 2x2 instead of a single 1x4 that would allow an upgrade later on?

The AMD APUs have shown that faster memory is a benefit. So either 1866 or a stick of the "magic" Samsung memory that over clocks like a monster.

I also wouldn't call Intel's chips APUs since their IGPs can only do a fraction of what AMD's can.

 

[NTMBK]

Why would you use 2x2 instead of a single 1x4 that would allow an upgrade later on?

The AMD APUs have shown that faster memory is a benefit. So either 1866 or a stick of the "magic" Samsung memory that over clocks like a monster.

I also wouldn't call Intel's chips APUs since their IGPs can only do a fraction of what AMD's can.

Using just one stick means you only use one of the two memory channels available. This means that you can only have half of the potential memory bandwidth, crippling your graphics performance.

 

[grimpr]

My post was about the products evaluated in the xbitlabs article, you can't say the same about yours.

If the power consumption of the reviewed products are as irrelevant as you claim then why did xbitlabs bother to collect and present the data in their review?

For some reason xbitlabs felt it was relevant and so they included the data in their review article...why am I not afforded the same without being characterized as making a fuss and being sentimental?

If you have a problem with the message (the data graphs I embedded) then take it up with the people responsible for the message (AMD and xbitlabs).

IDC, my post wasnt an attack against you or your post but a righteous response to the message in the original article where they almost declare glorious victory of the Core i3 chips based on power consumption alone & without making a fuss about the non appearance of the hardware AES extension on the dualcore Intel chips compared to the AMD lineup, as for being "sentimental" about power consumption in a hardcore cpu overclocking forum/thread, i find this "trend" since Sandy Bridge came to market to be relevant and irellevant in the same time, relevant for making constructive and high quality analysis such is your posts, thoughts and experiments on the matter concerning architecture and the inner workings of node process and irrelevant at the same time for this forum being a cpu & overclocking enthusiast one with a 10+ years old history of cpu enthusiast overclocking, yes power consumption matters but imho that isnt something that a cpu enthusiast and overclocker will cry and nag about it all the time, i didnt see it when Nehalem came about, everyone was cheerfull for the IPC improvements Intel gave after Conroe neglecting that it was a blast furnace to cool down and a power hog for that matter and dont see it when 3930k enthusiasts talk about their chips, my general question still remains though, what priority power consumption in the same process node has over overclocking & performance in pc desktop cpus in an enthusiast hardware forum or news site? Sorry for mistypes and thanks for reading. :thumbsup:

 

[inf64]

What is funny is that soon we will have 6700 65W model that will be faster than 5800K . So the "super bad power draw that comes at a "cost" of >2x better performance :$" will be a dead argument soon . Let's see the "Haswell will do this and that to Richland" comments flowing now :awe:

 

[USER8000]

If people are going to be made bankrupt by running an A10 5700,I think they probably have worse problems,then fretting over 10W to 30W.

 

[Vesku]

I'm not really bothered at all by a ~30W difference at full load in a regular tower computer. Can see why there weren't/aren't many FM2 mini-itx boards though.

 

[USER8000]

I'm not really bothered at all by a ~30W difference at full load in a regular tower computer. Can see why there weren't/aren't many FM2 mini-itx boards though.

I think its more a case of the CPU volumes Intel sells that makes them a more viable situation for motherboard companies(Shuttle for example said that) to produce SFF motherboards.

You can run a Core i7 3770K or Core i7 2600K a number of socket 1155 mini ITX motherboards,and the A10 5700 does consumes less power.

 

[grimpr]

What is funny is that soon we will have 6700 65W model that will be faster than 5800K . So the "super bad power draw that comes at a "cost" of >2x better performance :$" will be a dead argument soon . Let's see the "Haswell will do this and that to Richland" comments flowing now :awe:

That Richland SKU is sweet, the compact & silent budget gaming pcs i could built with it.:awe:

 

[Galatian]

If people are going to be made bankrupt by running an A10 5700,I think they probably have worse problems,then fretting over 10W to 30W.

You know that's not the point: AMD APUs certainly do deliver more graphical performance, but they also draw more energy. Is that a good trade-off?

Let's look at the games tested by X-Bit. The A10-5700 only delivers 30 FPS on average at 1080p and low quality settings. I don't consider that good or playable at all. So I might as well just go with the Intel CPU because neither processor delivers good enough performance graphical wise IMHO.

But let's look how much the price difference is: actually the i3-3225 is only 10€ more expensive here in Germany then the A10-5700. We pay (on average) 24 Cents for electricity. Running the CPU maybe like 6 hours a day I should have the price difference already recouped in like a month.

I don't think APUs are bad per se, but I don't think that the current version delivers. Maybe Richland will be better but since it is the same architecture I actually doubt it. Kaveri will be a different story all together and I'm actually looking forward to that one.

 

[Vesku]

I think its more a case of the CPU volumes Intel sells that makes them a more viable situation for motherboard companies(Shuttle for example said that) to produce SFF motherboards.

You can run a Core i7 3770K or Core i7 2600K a number of socket 1155 mini ITX motherboards,and the A10 5700 does consumes less power.

Yes but they also most likely be running a dedicated GPU, as well.

 

[USER8000]

You know that's not the point: AMD APUs certainly do deliver more graphical performance, but they also draw more energy. Is that a good trade-off?

Let's look at the games tested by X-Bit. The A10-5700 only delivers 30 FPS on average at 1080p and low quality settings. I don't consider that good or playable at all. So I might as well just go with the Intel CPU because neither processor delivers good enough performance graphical wise IMHO.

But let's look how much the price difference is: actually the i3-3225 is only 10€ more expensive here in Germany then the A10-5700. We pay (on average) 24 Cents for electricity. Running the CPU maybe like 6 hours a day I should have the price difference already recouped in like a month.

I don't think APUs are bad per se, but I don't think that the current version delivers. Maybe Richland will be better but since it is the same architecture I actually doubt it. Kaveri will be a different story all together and I'm actually looking forward to that one.

Is that the price per KWH in Euro??

Yes but they also most likely be running a dedicated GPU, as well.

I am running a SFF PC with a SB Core i5 in an ITX motherboard. Something like an A10 5700 under full CPU or GPU load is unlikely to be an issue,in a well design motherboard with a reasonable 4+1 or 4+2 phase VRM. Look at the load figures:

http://img.hexus.net/v2/cpu/amd/Trinity/desktop/A105700/Power2G.png

http://img.hexus.net/v2/cpu/amd/Trinity/desktop/A105700/Power3G.png

Llano had a few mini-ITX motherboards too,and the 65W TDP quad cores were rare.

The main issue,as Shuttle themselves stated,is that the Intel market is bigger. You generally have a better range of mini-ITX motherboards for Intel,and even if AMD ended up having lower load power consumption,than the equivalent price comparable Intel CPUs,I doubt the landscape would change.

I have mucked around with a few of the AMD APUs,under idle and low load conditions,they seem not too bad with regards to power consumption at the wall.
The IGP is a bit better than what I expected too at least for older games.

This was interesting too:

http://forums.anandtech.com/showthread.php?t=2295636

 

[Galatian]

Is that the price per KWH??

Yes and I just checked at a pricing portal...the lowest price here in Berlin is actually 23,55 Cents per kWh. I'm already paying 100€ per month just on energy on a 2 person household. Of course that's for green energy (100% water) and we have an electrical water boiler for warm water, so obviously it is a little inflated in my case. But all the more reason to save as much as possible. That's why I look for the sweetspot with my gaming rig and for work I actually have a laptop.

Edit: Ok I just calculated it using the 1 Thread power consumption from XBit which shows a difference of 19 W, running both for 6 hours a day at such a load. I will actually save 1.9 Cents a day or about 7€ per year. Not much true...
The full load (just CPU) should be like 3,2 Cents a day or about 12€ a year. Also not that much.

Seeing those numbers I have to admit it really is not as bad as I thought. I would really like to see a real world test, because of race to idle and such things.

In fact those numbers makes Richland all the more interesting as it will have better power saving options, so I stand corrected!

 

[USER8000]

Yes and I just checked at a pricing portal...the lowest price here in Berlin is actually 23,55 Cents per kWh. I'm already paying 100€ per month just on energy on a 2 person household. Of course that's for green energy (100% water) and we have an electrical water boiler for warm water, so obviously it is a little inflated in my case. But all the more reason to save as much as possible. That's why I look for the sweetspot with my gaming rig and for work I actually have a laptop.

Using a laptop is a good idea. I tend to use my laptop or tablet for most general things(its just more convenient),and I suppose these devices as a whole are made with power saving in mind which is a bonus.

With desktops,you kind of have to look at all the bits to get a truely efficient one,if that is your aim. CPU,graphics card,motherboard,PSU and even the monitor. It surprising how much you can add to power consumption by ignoring a few parts!

Although,TBH,this has to be balanced with the cost of the parts too,so after a certain degree,its not worth it IMHO.

Anyway,going back to the A10 and Core i3, idle and low load conditions there is hardly any difference between the two platforms,if you check the reviews out.

If you look at 6 hours per day over 365 days,that is 2190 hours a year.

Over a year at 30W more that is 15.51 Euro more,and at 10W more that is 5.17 Euro more.

That is making the assumption than you are running a game like Batman or doing video encoding 24/7,not a synthetic benchmark(Furmark and/or Prime95) and are using the IGP. It will be interesting to see what improvements,that Richland and Kaveri bring to the table.

Personally,the AMD APUs are more suited for a general purpose family rig,which might be needed to have a few bases covered.

As a dedicated gaming rig from the get go,I would be more inclined to get a low end Core i5 or FX6300 as a minimum,and use it when games need to be run.

 

[Idontcare]

IDC, my post wasnt an attack against you or your post but a righteous response to the message in the original article where they almost declare glorious victory of the Core i3 chips based on power consumption alone & without making a fuss about the non appearance of the hardware AES extension on the dualcore Intel chips compared to the AMD lineup, as for being "sentimental" about power consumption in a hardcore cpu overclocking forum/thread, i find this "trend" since Sandy Bridge came to market to be relevant and irellevant in the same time, relevant for making constructive and high quality analysis such is your posts, thoughts and experiments on the matter concerning architecture and the inner workings of node process and irrelevant at the same time for this forum being a cpu & overclocking enthusiast one with a 10+ years old history of cpu enthusiast overclocking, yes power consumption matters but imho that isnt something that a cpu enthusiast and overclocker will cry and nag about it all the time, i didnt see it when Nehalem came about, everyone was cheerfull for the IPC improvements Intel gave after Conroe neglecting that it was a blast furnace to cool down and a power hog for that matter and dont see it when 3930k enthusiasts talk about their chips, my general question still remains though, what priority power consumption in the same process node has over overclocking & performance in pc desktop cpus in an enthusiast hardware forum or news site? Sorry for mistypes and thanks for reading. :thumbsup:

Sorry grimpr, I vented on you because I was still irritated by the power stuff. I'm irritated because I know AMD APU's can do better, could be so much better, if they weren't stuck with that 32nm process.

AMD's discrete GPUs are 28nm at TSMC and have great performance/watt. It is such a step backwards to see the graphics performance from 32nm process tech when the APU products are tested.

They need to get their APU's to 28nm asap.

 

[USER8000]

The power consumption test for the IGP is under Furmark. Moreover,the A10-5800K seems a bit pointless too. It will be really interesting to see the voltages both A10 CPUs operate at.

 

[Fox5]

Sorry grimpr, I vented on you because I was still irritated by the power stuff. I'm irritated because I know AMD APU's can do better, could be so much better, if they weren't stuck with that 32nm process.

AMD's discrete GPUs are 28nm at TSMC and have great performance/watt. It is such a step backwards to see the graphics performance from 32nm process tech when the APU products are tested.

They need to get their APU's to 28nm asap.

Err, the APUs seem to have pretty good performance per watt, at least when you account for the gpus being tied to the power hungry AMD cpus. The gpu part doesn't seem to be suffering on 32nm.

 

[grimpr]

The power consumption test for the IGP is under Furmark.

Wow, the prime95 of gpu stress testers, totally synthetic and out of all real world usage scenarios, even if you hit an APU with Crysis 3 it wont stand nowhere near those levels of power consumption. And that makes me wonder, does the i3 with its igp even run Crysis 3?

 

[Idontcare]

Err, the APUs seem to have pretty good performance per watt, at least when you account for the gpus being tied to the power hungry AMD cpus. The gpu part doesn't seem to be suffering on 32nm.

Jaguar is 28nm for a reason ^2

 

[USER8000]

Wow, the prime95 of gpu stress testers, totally synthetic and out of all real world usage scenarios, even if you hit an APU with Crysis 3 it wont stand nowhere near those levels of power consumption. And that makes me wonder, does the i3 with its igp even run Crysis 3?

Hexus tested the A10 5700 with Batman:Arkham City(I don't think it is the earlier one),and it consumed 11W more than a Core i3 3225,and was over twice as quick.

The A10 5800K,OTH,is a failure in comparison,consuming far more power,for hardly any extra peformance.

 

[grimpr]

Hexus tested the A10 5700 with Batman:Arkham City(I don't think it is the earlier one),and it consumed 11W more than a Core i3 3225,and was over twice as quick.

The A10 5800K,OTH,is a failure in comparison,consuming far more power,for hardly any extra peformance.

Yes, but since the K models appeal to the overclocking crowd and a good one at that, hwbot has a whole section with world records for AMDs APU overclocking, power consumption doesnt matter. As for us building compact budget gaming family pcs, we pick the 65W SKUs, they are so much better.

 

[grimpr]

Sorry grimpr, I vented on you because I was still irritated by the power stuff. I'm irritated because I know AMD APU's can do better, could be so much better, if they weren't stuck with that 32nm process.

AMD's discrete GPUs are 28nm at TSMC and have great performance/watt. It is such a step backwards to see the graphics performance from 32nm process tech when the APU products are tested.

They need to get their APU's to 28nm asap.

Thanks for responding, i understand your pov of thinking since you're more interested in financial & business threads these days about the future of AMD as a company and as a scientist process engineer you understand in depth the technological and financial dynamics on process tech that undermines AMD at the moment, things that i fully agree with and very interested to read about, but in the end we like to have fun with our chips, process node problematic or not, and AMDs APUs are surely fun to build budget family desktop pc's for light gaming.

 

[Arkaign]

Good for HTPC
Good for Laptop
Crap for Gaming

^^

That is true of ALL AMD and Intel processors when compared to even low end stuff like 7770.

And when you use discrete, there's zero reason to choose an AMD CPU.

Basically, for gamers : Intel CPU + AMD GPU = win.

I guess there's some place for the APUs, but really they're pretty horrible on average. I tried to play Hitman on an A6 laptop at 1366x768, and it was unplayable even on garbage settings. Truly hideous.