AMD's Trinity APU Efficiency: Undervolted And Overclocked

[Cazalan]

Here's hoping this gets Intel to release faster clocked i3's or unlocked i3's.

 

[Guest]

There is nothing spun about lower idle power consumption. Most computers spend the majority of their life in an idle or near idle state. The only one who sounds biased is you. It was already mentioned that they are willing to do more benchmarks with idle power consumption factored in.

using benchmarks and data not related to the main goal of testing power consumption isn't bias. yet backing an article with that goal with more proof is?

yeah right.

sheep follow the herd off the cliff with the rest . . .

 

[army_ant7]

So.... this CPU is years later than the core-i3, and can't compete in terms of power efficiency, and it takes a 3.8-4.4ghz AMD CPU to meat or beat a "3.3Ghz" Core -i3???

Ummm.... Yeah.... I really hope AMD finds a new engineering team sometime soon.

The problem here seems to be that you may not be aware that there are actually different generations of the Core-i series.
1st gen = Nehalem/Westemere (not sure exactly which is which among the CPU's)
2nd gen = Sandy Bridge
3rd gen = Ivy Bridge

Ivy Bridge came out this year only, and the i3 Ivy Bridges (roughly) only a month ago more or less.

oneblackened:

"Because all desktop i5s are quad-core. You can't replace a full core with a second thread on a core. You get FAR less performance out of a double-threaded core than 2 single-threaded cores, all other things being equal. "

What are you talking about? Only 1/2 of the Core-i5's are quad core, the other half are dual core with HT enabled. Whoever told you all Core-5's were quad's was very mistaken.

You were right about there being dual-core i5's (mobile ones), but the problem is in bold in the quote. Just try double-reading first or not trying to be so rash or harsh with your comments.

If an unlocked i3 can hit around 5GHz, then it could easily match a locked stock i5 in highly threaded performance while beating it greatly in single and dual-threaded work, the same advantage that it has over stock AMD CPUs.

I'd think that your statement would be correct about it beating the i5's if the programs used can't evenly use 4 threads. I'm just thinking that theoretically, a quad-core 3.1GHz i5 (for example) has 12.4GHz worth of performance to be exploited (and I'm not sure if it has more cache or anything, too lazy to check), while a dual-core 5GHz Core i3 has only 10GHz to be used, though it has Hyper-threading to squeeze out more performance. I know that there may be other factors that would come into play like overhead and stuff from multi-threading (though I'm not sure, but the i3 may still experience those still) and possibly other stuff. But considering all that, I think it's still pretty tough to conclude that a 5GHz i3 would beat a lower-end desktop i5 at programs using multi-threading at its best for 4 or more threads, unless I'm missing. I bet it would come close though. Single-threaded performance is a no brainer in this case so you won't hear arguments from me there. Looking forward to your reply.

 

[army_ant7]

Also, to anyone who pretty much said that the i3's can't overclock (and also possibly mocked others because they said they would've liked to have seen it in this article), I'd have to tell you to look up BCLK overclocking. Though there might be little to gain, it still is overclocking. (I've heard of a 5MHz increase (105MHz) on Sandy Bridges, which would mean a 5% increase, which is still something.) It would be somewhat unfair because Chris didn't change the BCLK's of the Trinities to squeeze the best possible overclock in this test which is understandable possibly due to time constraints, plus I hear the limit highly depends on your mo-bo since other components on it would run faster and would eventually become unstable. (I do wonder how much the BCLK of the Trinities can go, along with Piledrivers.)

I'm not certain about this, but aren't all Intel HD graphics unlocked? I remember reading something like this. Would anyone care to testify?

 

[techcurious]

[citation][nom]army_ant7[/nom]Also, to anyone who pretty much said that the i3's can't overclock (and also possibly mocked others because they said they would've liked to have seen it in this article), I'd have to tell you to look up BCLK overclocking. Though there might be little to gain, it still is overclocking. (I've heard of a 5MHz increase (105MHz) on Sandy Bridges, which would mean a 5% increase, which is still something.) It would be somewhat unfair because Chris didn't change the BCLK's of the Trinities to squeeze the best possible overclock in this test which is understandable possibly due to time constraints, plus I hear the limit highly depends on your mo-bo since other components on it would run faster and would eventually become unstable. (I do wonder how much the BCLK of the Trinities can go, along with Piledrivers.)I'm not certain about this, but aren't all Intel HD graphics unlocked? I remember reading something like this. Would anyone care to testify?[/citation]
Oh no, we are aware of the BCLK overclocking. It's just that a 5% CPU overclock is not worth mentioning.. and not worth the hours of stress testing to determine stability. As far as CPU overclocking goes, 5% (if you are lucky to even get that much out of BCLK overclocking alone) is pathetic and hardly worth the effort.

 

[army_ant7]

Oh no, we are aware of the BCLK overclocking. It's just that a 5% CPU overclock is not worth mentioning.. and not worth the hours of stress testing to determine stability. As far as CPU overclocking goes, 5% (if you are lucky to even get that much out of BCLK overclocking alone) is pathetic and hardly worth the effort.

I wouldn't disagree with you, that it may not be worth it, but hey, people are different. Some people may be willing to go through the trouble, possibly because it's fun for them to do it along the way. One view on it could be that for example, upping the BCLK by 3MHz on a 3.3GHz i3-3220 would make it pretty much 3.4GHz which is the same as the substantially more expensive i3-3240 (which, yes, I know, is unreasonably priced).

It may seem silly, but we just have our own idiosyncrasies and they weren't spreading any false info. We should respect their opinions but it isn't really bad to give our own. See what I'm saying? Also, though you admit that (at least) you're aware of the BCLK overclocking, others shouldn't act like it's nonexistent. It could mislead other people.

 

[proffet]

I do not agree with raising the BLK in overclocking, not good and causes too much stress.
but I have messed with it once or twice when I put together a few i3-2120 units.
sometimes the 5% increase is too much and will become unstable real quick.
Intel locked those non-K's up pretty good but if you have a P or Z series board no harm in playing around with it.
maybe you can get it stable.

edit: luciferano corrected...

 

[kamber87]

What should be taken into consideration is that with Intel's HD4000, you're not going to do any gaming at all. So anyone who opts for core i3 3225 should buy a discrete graphics card for gaming. Then (with a discrete graphics card) you can compare power consumption of the WHOLE intel platform and its price to that of AMD's. Only then the comparison could be fair I think.

 

[luciferano]

[citation][nom]proffet[/nom]I do not agree with raising the BLK in overclocking, not good and causes too much stress.but I have messed with it once or twice when I put together a few i3-2120 units.sometimes the 5% increase is too much and will become unstable real quick.Intel locked those non-K's up pretty good but if you have a P or K series board no harm in playing around with it.maybe you can get it stable.[/citation]

K series board? Do you mean Z?

Also, I've heard of a few people getting around 10% BLCK overclocks on LGA 1155 systems (and also some people not being able to increase or decrease it at all, but I don't hear that very often either). Have any of you heard of that too? I think it depends on the CPU as well as the motherboard, but I have to agree with people saying that it's kinda not worth it for the most part. However, worth it or not, it could be tested and maybe there are tricks or we can at least learn the variables in different LGA 1155 systems that allow the BLCK overclocking headroom to vary so much.

 

[luciferano]

[citation][nom]kamber87[/nom]What should be taken into consideration is that with Intel's HD4000, you're not going to do any gaming at all. So anyone who opts for core i3 3225 should buy a discrete graphics card for gaming. Then (with a discrete graphics card) you can compare power consumption of the WHOLE intel platform and its price to that of AMD's. Only then the comparison could be fair I think.[/citation]

I wouldn't game on an Intel IGP specifically because they still have driver issues with a lot of games, but sure, the low performance is also crippling. However, as army_ant7 has reminded us, the Intel IGPs can be overclocked. If they have a lot of safe headroom on the Ivy Bridge IGPs, then maybe we could get some decent performance out of them. I do agree though, there should have been changes made to the Intel system for it to be a gaming system versus gaming system comparison instead of a low-end productivity system versus gaming system comparison.

The two systems here wouldn't be used in the same way, so comparing them directly is unfair to both in differing ways. The AMD system gets higher power consumption compared to the i3 system than it would if the i3 system was given a cheap graphics card, albeit the i3 system would be too expensive. Tom's could have taken this opportunity to do another Pentium/Celeron with discreet versus AMD APU (both systems with the same budget, at least for the board/memory/CPU/graphics) article with the newest generation of each. This test doesn't really prove much except that the i3 system uses less power (something that we shouldn't even need to test because that's just obvious) when it's not built for the same purpose and that the i3 system also can't come close in gaming performance as a result of that.

 

[uglynerdman]

uh since i actually buy parts rather than play fantasy cpuball like most on here. id go for the i3 with a discreet card. if im gonna game might as well get a good gpu. all i care is a i get stable fps right? and all other productivity software will run on a i3 faster.

The big thing is the power savings for someone who actually pays their electric bills and worries about money managing a home.

 

[drinvis]

Chris Angelini,
Can you please clarify some of my doubts regarding photoshop cs6.
Does all the functions where OpenCL is implemented in photoshop cs6 do well on both amd a10 5800k
and intel i3 3220/25.Or does the IGP of trinity has some advantage?
Regardless pf the above,I think all the functions where OpenGL is used/implemented I think the
a10 5800k would be better in performance.Please correct me if I am wrong.

 

[luciferano]

[citation][nom]uglynerdman[/nom]uh since i actually buy parts rather than play fantasy cpuball like most on here. id go for the i3 with a discreet card. if im gonna game might as well get a good gpu. all i care is a i get stable fps right? and all other productivity software will run on a i3 faster. The big thing is the power savings for someone who actually pays their electric bills and worries about money managing a home.[/citation]

With a comparably performing discrete card counted and a power supply unit that doesn't have between double and triple the proper wattage, power consumption would be nearly identical between the i3 system and the A10 system. With any card less than a 6670, the AMD system has Dual-graphics and would thus continue to outperform the Intel system greatly in most games and with a greater card, you could disable the AMD IGP and power consumption on that APU could drop like a rock, especially if you undervolt.

 

[luciferano]

[citation][nom]drinvis[/nom]Chris Angelini,Can you please clarify some of my doubts regarding photoshop cs6.Does all the functions where OpenCL is implemented in photoshop cs6 do well on both amd a10 5800kand intel i3 3220/25.Or does the IGP of trinity has some advantage?Regardless pf the above,I think all the functions where OpenGL is used/implemented I think thea10 5800k would be better in performance.Please correct me if I am wrong.[/citation]

Intel's superior memory bandwidth is probably a huge factor in how well they competed in non-gaming GPU performance with AMD. Remember, Intel's memory controllers get a lot more bandwidth out of the same memory configuration than AMD.

 

[proffet]

K series board? Do you mean Z?

Also, I've heard of a few people getting around 10% BLCK overclocks on LGA 1155 systems (and also some people not being able to increase or decrease it at all, but I don't hear that very often either). Have any of you heard of that too? I think it depends on the CPU as well as the motherboard, but I have to agree with people saying that it's kinda not worth it for the most part. However, worth it or not, it could be tested and maybe there are tricks or we can at least learn the variables in different LGA 1155 systems that allow the BLCK overclocking headroom to vary so much.

P series included, P67 board in-particular but for the most part your right too.
it's about the Z series now.

I was referring to 'K' series chips and not motherboards.

 

[luciferano]

[citation][nom]proffet[/nom]P series included, P67 board in-particular but for the most part your right too.it's about the Z series now.I was referring to 'K' series chips and not motherboards.[/citation]

Intel locked those non-K's up pretty good but if you have a P or K series board no harm in playing around with it.maybe you can get it stable.

I'm absolutely sure that you had a typo here.

 

[proffet]

nice find...
corrected.

 

[army_ant7]

You have to wonder, why do the a10-5800K and 5700 have the same price?

 

[luciferano]

[citation][nom]army_ant7[/nom]You have to wonder, why do the a10-5800K and 5700 have the same price?[/citation]

Look at the TDPs. The 5700 has a 65W TDP and the 5800K has a 100W TDP. That's probably the reason for the price similarity.

 

[tajisi]

[citation][nom]esrever[/nom]Most PC are idle or semi idle when people have them on. 90% of the time I use my PC, I do web surfing or watch video or a text editor for work, my pc is not loaded with benchmarks 24/7. If you look at idle power consumption, the trinity APUs are amazing. They easily beat out intels offerings. If you are looking at the power consumption over a month, the trinity will be much more energy efficient than the i3 for most people.[/citation]

The problem with that being that if the end user spends 90% of their time idle and 10% doing anything taxing, the larger power draw of the APU quickly eats away any savings that would have accrued. I can get 33 MPG on the highway if I drive carefully, but I don't save any in gas because I get half that when I drive in town. Same difference.

You can't have a savings unless the computer becomes an expensive paperweight constantly idle.

 

[army_ant7]

@luciferano
You may have missed this comment of mine, but it's not a big problem.

I'd think that your statement would be correct about it beating the i5's if the programs used can't evenly use 4 threads. I'm just thinking that theoretically, a quad-core 3.1GHz i5 (for example) has 12.4GHz worth of performance to be exploited (and I'm not sure if it has more cache or anything, too lazy to check), while a dual-core 5GHz Core i3 has only 10GHz to be used, though it has Hyper-threading to squeeze out more performance. I know that there may be other factors that would come into play like overhead and stuff from multi-threading (though I'm not sure, but the i3 may still experience those still) and possibly other stuff. But considering all that, I think it's still pretty tough to conclude that a 5GHz i3 would beat a lower-end desktop i5 at programs using multi-threading at its best for 4 or more threads, unless I'm missing. I bet it would come close though. Single-threaded performance is a no brainer in this case so you won't hear arguments from me there. Looking forward to your reply.

Other than this, I do wonder how much Intel HD graphics overclock, and how they would perform. Just for knowledge's sake of course and the possibility that it may satisfy even a few number of users for budget gaming and old games.

BTW, I am wondering, does anyone know how the HD (Pentium/Celeron), 2000, 2500, 3000, and 4000 compare to the old Nvidia Ion 2 graphics that came with Atom PC's? I guess you could somewhat deduce it using the info on the Best Gaming Graphics hierarchy chart and this article. The graphics chart admits that isn't perfect especially when it comes to mobile and integrated graphics I think. So if the E-350 has a Radeon HD 6310 and the chart says it's in the same tier as the Intel HD 3000, then the Ion 2 might compare to the latter as it did the former in the article.
I'm not sure how the HD (Pentium/Celeron) and HD 2000 would compare to it, but I bet the HD (Pentium/Celeron) would suck compared to it. I think the HD 2500 might still be weaker than the HD 3000, albeit has more features like DirectX11 and true OpenCL support, and possibly even better Quick Sync performance though I don't know if it differs any from the HD4000. The HD 4000 probably beats it (Ion 2) by a long shot.
That article with the Ion 2 may prove relevant as it has a wide selection of games it could give users an idea of what they can expect from the newer integrated graphics.

 

[esrever]

[citation][nom]tajisi[/nom]The problem with that being that if the end user spends 90% of their time idle and 10% doing anything taxing, the larger power draw of the APU quickly eats away any savings that would have accrued. I can get 33 MPG on the highway if I drive carefully, but I don't save any in gas because I get half that when I drive in town. Same difference. You can't have a savings unless the computer becomes an expensive paperweight constantly idle.[/citation]
Trinity:
0.9*32+0.1*122.6= 28.8+12.26=41.06
I3:
0.9*39.9+0.1*80.9=35.91+8.09=44

the i3 will use almost 3 more watts every hour if you use the base of 90% idle time. Note, idle probably includes any time where the CPU usage is under 10% which is a great deal of time from what I can tell from my personally usage.

 

[army_ant7]

Look at the TDPs. The 5700 has a 65W TDP and the 5800K has a 100W TDP. That's probably the reason for the price similarity.

I did notice that, but, as we know, TDP doesn't necessarily mean it will consume that much electricity or generate as much. Because the K-series are also unlocked, you can adjust them (undeclocking and undervolting in this case) to beat or meet those non-K APU's in terms of power and heat. You could also adjust it to be as powerful as your system solution can handle/has to be (in heat, noise, and/or power consumption). I'm thinking this should be true since they are of the same architecture or possibly the same chip configured differently. The K-series might even have higher binning.

Even regular people would look at the clockrate and think it's better. Maybe the lower TDP rating would appeal to them more if they aren't aware of what I said above, or this could be a marketing scheme for AMD to increase initial sales because people might think it's a deal to get the more powerful APU for the same price as a weaker one.

 

[luciferano]

[citation][nom]army_ant7[/nom]@luciferanoYou may have missed this comment of mine, but it's not a big problem. Other than this, I do wonder how much Intel HD graphics overclock, and how they would perform. Just for knowledge's sake of course and the possibility that it may satisfy even a few number of users for budget gaming and old games.BTW, I am wondering, does anyone know how the HD (Pentium/Celeron), 2000, 2500, 3000, and 4000 compare to the old Nvidia Ion 2 graphics that came with Atom PC's? I guess you could somewhat deduce it using the info on the Best Gaming Graphics hierarchy chart and this article. The graphics chart admits that isn't perfect especially when it comes to mobile and integrated graphics I think. So if the E-350 has a Radeon HD 6310 and the chart says it's in the same tier as the Intel HD 3000, then the Ion 2 might compare to the latter as it did the former in the article.I'm not sure how the HD (Pentium/Celeron) and HD 2000 would compare to it, but I bet the HD (Pentium/Celeron) would suck compared to it. I think the HD 2500 might still be weaker than the HD 3000, albeit has more features like DirectX11 and true OpenCL support, and possibly even better Quick Sync performance though I don't know if it differs any from the HD4000. The HD 4000 probably beats it (Ion 2) by a long shot.That article with the Ion 2 may prove relevant as it has a wide selection of games it could give users an idea of what they can expect from the newer integrated graphics.[/citation]

Hyper-Threading makes up for the minor loss between a 5GHz i3 and a 3.1GHz quad-core i5 (of the same micro-architecture) in quad-threaded work. For any software that does not scale roughly 100% across four threads, the i3 would beat the i5 in this example and even in roughly 100% quad-threaded scaling, they would be on-par. Any enthusiast who can buy an unlocked i3 for say $140-160 would probably do so instead of spending more on a locked i5 that can't beat the i3 in much of anything, if anything at all.

A locked i5 still has Turbo overclocking, but that's usually only a 20% gain if you don't include BLCK overclocking that the i3 could also do anyway and it would then be a trade off between huge lightly threaded performance with great highly threaded performance against great lightly threaded performance and huge highly threaded performance. The i3 option is more favorable for most current gaming software and even for future software that will be more well-threaded, the i3 would be quite good and the lower price is a good arguing point. It might not be the best for everyone, but it would decimate i5 sales and for good reason (hence Intel will probably not do it).

 

[proffet]

I did notice that, but, as we know, TDP doesn't necessarily mean it will consume that much electricity or generate as much. Because the K-series are also unlocked, you can adjust them (undeclocking and undervolting in this case) to beat or meet those non-K APU's in terms of power and heat. You could also adjust it to be as powerful as your system solution can handle/has to be (in heat, noise, and/or power consumption). I'm thinking this should be true since they are of the same architecture or possibly the same chip configured differently. The K-series might even have higher binning.

Even regular people would look at the clockrate and think it's better. Maybe the lower TDP rating would appeal to them more if they aren't aware of what I said above, or this could be a marketing scheme for AMD to increase initial sales because people might think it's a deal to get the more powerful APU for the same price as a weaker one.

nice post...
+1