AMD Trinity On The Desktop: A10, A8, And A6 Get Benchmarked!

[army_ant7]

[citation][nom]bah_humbug76[/nom]Wow hang on a minute, so your saying with a 25% improvment Piledriver would still be 50% slower than IB, so what your saying is IB is 75% faster than Bulldozer..... Have a word with yourself.[/citation]
Bulldozer = BD
Ivy Bridge = IB
Trinity = T = 1.25BD
Piledriver = PD = 1.10 T =
1.375BD = 0.5IB
1.375BD/0.5=IB
2.75BD=IB

I'm just doing the math and anyone correct me if I'm wrong. If ever what oxford373 says is true, Ivy Bridge is 2.75x faster than Bulldozer. Is anyone buying this? I don't mean to say that oxford373 is wrong, but I'm having a hard time believing this and if anyone can come up with benchmarks, you're welcome to share.

Addition: When I say faster, I mean IPC since that's what oxford373 gave. 2.5x is also another way of saying 150% faster.
Actually, now that I remembered that it was IPC, it does sound plausible, though a Bulldozer vs. Ivy Bridge IPC tests is still in order, and I'm not sure if that's an easy task since applications don't always use all the IPC that a CPU has to offer I think. I wonder if there's a benchmark for this.

 

[bah_humbug76]

[citation][nom]army_ant7[/nom]Bulldozer = BDIvy Bridge = IBTrinity = T = 1.25BDPiledriver = PD = 1.10 T = 1.375BD = 0.5IB 1.375BD/0.5=IB 2.75BD=IBI'm just doing the math and anyone correct me if I'm wrong. If ever what oxford373 says is true, Ivy Bridge is 2.75x faster than Bulldozer. Is anyone buying this? I don't mean to say that oxford373 is wrong, but I'm having a hard time believing this and if anyone can come up with benchmarks, you're welcome to share. Addition: When I say faster, I mean IPC since that's what oxford373 gave. 2.5x is also another way of saying 150% faster.Actually, now that I remembered that it was IPC, it does sound plausible, though a Bulldozer vs. Ivy Bridge IPC tests is still in order, and I'm not sure if that's an easy task since applications don't always use all the IPC that a CPU has to offer I think. I wonder if there's a benchmark for this.[/citation]
http://www.tomshardware.co.uk/a10-5800k-a8-5600k-trinity-apu,review-32476.html

Piledriver looks pretty evenly matched to the i3 to me, it beats the i3 in 2 out of 5 Sandra aspects and the other 3 are not far behind, it also beats the i3 in CS6 and Handbreake, it get close in CS5, the whole thing is give and take or close even steven.....

oxford373 statment there is just a vast exaguration...

 

[SuperVeloce]

exaggeration? 2core Intel vs. 4core AMD? How is it then, that Ivy IPC is an exaggeration? I know it's not traditional fully fledged 4 core cpu, but AMD marketing says it is

 

[blazorthon]

[citation][nom]SuperVeloce[/nom]exaggeration? 2core Intel vs. 4core AMD? How is it then, that Ivy IPC is an exaggeration? I know it's not traditional fully fledged 4 core cpu, but AMD marketing says it is[/citation]

It is a quad core. It only has two FPUs, but it does in fact have four integer cores, although they are in two modules of two tightly linked integer cores. Also, how many cores there are in any given CPU has little to do with performance per core and it most certainly has nothing to do with IPC.

If all four cores are stressed, even the FX-4100 can outmatch an i3 by a little at stock frequencies and substantially when overclocked. A Trinity version or even better, a Vishera version, could be much faster than an i3 and a Vishera quad core is likely to be truly closing in on performance per core compared to AMD's previous generation.

 

[SuperVeloce]

You do not need to explain this to me. We were talking about IPC, not core count and frequency difference. People do not understand that, thats why someone downgraded my comment. If you talk about IPC, then you cannot compare 4 core AMD vs. 2 core Intel with different clocks and say "AMD is faster 2 out of 5". IPC is not the same as "value for money" as someone would suggest with their comparisons. Again, we are talking about IPC of different arhitectures (instructions per cycle: one core, same freq.). I actually argued against comparison of cpus with different core count. I argued against bah_humbug76 comment

 

[blazorthon]

[citation][nom]SuperVeloce[/nom]You do not need to explain this to me. We were talking about IPC, not core count and frequency difference. People do not understand that, thats why someone downgraded my comment. If you talk about IPC, then you cannot compare 4 core AMD vs. 2 core Intel with different clocks and say "AMD is faster 2 out of 5". IPC is not the same as "value for money" as someone would suggest with their comparisons. Again, we are talking about IPC of different arhitectures (instructions per cycle: one core, same freq.). I actually argued against comparison of cpus with different core count. I argued against bah_humbug76 comment[/citation]

I'm not the one whom downvoted your comment. Besides, Ivy Bridge's integer performance per Hz (not exactly the same as IPC despite many people mistaking them) per core is only about 60% to 65% ahead of BD. It's not even double. With even Trinity being ahead of Bulldozer, That gap closes significantly, granted Haswell is on the way and could refute that gain fairly well.

Regardless, there no way to compare these Bulldozer and Trinity CPUs against Intel CPUs of the same core count in a fair way because of the budget differences. Intel cheapest Sandy Bridge quad core is more expensive than AMD's cheapest eight core and AMD's cheapest 6 core is only somewhat more expensive than Intel's cheapest dual core CPUs. If you think that it's fair to compare an ~$200 i5 to aan $100 FX-4100 or similar jsut because they're both quad core CPUs, well, I'll have to disagree completely. Sure, the i3s and the FX-4xxxs excel in different workloads, but they are in the same price range and are thus the competitors. The FX-4xxxs and such do not compete with the i5s and the i7s and their having the same core count changes nothing.

 

[army_ant7]

Okay, what I'm about to say is something that comes off as a result of being unbiased. And before I forget, thanks blaz about that info about the newer Intel memory controllers. (EDIT: Oops. That "thank you" should've been on the other forum. Got them mixed up. Hehe...)

This is a question about IPC so even though your argument about it being unfair to compare the architectures directly due to price differences, is sound, it might've not been the best thing to retort to SuperVeloce as he wasn't talking about that, though again, it is a sound point.

You were possibly right IMO about there being no way to compare these module-based architectures to the more traditional architectures employed by the Bridges. Here's my take on it though. Traditionally, you could just compare one core to another core because a core back then used to be built with the same components/units and weren't so intertwined that they shared something as vital as an FPU. So IPC could be more observable through the correlated performance per cycle (PPC) which is evident in benchmarks. Again, that's why I was wondering if there really is a benchmark that could rate a CPU's IPC fairly accurately. Anyway, a one-to-one core comparison seemed practically valid then, though when comparing CPU's you should really be idealistically comparing the CPU's as a whole, IMO. I'm not sure if anyone would agree with me. Though again, since a core was pretty much an exact fraction of the whole CPU, it would yield practical results in tests when compared to another core.

But now, AMD's module-based architectures arrived, and we know that a core isn't the same as it used to be. 1 core isn't a perfect fraction of the CPU as a whole anymore since you'd either have to have it come with an FPU or not, and maybe other parts/units as well, in our current cases.
I don't think it's fair deriving IPC comparisons by comparing:
1) 1 module with a disabled the other (integer) core in the module since the FPU was designed to be shared between those 2 cores in that 1 module and whatever implications that may have on its IPC should be given value.
2) 1 module to one Hyper-Threaded core.
3) 1 module (2 cores) to two cores of another more traditional CPU.
4) 1 CPU to another CPU... Oh wait! It does seem fair...kind of. Just like what I proposed, comparing CPU's as a whole. Getting a benchmark that can take advantage of both traditional cores and module-based ones to their fullest extent would be nice. It's synthetic though and not real-world, but that's part of sticking strictly to IPC is IMO. One caveat though is not being really able to compare architectural capabilities like before without taking price into the matter, just like what blaz seems to hold strongly for, to compare the CPU's as a whole (for their price points). Another caveat is comparing specific CPU's (SKU's) instead of the architecture itself. Other comparisons like performance/temperature, performance/die size, energy efficiency may still be used for architecture comparisons I would think.

I do agree with you about what I think you feel about AMD's marketing. This would hopefully serve them a lesson about possibly deceiving people who are less knowledgeable about this stuff and would advertise their CPU's/APU's by modules not cores for they can be and I bet are confused with traditional cores.

Sorry for being so thorough, because you guys probably know more than me on this matter already, which is evident since I've learned so much from you guys, especially blaz. It just happened to come with my take on it.

A final thought is that oxford373 shouldn't have said what he/she did about comparing percentages with everything I've said in mind IMO.

 

[blazorthon]

...
Okay, what I'm about to say is something that comes off as a result of being unbiased. And before I forget, thanks blaz about that info about the newer Intel memory controllers. (EDIT: Oops. That "thank you" should've been on the other forum. Got them mixed up. Hehe...)
...

Well, at least generally, with IPC and such, we're talking purely about integer performance of each core at a given clock frequency (granted that as I've said before, this isn't really exactly what IPC is, just what most people like to treat it as), not so much FP performance. Comparing CPUs as a whole when they have similar prices is what I'd call fair comparisons so long as they aren't used to come to ridiculous conclusions. For example, people saying that the FX-8120 or the 8150 is extremely slow jsut because it doesn't beat other CPUs in every single workload. Heck, that's no better than saying that the i5s are slow just because they can't keep up with the 8120/8150 and the i7s in highly threaded integer work. Basically, yes, you expanded on the main point here in that a CPU should be compared to another CPU as a whole, or more specifically, compared to another CPU as a whole so long as that other CPU is in the same or at least a similar price range.

 

[Guest]

Badass! Can't wait to find out what will Hybrid Crossifire whith this.

 

[abitoms]

In the URL below, there is an earlier comparison between 6670 against Llano 3870k.

http://www.tomshardware.com/reviews/pentium-g620-amd-a8-3870k-radeon-hd-6670,3140-6.html

As you can see, at 1440 x 900, the 3870k with its weaker-than-Trinity GPU puts out around 48 FPS while in this Trinity preview, the stronger 7660D puts out less than 45 FPS.

What gives...can someone throw light on this pls? Did I misinterpret anything, or is it due to non-comparable test setup ...

 

[blazorthon]

[citation][nom]abitoms[/nom]In the URL below, there is an earlier comparison between 6670 against Llano 3870k.As you can see, at 1440 x 900, the 3870k with its weaker-than-Trinity GPU puts out around 48 FPS while in this Trinity preview, the stronger 7660D puts out less than 45 FPS.What gives...can someone throw light on this pls? Did I misinterpret anything, or is it due to non-comparable test setup ...[/citation]

The tests in the two articles aren't the same.

 

[Wisecracker]

Badass! Can't wait to find out what will Hybrid Crossifire whith this.

Turks Pro GPU -- Radeon HD 6570 DDR3

OC the GPU/RAM clocks as high as they'll go -- sync 'em up the best yah can.

 

[Guest]

Batman Arkham City has some serious issues with DX11...not sure why. That could explain why dual graphics didn't help it.

 

[jonahkirk]

I'm still amazed it can't do better in Skyrim. My laptop with a core I-7 2720 and 6770M (clocked to 7690XT speeds) plays Skyrim at high settings at 1080P. I would think a desktop proc could push a little harder. I would think the graphics engines would be more comparable. Then again, my llano 3850 could game unplugged for a long time (had to return), this thing dies in no time unplugged even using intel's graphics.

 

[jonahkirk]

I'm still amazed it can't do better in Skyrim. My laptop with a core I-7 2720 and 6770M (clocked to 7690XT speeds) plays Skyrim at high settings at 1080P. I would think a desktop proc could push a little harder. I would think the graphics engines would be more comparable. Then again, my llano 3850 could game unplugged for a long time (had to return), this thing dies in no time unplugged even using intel's graphics.

 

[luciferano]

It's still nice to see how Trinity consistently beats Bulldozer by 15% per core at the same frequency despite Bulldozer having a large L3 cache advantage. Vishera could easily have a more than 25% performance per Hz advantage over Bulldozer while hitting higher frequencies too and might even scale better with CPU/NB frequency overclocking as a result of that.

With one core per module disabled or something similar to alleviate the front-end bottle-neck of each module and the CPU/NB frequency overclocked to further reduce it in other ways, this might even be able to keep up with Haswell in overclock versus overclock gaming performance. It should have no trouble beating Sandy/Ivy Bridge when used in this way except maybe if you took Ivy Bridge and swapped out Intel's paste for some AC5 or better.

AMD most certainly has a lot of hope in them, especially if we also consider that Steamroller is set to have another 20-30% performance improvement over Piledriver and Excavator is supposed to have an even greater improvement over Steamroller, all each within about a year of each other. So, each would possibly be comparable to Intel's tocks, except about twice as often. Intel would be forced to step up their game or else be passed up by AMD.

The CPU market might get more interesting in the next few years, especially more so with how modern software is getting more and more well-threaded, IE more in favor of AMD's MOAR CORES approach.

 

[luciferano]

[citation][nom]jonahkirk[/nom]I'm still amazed it can't do better in Skyrim. My laptop with a core I-7 2720 and 6770M (clocked to 7690XT speeds) plays Skyrim at high settings at 1080P. I would think a desktop proc could push a little harder. I would think the graphics engines would be more comparable. Then again, my llano 3850 could game unplugged for a long time (had to return), this thing dies in no time unplugged even using intel's graphics.[/citation]

Well, the Intel mobile i7s aren't much slower than the top i3s and a Radeon 6770M is too powerful to let any integrated graphics really show it up just yet. Also, to be fair, the A10 APUs probably uses no more power than your Intel laptop uses if you combine the CPU and the GPU power consumption in a comparison against the Trinity A10 APUs, which is kinda important when you consider that mobile i7s are highly hardware is very highly binned to not use too much power while performing decently.

 

[abitoms]

"luciferano:
....Vishera could easily have a more than 25% performance per Hz advantage over Bulldozer while hitting higher frequencies too and might even scale better with CPU/NB frequency overclocking ....."

I'd say let's hold our horses abt the performance estimates, in just 7 days we all will know. the reason I'm saying this is .... i know you came up with the estimates in a well-meaning sense and we both and others want amd to do better than it now currently.

I'm only scared of some ppl taking our friendly, next-door estimates and running away with them as if they were official

 

[Reynod]

Sweet article ... the numbers match what I was able to translate from the first example that popped up on that Taiwanese site - the mobo guy who does the bios programming.

These look quite good though the power jumps a bit high for the fastest cpu in the lineup ... I guess its ok for a desktop part in defence.

I hope AMD can ramp these up in terms of production ... that was the issue with the previous iteration ...

 

[re-play-]

where is the FX8150 in those test as u listed the cpu on the test benchmark?

 

[Wisecracker]

Sweet article ... the numbers match what I was able to translate from the first example that popped up on that Taiwanese site - the mobo guy who does the bios programming.

These look quite good though the power jumps a bit high for the fastest cpu in the lineup ... I guess its ok for a desktop part in defence.

I hope AMD can ramp these up in terms of production ... that was the issue with the previous iteration ...

Not sure of the exact time-frame, but my understanding is that AMD Trinity has been in final production since February. If yields are respectable they've had the wafer-starts to bin all types of desktop/mobile product -- even a line of rumored Trinity Optys.

And ... even if a chip is rated up to 100w it does not mean modules or even the SIMD-engine graphics cannot be run in an efficient low-power state.

AMD has been quite fancy as of late with their Zero-Core and PowerTune energy stuff - hopefully, they can bring some of it together on Trinity (if not take it even further with Kaveri).

 

[spamu2]

I'm not seeing the logic behind republishing this review. If there is nothing new added, why bother?

 

[luciferano]

[citation][nom]spamu2[/nom]I'm not seeing the logic behind republishing this review. If there is nothing new added, why bother?[/citation]

If desktop Trinity is launching soon, then bringing it back up again is a good idea.

 

[GuiCR]

Nice! I Love AMD!

 

[penguintech]

Very Very nice.... too bad I am not an AMD fan.