Battle At $140: Can An APU Beat An Intel CPU And Add-In Graphics?

[blazorthon]

I consider the A-8 3870K to be equal to the i3-2100, both in price and clockspeed, if not in performance as well. When I combine a HD 6670 with the A8-3870K I get more graphics bang for my buck than when I combine it with a i3-2100.
Would I choose another card than the HD 6450, 6570 or 6670, I'd loose the benefit of Virtual Crossfire, but still the HD 6550D of the A8-3870K will be much better than the Intel HD 2000 of the i3-2100.

The A8-3870K is not even close to the i3-2100 in CPU performance in games. It's comparing a greater than $100 APU (sub $100 CPU plus sub $100 GPU) to a $greater than $100 CPU and there's no competition. The i3 is a much higher end CPU for gaming and that is why it is not cheaper to make up for it's IGP being weaker. Clock speed does not indicate performance because the Sandy Bridge CPUs have nearly 60% more performance per core than Llano does at the same clock frequency and Hyper-Threading is enough to take a dual core i3 beyond a quad core A8 in lightly threaded (although highly threaded performance would be pretty close). The i3-2100 is not for being combined with a 6670 or weaker card, it's for being combined with a stronger card where it's higher performance is put to good use. The Pentiums and Celerons are what compete with Llano in CPU performance in games (and still win most of the time) and should be compared to Llano paired with a 6450, a 6570, and a 6670 when the Celeron or Pentium is paired with a 6570, a 6670, and a 6750 (or thereabouts).

 

[Dirk_p_broer]

I own both an i3-2100 and an A-3870K and I can assure you that in real life the performance is much more equal than in a few selected Intel-optimized games. Check e.g. some benchmark scores for both CPUs on CPU-World.
FYI: I've combined the i3-2100 with a HD 4770 and the A8-3870K with a HD 6670, and the CrossfireX combo can beat the i3-2100/HD 4770 combination in quite some applications.
My main interest is Distributed Computing, so-called "crunching" and not gaming.
I do not expect much more performance out of the Socket 1155 mobo till I buy an i5 or i7, and/or a heavy Radeon/nVidia. On the other hand my FM1 mobo can now only be upgraded CPU-wise with a heavy Radeon/nVidia to give more performance, might as well buy a FM2 mobo in a few months (which will fit both Trinity and Piledriver).

 

[blazorthon]

[citation][nom]Dirk_P_Broer[/nom]I own both an i3-2100 and an A-3870K and I can assure you that in real life the performance is much more equal than in a few selected Intel-optimized games. Check e.g. some benchmark scores for both CPUs on CPU-World.FYI: I've combined the i3-2100 with a HD 4770 and the A8-3870K with a HD 6670, and the CrossfireX combo can beat the i3-2100/HD 4770 combination in quite some applications. My main interest is Distributed Computing, so-called "crunching" and not gaming. I do not expect much more performance out of the Socket 1155 mobo till I buy an i5 or i7, and/or a heavy Radeon/nVidia. On the other hand my FM1 mobo can now only be upgraded CPU-wise with a heavy Radeon/nVidia to give more performance, might as well buy a FM2 mobo in a few months (which will fit both Trinity and Piledriver).[/citation]

There are no Intel optimized games and saying that there is is a failure of understanding games and CPUs. Games that prefer the Sandy Bridge CPUs with lower core counts aren't Intel optimized because games don't need to be optimized to work well on Intel. Games often prefer Intel over AMD because they aren't specifically optimized for using many threads and AMD has such poor performance per thread that without being able to use more physical threads than Intel, Intel soars past AMD. So, games need to be specifically optimized to run better on the AMD processors than on Intel, whereas Intel does not need as much optimization and older programs that are often not optimized for high thread counts also run better on Intel.

So, it's not like the games that are optimized for Nvidia over AMD, games simply aren't specifically optimized for AMD, so AMD loses more often because you need to optimize more for AMD than you do for Intel. AMD is working on this problem with Piledriver and Steamroller.

The i3-2100 is about 60% to 70% faster than the A8-3870K for single and dual threaded games (and quad threaded games that can only make good use of one or two out of their four threads, a common aspect of many quad threaded games). Only when a game is specifically optimized to make good use of four threads can the A8-3870K more or less match the i3-2100.

 

[Dirk_p_broer]

When a game uses SSSE 3, SSE 4.1, SSE 4.2 and/or AVX you might as well call it optimized for Intel, as only Bulldozer from AMD's CPUs has these instruction sets.

 

[blazorthon]

When a game uses SSSE 3, SSE 4.1, SSE 4.2 and/or AVX you might as well call it optimized for Intel, as only Bulldozer from AMD's CPUs has these instruction sets.

If you know a game uses AES encryption (AVX is for AES acceleration if I remember correctly), I'd be surprised. Also, Phenom II, Athlon II, Sempron, and Llano have SSE 3.x support, they lack SSE4.x support and games don't use that yet anyway (game programmers want their games to be compatible with all modern platforms and some older ones, so they woulsn't use instructions that the computers of many gamers don't support). Sure, if they did then you could call games that use SSE 4.x as Intel and Bulldozer (and up) optimized, but games don't use SSE 4.x and they don't use AVX either, so that doesn't matter. Name one game that uses SSE 4.x and I'll concede, but I'm not aware of it being used in games right now.

 

[Dirk_p_broer]

Phenom II, Athlon II, Sempron, and Llano have SSE 3.x support

SSSE 3 is a supplemental set to SSE 3 that is not to be found in any AMD CPU before Bulldozer, like SSE 4.1 and SSE 4.2.

Games that make use of these sets can be found on e.g. http://www.intel.co.uk/content/www/uk/en/gaming/overclocking/games-optimized-for-intel.html

You might want to read up about optimzed code on http://wiki.pcsx2.net/index.php?title=GSDX

 

[blazorthon]

SSSE 3 is a supplemental set to SSE 3 that is not to be found in any AMD CPU before Bulldozer, like SSE 4.1 and SSE 4.2.

Games that make use of these sets can be found on e.g. http://www.intel.co.uk/content/www/uk/en/gaming/overclocking/games-optimized-for-intel.html

You might want to read up about optimzed code on http://wiki.pcsx2.net/index.php?title=GSDX

Alright then, that's four games listed as using SSSE 3, only one of which is in Tom's regular benchmarks. One out of what, seven, eight in their usual benchmarks? Furthermore, it doesn't even make a big difference. The very fact that AMD has abandoned their outdated AM3 and older processors means that not only is this not a big deal for them, but that it will also be even less of a problem as time goes on. Furthermore, we can see how the Phenom IIs do well in SC2 regardless of their *disadvantage* of lacking SSSE 3 that according to your links, SC2 uses.

http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120-6.html

It's not a well threaded game, so that they are literally right behind the i3s, instead of ahead of the i3s, should not come as any surprise at all. So, Intel optimized or not, they STILL did almost exactly as good as the i3. The difference is hardly even noticeable because they're so close! So, your argument seems to have little weight to it.

Also, your other link clearly states that you can just use SSE2 DLLs instead of SSE3 and SSE4 and that it won't make a difference, except in what is stated as rare parts that would otherwise be optimized for SSSE3 and SSE4 in which case the difference is still not great. That's quite the unfair advantage that Intel has there, especially with AMD not producing CPUs that aren't incompatible with these instruction set extensions anymore, isn't it?

 

[studio2308]

this whole review is Intel biases BS, I have the A-8 3870k and it preforms worlds better than what you have in this article, in fact i would not be surprised if this was a fake review all together.

 

[cleeve]

[citation][nom]studio2308[/nom]this whole review is Intel biases BS, I have the A-8 3870k and it preforms worlds better than what you have in this article, in fact i would not be surprised if this was a fake review all together.[/citation]

So you're saying because you own a CPU, and haven't compared it against another CPU, or even benchmarked your CPU... you conclude the review is fake?

Please, provide benchmarks of your machine using the apps we tested and the same testing methods.

In short, I think you're full of it, and I call you out.

 

[Wisecracker]

So you're saying because you own a CPU, and haven't compared it against another CPU, or even benchmarked your CPU... you conclude the review is fake?

Please, provide benchmarks of your machine using the apps we tested and the same testing methods.

In short, I think you're full of it, and I call you out.

Down goes Frazier ...
Down goes Frazier ...


Any of you guys old enough to remember that call?

😛

 

[army_ant7]

I own both an i3-2100 and an A-3870K and I can assure you that in real life the performance is much more equal than in a few selected Intel-optimized games. Check e.g. some benchmark scores for both CPUs on CPU-World.
FYI: I've combined the i3-2100 with a HD 4770 and the A8-3870K with a HD 6670, and the CrossfireX combo can beat the i3-2100/HD 4770 combination in quite some applications.
My main interest is Distributed Computing, so-called "crunching" and not gaming.
I do not expect much more performance out of the Socket 1155 mobo till I buy an i5 or i7, and/or a heavy Radeon/nVidia. On the other hand my FM1 mobo can now only be upgraded CPU-wise with a heavy Radeon/nVidia to give more performance, might as well buy a FM2 mobo in a few months (which will fit both Trinity and Piledriver).

Um... I'm just wondering, wouldn't those two setups you provided pave the way for GPU bottlenecks for games? Maybe that's why you think those two processors perform the same. blaz's reply seemed to indicate this.

Clock speed does not indicate performance because the Sandy Bridge CPUs have nearly 60% more performance per core than Llano does at the same clock frequency and Hyper-Threading is enough to take a dual core i3 beyond a quad core A8 in both lightly threaded and highly threaded performance (although highly threaded performance would be pretty close).

The i3-2100 is about 60% to 70% faster than the A8-3870K for single and dual threaded games (and quad threaded games that can only make good use of one or two out of their four threads, a common aspect of many quad threaded games). Only when a game is specifically optimized to make good use of four threads can the A8-3870K more or less match the i3-2100.

Which one of the two is really true?

 

[blazorthon]

Um... I'm just wondering, wouldn't those two setups you provided pave the way for GPU bottlenecks for games? Maybe that's why you think those two processors perform the same. blaz's reply seemed to indicate this.



Which one of the two is really true?

My mistake, sorry, the i3 would still lose in highly threaded performance slightly. I don't know what I was thinking there 🙁

 

[army_ant7]

My mistake, sorry, the i3 would still lose in highly threaded performance slightly. I don't know what I was thinking there 🙁

Oh alright. I was guessing that was the right one based on benchmarks and possibly posts of yours and others around the forums. Thanks for clearing that up.

 

[Raiddinn]

Level 3 cache in processors is pretty useful in games and the A8s don't have it.

Here are some benchmark tests of 2120 vs 3870k both OCd and not:

http://tinyurl.com/8tmtyqz

The video card used alongside each processor is a GTX 460.

Looking through there you can see that the A8 is behind in pretty much every area non-OCd. It is still behind in almost everything even when OCd, although OCing does close a lot of the gap.

The games in particular overwhelmingly favor the 2120 over the A8 even if the A8 is OCd.

Now, these aren't a bunch of cherry picked highly threaded games designed to try and get the A8 to come out ahead, so you have to take these benchmarks with a grain of salt. That being said, there aren't a whole lot of highly threaded games anyway.

The only one I can think of off the top of my head that makes good use of multiple cores is BF3 multiplayer. The multiplayer aspect of games is particularly hard to benchmark, though, since its really tough to have the exact same conditions over and over when there are more people involved than just the tester.

If the game is highly threaded, they are probably about the same. If not its a landslide towards the 2120. Based on that, there is really no downside to getting the 2120 instead of the A8-3870k.

The only reason I would say to get a 3870k at all would be if you wanted to have one less piece of equipment in the equation (ie no video card). Even then you aren't looking at a cost advantage (having to buy high Mhz RAM to take advantage of the A8's in chip graphics) in the direction of the A8.

The only advantage I see is that it leads to a simpler system for low end purposes.

That being said, there is something to say for having a simpler and more error free design.

Having no video card allows for better airflow in the PC and it ensures that neither a bad video card nor a bad PCIE slot will create problems for you when you try to build a new PC.

For someone that doesn't really care to do serious gaming and just wants something easy to slap together that will work right the first time, the advantage of the A8 can be pretty significant.

 

[army_ant7]

+1'ed you Raiddin, though it's not really new info to me.

If the game is highly threaded, they are probably about the same. If not its a landslide towards the 2120. Based on that, there is really no downside to getting the 2120 instead of the A8-3870k.

There is one downside. It's price + the price of a GPU. That's why they used a Pentium here.

 

[blazorthon]

Level 3 cache in processors is pretty useful in games and the A8s don't have it.

Here are some benchmark tests of 2120 vs 3870k both OCd and not:

http://tinyurl.com/8tmtyqz

The video card used alongside each processor is a GTX 460.

Looking through there you can see that the A8 is behind in pretty much every area non-OCd. It is still behind in almost everything even when OCd, although OCing does close a lot of the gap.

The games in particular overwhelmingly favor the 2120 over the A8 even if the A8 is OCd.

Now, these aren't a bunch of cherry picked highly threaded games designed to try and get the A8 to come out ahead, so you have to take these benchmarks with a grain of salt. That being said, there aren't a whole lot of highly threaded games anyway.

The only one I can think of off the top of my head that makes good use of multiple cores is BF3 multiplayer. The multiplayer aspect of games is particularly hard to benchmark, though, since its really tough to have the exact same conditions over and over when there are more people involved than just the tester.

If the game is highly threaded, they are probably about the same. If not its a landslide towards the 2120. Based on that, there is really no downside to getting the 2120 instead of the A8-3870k.

The only reason I would say to get a 3870k at all would be if you wanted to have one less piece of equipment in the equation (ie no video card). Even then you aren't looking at a cost advantage (having to buy high Mhz RAM to take advantage of the A8's in chip graphics) in the direction of the A8.

The only advantage I see is that it leads to a simpler system for low end purposes.

That being said, there is something to say for having a simpler and more error free design.

Having no video card allows for better airflow in the PC and it ensures that neither a bad video card nor a bad PCIE slot will create problems for you when you try to build a new PC.

For someone that doesn't really care to do serious gaming and just wants something easy to slap together that will work right the first time, the advantage of the A8 can be pretty significant.

The games where the A8 lost significantly in, it still got from 54FPS to around 70FPS. If the A8 couldn't get near or above 60FPS, then I'd call it a loss, but it can, so that's irrelevant. That overclocking would then let it make up most of the lost ground (and there's more that can be done than mere overclocking for the more adventurous) and propel it well over 60FPS in all tested titles except maybe for Crysis 2 does not make the i3's case any better. Given that those tests only did a 20% overclock, this much is obvious.

Also, the performance differences are not impacted by the cache differences too much. The A8 still has more cache than the i3 has anyway. It has 1MiB of L2 per core whereas the i3 has 256KiB per core with a small L3 cache shared among both cores. The performance difference here is primarily Sandy Bridge's performance per core advantage.

 

[ezzep123]

Good grief...you all are still fighting over this???

There's a reason I don't come to Tom's Hardware anymore...all you guys do is fight like your lives depend on it! I try to get advice on something, and people take the thread over..

That being said, my wife has a laptop...Toshiba Satellite p775 or something like that...2.4ghz i5 SB w/6gb of ram and an nvidia 540m. My a8 can hold it's own against it, by overclocking the GPU part. You might think "What kind of crap are you throwing??" How many of you have real-tested an i5 and an a8 side-by-side?? Huh?? How many? None? You're all going by what Tom's Hardware's tests say? How did Intel convince Dell to sell Intel products exclusively? By giving them really good products? Nope, they paid Dell off. I don't know if the same thing is happening on Tom's Hardware, but I will say this--do some real-world comparisons YOURSELVES and quit arguing like a bunch of kindergarteners.

 

[cleeve]

You might think "What kind of crap are you throwing??

No, but I am curious if you actually benchmarked it.
Sounds like you just used them one after the other. That's not really objective.

Do some real-world *OBJECTIVE, MEASURED, AND REPEATABLE* comparisons YOURSELF before pointing fingers, please.

You're all going by what Tom's Hardware's tests say? How did Intel convince Dell to sell Intel products exclusively? By giving them really good products? Nope, they paid Dell off. I don't know if the same thing is happening on Tom's Hardware, but I will say this...

It also sounds like you're making a vague implication that Tom's is being paid off because your seat-of-the-pants benchmark method leads you to believe your A8 is as fast as the 540M (which it probably is by the way, the 540M is substantially slower than the 6670 DDR3 used in this article. A working knowledge of the hardware in the playing field helps when you're critiquing integrity, too).

Incidentally, we've already proved an overclocked A8 is able to match the faster 6670 DDR3... check the newer article, http://www.tomshardware.com/reviews/a8-3870k-apu-overclocking-guide,3260.html

It's the CPU side that's holding the A8 back in these benches, probably. There's a ton of evidence that Sandy Bridge kicks out higher IPC than Llano. Unless every reviewer who ever tested them is being paid off, I guess.

Are you saying that Sandy Bridge has lower IPC than Llano and everyone who ever tested them is paid off to say otherwise? Based on your seat-of-the-pants benchmark method?
Is that your platform?

 

[ezzep123]

Wait...you tested this using 1600mhz ram...not 1866. Mine uses 1866, not 1600. Yeah, you're right....my benchmarks would be wrong...after I OC'ed the CPU by .5ghz, and then upped the GPU 250mhz. Yeah, my benchmarks would be different than yours.
Did you ever test with 1866? No, ok then, you're telling me my seat-of-the-pants method is a failure then? You're telling me that my machine isn't capable of doing what I want it to do, when you haven't used the faster ram?

And I truly don't see the big deal about Sandy Bridge/Ivy Bridge. You gotta understand...I grew up gaming with an athlon that kicked the crap out of the pentium 4s..and the athlon was clocked 1.5ghz lower than the p4. So when I saw this article way back when, about you guys comparing a pentium to an a8, it ticked me off. It was like you or whomever was saying "Jeez, AMD sucks so bad that we're going to compare it to a Pentium." So, I realize now that you're not meaning to say that. It's just that all the Pentiums I use at work seriously suck...with HT or not.

How about you guys run an article comparing the APU C series or E series vs. the Atom? That'd be a fair comparison. And I'm not just meaning netbook. How about the Atoms that have been built into the Google TVs, if they're still being made.

 

[blazorthon]

Wait...you tested this using 1600mhz ram...not 1866. Mine uses 1866, not 1600. Yeah, you're right....my benchmarks would be wrong...after I OC'ed the CPU by .5ghz, and then upped the GPU 250mhz. Yeah, my benchmarks would be different than yours.
Did you ever test with 1866? No, ok then, you're telling me my seat-of-the-pants method is a failure then? You're telling me that my machine isn't capable of doing what I want it to do, when you haven't used the faster ram?

And I truly don't see the big deal about Sandy Bridge/Ivy Bridge. You gotta understand...I grew up gaming with an athlon that kicked the crap out of the pentium 4s..and the athlon was clocked 1.5ghz lower than the p4. So when I saw this article way back when, about you guys comparing a pentium to an a8, it ticked me off. It was like you or whomever was saying "Jeez, AMD sucks so bad that we're going to compare it to a Pentium." So, I realize now that you're not meaning to say that. It's just that all the Pentiums I use at work seriously suck...with HT or not.

How about you guys run an article comparing the APU C series or E series vs. the Atom? That'd be a fair comparison. And I'm not just meaning netbook. How about the Atoms that have been built into the Google TVs, if they're still being made.

Modern Pentiums don't have HTT and Sandy/Ivy Bridge have far higher performance per core per Hz of the CPU frequency.

 

[ezzep123]

How modern do they have to be for me to forget how bad the P4 sucks? Well, they have to get rid of the name entirely.

Yeah, I'm not suggesting that Ivy/SB suck. I'm just not sold out on them. If I was, I would've built a system for my own. But I refuse to let my money go to Intel. I really don't care how much better they are.

 

[ezzep123]

OK, so I'm supposed to use Adobe Photoshop CSS for a benchmark? Yeah, I'm going to go out of my way and budget, and buy a $1000 program just to prove you wrong...whatever.

Why do you use Winzip for reviews? Who uses Winzip these days? Windows 7 has a built-in extracting program. If that's not enough, you use 7zip as well.

Yeah, if you would provide me with a check or rebate to call you guys out, then I might do it. I mean, seeing the review is free for me to read and review.

 

[ezzep123]

One more--You used a version of 3DS that was from 2010. It's 2012 now. Is there an updated version you could use? AMD's Fusion series wasn't released until last year.

 

[Wisecracker]

Somewhere deep in a dark corner of the THG Mad Scientists Lab, our intrepid author is benching an over-clocked AMD Trinity A10-5800K APU on a Giga-byte GA-F2A85X-UP4 motherboard, in preparation for an update .....

Go, team THG! Go!

[:lutfij:4]

 

[army_ant7]

Somewhere deep in a dark corner of the THG Mad Scientists Lab, our intrepid author is benching an over-clocked AMD Trinity A10-5800K APU on a Giga-byte GA-F2A85X-UP4 motherboard, in preparation for an update .....

I really hope you're right! 😀