AMD's Kabini: Jaguar And GCN Come Together In A 15 W APU

[mvoinea]

[citation][nom]alextheblue[/nom]I would never run anything artificially restricted to one thread, so in a purely real-world usage scenario, single-thread performance of software capable of running multiple threads is less-than-useful. However, as a complete nerd I still like to see the data. I just hate when those numbers get used and abused (not by you), when it doesn't even matter in many cases anymore.[/citation]
Well, in REAL world you are wrong and every programmer knows it. Most tasks are single-threaded dependant; even if program is multithreaded most internal tasks are single threaded. You can think of professional programs like Photoshop (some filters are not threaded well), CAD, Office, etc. So it's important that we have BOTH single thread performance and multithreading capabilities (just look at Handbrake performance). That's why that in a phone form factor a Cloverfield is generally better in applications than 8-core ARM.
Of course, this days any notebook CPU lower than 4 threads will be to slow, not because of multithreaded programs but because of multiple programs/applications in backgroud (multitabs browser, antivirus, etc).

 

[alextheblue]

[citation][nom]mvoinea[/nom]Well, in REAL world you are wrong and every programmer knows it. Most tasks are single-threaded dependant; even if program is multithreaded most internal tasks are single threaded. You can think of professional programs like Photoshop (some filters are not threaded well), CAD, Office, etc. So it's important that we have BOTH single thread performance and multithreading capabilities (just look at Handbrake performance). That's why that in a phone form factor a Cloverfield is generally better in applications than 8-core ARM.Of course, this days any notebook CPU lower than 4 threads will be to slow, not because of multithreaded programs but because of multiple programs/applications in backgroud (multitabs browser, antivirus, etc).[/citation]I'm saying that in well-threaded tasks, the final performance is all that matters to the user. In other words, scaling. If a program scales well to four or more cores, single threaded performance numbers aren't that important in the real world if you're running 4+ cores. So taking such a program and running only one thread (or otherwise restricting it to one core) might produce a certain result, but is that the performance you'll actually get if you didn't restrict the program? It's neat to geek over, but the final multi-threaded performance is more indicative of the real-world experience the user gets on a particular device/chip.

If a program doesn't scale well at all (bottlenecked), or is single-threaded, then in that program it is relevant. But here's the big rub: In such a scenario, the single-thread performance advantage/deficit will show up regardless, and you wouldn't need to restrict it artificially or DO anything to produce those results. If you take something that DOES scale well and force it to not scale, you're not showing what would really happen on that chip. It's just a commentary... I like seeing the numbers, they're neat to look at, but just as an aside, they're not always something to get fixated with.

As for background tasks like antivirus, anti spyware, firewall, etc., the normal and basic stuff eats up next to no cycles. If you're talking about running a lot of different CPU-intensive stuff at the same time, well that just reinforces my point, since those multiple programs will be spreading it out amongst multiple threads. As for single-thread performance being vital on phones, that's because most existing phone apps (and OS) are behind the times and are poorly threaded, if at all. That's why I stuck to a dual core this go around.

 

[InvalidError]

As for background tasks like antivirus, anti spyware, firewall, etc., the normal and basic stuff eats up next to no cycles.

Depends on which antivirus you are talking about and whether or not live protection is enabled. A few months ago, I discovered that Windows Defender's live protection could end up consuming 60-80% of my i5-3470 when I have multiple programs accessing files at the same time so I had to turn Defender off to prevent my PC from becoming awfully sluggish.

With live protection disabled, CPU load while running said programs drops from 100% to ~50%.

 

[mvoinea]

[citation][nom]alextheblue[/nom]I'm saying that in well-threaded tasks, the final performance is all that matters to the user. In other words, scaling. [/citation]
And I say that most algorithms used in usual programs are not/can't be threaded well. How about Handbrake results here? A relative good CPU threaded program?
About background programs, it's the nature of actual bloatware/bad programing in Windows (also in Android). I sometime have damned Acroread32.exe eating 15% CPU from my i7 notebook (no pdf opened). I have enough power and I sense it when my fan goes up and I kill it (same recomandations from Adobe, like shit happens). And it's one example, for many non-IT users having a low power/core count CPU with standard manufacturer bloatware preinstalled brings simple tasks like browsing to crawl. More, even on clean OS install browsing on many sites isn't smooth because of bad/aggressive flash/ajax bloatware.
At the end it's kind of sad that many times a CPU/GPU is not good enough because of lazy programmers or marketing directions.

 

[cleeve]

[citation][nom]cleeve[/nom]My goal is to see what these platforms are capable of, not to see how I can cripple them.In the same vein, do you also find it odd that we included benchmarks that make use of more than two threads? Should I have eliminated those benchmarks because the Pentium is restricted to two, despite Kabini's ability to handle four?I don't understand your objection.[/citation]

[citation][nom]alextheblue[/nom]Here's the difference, and why you're a bit off-base here. More and more real world software uses 4+ threads, that's a fact of life. Especially software that is really demanding, where it matters. However, I don't think you'll often see Kabini (with it's single memory channel) equipped with DDR3 1333. [/citation]

I'm sorry, I need you to clarify/simplify your point here this because I'm having trouble following your line of thought.

First, AMD supplied the 1333 MHz RAM for the laptops. That's the most common laptop RAM speed by far. Right off the bat, I think it's unrealistic for you to say "I don't think you'll often see Kabini (with it's single memory channel) equipped with DDR3 1333". That's what midrange and low-end AMD stuff is outfitted with. Show me a Brazos that came from the factory with DDR 800 or better. Probably not going to happen. In fact I did a Newegg search for *every* AMD A-series laptop between $300 and $500, and *every one* that listed RAM speed came in at 1333 MHz. Surprisingly, the situation changes for Intel, where it was surprisingly easy to find 1600 MHz RAM in a laptop.

Regardless, your original objection was that the Intel platforms benefited from dual-channel RAM, while Kabini has a single channel, so I shouldn't use two sticks of RAM. Then I replied as above.

From what I can see, you're saying I'm 'off base' suggesting that I shouldn't purposefully cripple one platform because 4+ threads is a 'fact of life' (implying multi-threading is important), while I should hamstring Intel dual-channel memory bandwidth (implying that memory bandwidth is *not* important)?

But then you say:

[citation][nom]alextheblue[/nom]One more way of saying this is that if you're benching the Platform, it only makes sense to run it as it was intended, in its best form.[/citation]

I totally agree, but this part really confuses me in context because it implies that I *should* run the Intel platforms with dual-channel RAM, as they support it. But that is what you protested in the first place, that Intel was benefiting from dual-channel RAM which it supports, while Kabini doesn't support it...?

Unless you're saying that I should only run Kabini in it's 'best form', while crippling Intel?

But, if that's your suggestion, why is that a good idea? That seems to conflict with your statement quoted immediately above.

 

[ingtar33]

Depends on which antivirus you are talking about and whether or not live protection is enabled. A few months ago, I discovered that Windows Defender's live protection could end up consuming 60-80% of my i5-3470 when I have multiple programs accessing files at the same time so I had to turn Defender off to prevent my PC from becoming awfully sluggish.

With live protection disabled, CPU load while running said programs drops from 100% to ~50%.

wow... i didn't think anyone used windows defender. learn something new every day i guess.

Bro, get AVAST! or AVG something on your system. Then grab Comodo Firewall. all are free programs, and Avast! and AVG far superior to that mess that is windows defender; Comodo firewall is the best firewall program you can get.

 

[InvalidError]

wow... i didn't think anyone used windows defender. learn something new every day i guess.

There likely are far many more than you think.

I have used AVG and Avast in the past. Their "Buy Now" splash screen annoyed the heck out of me. As far as live protection goes, I vaguely remember AVG and Avast being just as bad with some of my programs and disabling their live protection as well.

Ironically enough, Windows Defender is the only AV that managed to clean Confliker off my laptop's HDD.

The main reasons I use Defender are that it is included with Windows and does not have nag-screens... and it succeeded where all other AVs I tried have failed me. Since anything is better than nothing, I have no problem sticking to that.

 

[alextheblue]

[citation][nom]cleeve[/nom]I'm sorry, I need you to clarify/simplify your point here this because I'm having trouble following your line of thought.First, AMD supplied the 1333 MHz RAM for the laptops. That's the most common laptop RAM speed by far. Right off the bat, I think it's unrealistic for you to say "I don't think you'll often see Kabini (with it's single memory channel) equipped with DDR3 1333". That's what midrange and low-end AMD stuff is outfitted with. Show me a Brazos that came from the factory with DDR 800 or better. Probably not going to happen. In fact I did a Newegg search for *every* AMD A-series laptop between $300 and $500, and *every one* that listed RAM speed came in at 1333 MHz. Surprisingly, the situation changes for Intel, where it was surprisingly easy to find 1600 MHz RAM in a laptop.Regardless, your original objection was that the Intel platforms benefited from dual-channel RAM, while Kabini has a single channel, so I shouldn't use two sticks of RAM. Then I replied as above.From what I can see, you're saying I'm 'off base' suggesting that I shouldn't purposefully cripple one platform because 4+ threads is a 'fact of life' (implying multi-threading is important), while I should hamstring Intel dual-channel memory bandwidth (implying that memory bandwidth is *not* important)?But then you say:
I totally agree, but this part really confuses me in context because it implies that I *should* run the Intel platforms with dual-channel RAM, as they support it. But that is what you protested in the first place, that Intel was benefiting from dual-channel RAM which it supports, while Kabini doesn't support it...?Unless you're saying that I should only run Kabini in it's 'best form', while crippling Intel?But, if that's your suggestion, why is that a good idea? That seems to conflict with your statement quoted immediately above.[/citation]OK first of all you're getting something mixed up here. I never demanded you run only a single channel on the Intel systems. Maybe someone else did. By all means, run the best memory setup possible for all three! Dual channel 1333 for the Pentium, dual channel 1600 for the i3, and single channel 1600 for Kabini. The only thing I said was that having only a single channel, moving Kabini from DDR3 1600 to DDR3 1333 would have a larger impact (slightly) as a percentage, in games. Of course it's probably a moot point without a LV 13-17W Celeron or Pentium to bench it against.

Second, AMD might have supplied the memory for the laptops... they could have supplied the coffee and whiskey for all I care. But the prototype shipped with DDR3L 1600 AFAIK - that's per Anandtech. Maybe they lied/are mistaken, or maybe YOUR prototype got inferior RAM. Either way I'd like to see them all run with their best memory setups. For the record I've seen plenty of Brazos systems with more than 800Mhz memory. Of course, with the exception of the E-450 (and it's two Brazos 2.0 E2 successors 1800 and 2000), they only supported up to 1066. Which is why it is impressive that at such a low total SoC power envelope, Kabini supports 1600 out of the box.

I've even seen a really cheap $279 15.6" Toshiba with a C-50 (and this was early 2011 before C-60 hit) that had a single 2GB stick of 1066 - the fastest it could support. But I digress... I think most Kabini systems, especially the A4 and up models, will ship with whatever they can support (1333 for the lowend ones, 1600 for E2-3000 and up). I don't feel that strongly about it, it's just my feeling at this point - I don't trust OEMs to do this consistently. So I could very well be wrong on this point. But a 14" Ultrathin prototype with an A4-5000 at a target price of $500? Yeah, I'd be surprised if that one doesn't ship with 1600. The prototype did, according to other reviews.

 

[cleeve]

[citation][nom]alextheblue[/nom]OK first of all you're getting something mixed up here. I never demanded you run only a single channel on the Intel systems. Maybe someone else did. [/citation]

You replied - to my reply - to someone asking why I ran the Intel systems in Dual channel. So you can understand why I might get the impression you were addressing the original point.


[citation][nom]alextheblue[/nom]But the prototype shipped with DDR3L 1600 AFAIK - that's per Anandtech. Maybe they lied/are mistaken, or maybe YOUR prototype got inferior RAM.[/citation]

I wouldn't assume they lied or are mistaken, demo laptops are often shipped with different components. In this particular case it was hard to nail down as Windows task manager reported it as 800 MHz, but the BIOS reported it as 667 MHz, and CPU-Z refused to report memory speed at all. I erred on the side of caution and reported it as the BIOS did (which I could not adjust), and forced that speed on the Intel models. Perhaps Windows 8 was right and it was running at 800 MHz, giving the Kabini laptop an edge. I suspect Anand went by the Windows Task Manager number, assuming we got the same hardware.

Nevertheless, each and every $300 to $500 A4 system on Newegg that lists the specs lists 1333 MHz RAM at the moment. I wouldn't dismiss that out of hand as insignificant, it seems to be a clear trend.

 

[alextheblue]

[citation][nom]cleeve[/nom]Nevertheless, each and every $300 to $500 A4 system on Newegg that lists the specs lists 1333 MHz RAM at the moment. I wouldn't dismiss that out of hand as insignificant, it seems to be a clear trend.[/citation]It's kind of hard to compare the two since current A4 models are all but gone from the market. Anyway, mobile A4 Llano - all but one model supports only up to 1333. IIRC, the only exception in Llano A4 mobile was the rare top-speed A4 MX 45W. Trinity, well there's the A4-4355M... also supports 1333 max. A4-4300M does support 1600, however.

Looking at the few laptops still using 4300M this on Newegg... I see HP's soon-to-be-replaced G6 and G7 which might be 1333. Dunno, they don't say and I wouldn't trust HP. There used to be more options for A4 though, more laptops with 4300M, they've been partially supplanted by A6 Trinity and future A4/A6 (Kabini and Richland) models. Funny enough, the low-power Richland models only support 1333 as well, though with two channels I'm sure.

However, some time ago I looked over Toshiba and Samsung A4 Trinity models which had DDR3 1600. I think there were some reasonably priced Lenovos similarly equipped too, and maybe the odd Acer (though Acer strangely seemed to price their stuff too high as a rule).

Regardless, I hope they use 1600 on the single-channel solutions that are capable of this, at the very least.

 

[Kynn]

Okay, so you're looking for performance on a low price chip; well that's ok, but then let's take the best one ==> Kabini 5200 which is 33% on cpu speed (2Ghz instead of 1.5GHz) and 20% faster on gpu speed.

There, we'll have a look on the performance on games. (Even if TDP is 25W, when I look at your tests at power consumption pentium & core i3 seem much more higher in games.. So it's still interesting to look at)

Something else which I don't like that you do not say: the kabini isn't even available in laptops on the market, so it is CLEARLY to expect that graphic drivers are shit. (knowing AMD)

It'be great to refresh these tests when correct drivers would be released.

But i must say: I don't even understand how you can possibly show an AMD APU with a lower score on a graphic test than an Intel Chip.

I clearly recall all of their shitty chipsets which doesn't even launch a game. I don't think it's possible for them to beat AMD on this side; which explains why I really believe next drivers will do something about it.

 

[Kynn]

Mother of God, I just saw the DDR3 used in your tests: you kiddin' right?!

Even Zacate which is 2 or 3 years old used 1,3GHz DDR3!

And i'm sure you perfectly know Kabini can use up to 1,6GHz DDR3, which clearly gives an advantage for the APU which is using the RAM as GPU Shared Memory.

What does it mean? It means your tests are heavily cpu related and not gpu since you're prioritizing cpu work during games by lowering the gpu's frequency which is bound by the RAM frequency! So yes, you easily give victory to Intel Chips in fact!

Guys! What is that? Is it worth calling it a test for APU????

You SHOULD REALLY change the methods you use for your tests which are clearly not adapted to test an APU. (and NOT a CPU)

 

[esrever]

@cleeve I don't think you should be comparing this to $400 i3 laptops, I seen some i7 refurbished laptops at less than $400. Its the same price bracket, and what is the refurbish even matter when you look only at the cpu. I mean $600 i3 ulvs on sale because they have absolutely no build quality only have 1 thing you test them for, the CPU. For that, you might as well test the i7s.

On that note, I think I seen some desktop i7s at that price too. I don't think its exactly the same thing but the price segment is truly the only thing to really look at. Im sure a i7-2600 desktop on sale for $400 is less portable but a computer is a computer. And since you aren't really testing for the platform and how one would use it, you might as well test the i7. Its also $400, and does that same things.

But wait, you can but a GTX 680 for less than $400, sure you can't run the thing by itself but the price is all that matters and you can potentially get hundreds of times more gaming performance out of a GTX 680 than that kabini chip. This could be an amazing comparison. Im sure if you put this in the i7 desktop, it will be better than 2 kabini laptops. How is AMD suppose to compete with this kind of performance with their NEW apu? Both the GTX680 and the i7 2600 is more than a year old and AMD isn't even close. Even if you tape 20 kabini laptops together, you can't accomplish as much as a i7 and a gtx680. 20 kabini laptops costs $8000 and it still can't be a $800 machine.

Also the comparisons just shouldn't end at the $400 price point, I saw a beat up old car for $400 a while ago on craigslist, not sure if it works. I'm sure it can travel faster than kabini if you push it down a hill. AMD doesn't even look like its trying to compete here but you know, that $400 price point. It can't be a bad comparison.

 

[tajisi]

So many people missing the point. What the reviewer is trying to say, I believe, is along the lines of "AMD's APU tries to do its best, but isn't particularly good at what it was designed to do at this price point." He's not saying it's a bad chip. It's at best an average chip that makes too many compromises to achieve low power usage. In other words it's exactly what some of you griped about when Intel released Atom, but oh how the tides turn on what's acceptable as soon as the logo gets painted green. Cut the author some slack.