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

[army_ant7]

Oh... When you said CF dongle, you meant the thing bridge attaches to. I thought you meant it as the actual bridge. Well I guess since it plugs into the bridge...

Thank you again for going through my questions! As always, the info you share is greatly appreciated! With all these questions I ask, I hope other people at least read some of the info so that your effort isn't just used up on me. Anyway, will stay on stand-by 'til something comes up again.

 

[blazorthon]

[citation][nom]army_ant7[/nom]Oh... When you said CF dongle, you meant the thing bridge attaches to. I thought you meant it as the actual bridge. Well I guess since it plugs into the bridge... Thank you again for going through my questions! As always, the info you share is greatly appreciated! With all these questions I ask, I hope other people at least read some of the info so that your effort isn't just used up on me. Anyway, will stay on stand-by 'til something comes up again. [/citation]

Yes, when I said dongle, I meant the connector (or as the case may be, the two connectors) on the card that the bridge plugs into.

As I've said before, ask anyway anytime you want to; I'll answer to the best of my ability and knowledge, although it's not always as good as I'd like it to be Some of your questions just go beyond what I have read up on. However, please trust me when I say that for the most part, the answers are there if you can find out where to look. Google might not always be your friend, but can often help anyway. Sometimes, I'd read through several dozen pages of Google search results to find something that I'm looking for. It might take a few minutes (or even a few hours), but if you want to learn about something, then there is usually info out there somewhere.

However, I admit that some subjects are definitely much harder to look up than others... Software CF reviews with newer cards and the most recent drivers is bound to be more or less impossible to find, especially on cards other than the 7750. It's just one of those things that a lot of people don't care to look into and review. I often find it kinda disappointing that some of the most interesting topics are often the most ignored.

 

[Guest]

Chris, can you run some trinity benches with a discrete graphics card? It would be interesting to see how 5800k runs against higher end CPUs like FX-6100/FX-8150 or i3s/i5s on games (specially CPU bound ones, like WOW or Skyrim), just to get a better picture about IPC improvements.

 

[blazorthon]

[citation][nom]LMUY[/nom]Chris, can you run some trinity benches with a discrete graphics card? It would be interesting to see how 5800k runs against higher end CPUs like FX-6100/FX-8150 or i3s/i5s on games (specially CPU bound ones, like WOW or Skyrim), just to get a better picture about IPC improvements.[/citation]

They'll get to that when they have a proper platform instead of a preview/beta setup. This has already been explained; what we've seen here and would see from further testing of the current hardware may not necessarily be representative of what we will see when Trinity hits retail, so this stuff is useless for making valid comparisons against other platforms.

 

[buzznut]

Very interesting article, although its just a preview, I had hoped to see more benchies with the FX 8150 thrown in for comparison. After all, you mention the one thing most AMD fans want to know is how much better will piledriver be?

 

[army_ant7]

Hm... BTW, Chris. I'm wondering if we'd be seeing that video update on this article or on a future article. I really do keep checking back on the article to see if it's there already. A notice would be really appreciated. Thanks!

 

[hikari82]

Can tom hardware also benchmark this apu with emulation software like PCSX2 or Dolphin.. Please.
On A8-3850 only a few title that can be play 60 FPS on 1080P.

 

[hikari82]

I have an A8-3850 .. honestly i want to upgrade .. but AMD desicion to change the socket is not the "hear the loyal fan voice" imho.

 

[SuperVeloce]

Using the same socket means new features needs to be left out. Why would they do this just because you want to use it on a 5 years old motherboard?

 

[blazorthon]

[citation][nom]hikari82[/nom]I have an A8-3850 .. honestly i want to upgrade .. but AMD desicion to change the socket is not the "hear the loyal fan voice" imho.[/citation]

They had to change the socket... You can't add additional display outputs to a system with an on-die IGP without changing the socket to allocate some pins to this. Theoretically, maybe AMD could have at least made FM2 back-wards compatible with FM1 APUs so that current motherboards would support both, but maybe AMD couldn't. If I had to guess, then I'd think that AMD could have done this, but decided not to in order to sell as many Trinity APUs as reasonably possible. However, that has no bearing on the fact that we need new sockets for new features, such as additional display outputs, so sockets change occasionally.

 

[blazorthon]

[citation][nom]tourist[/nom]What new display output are you referring to ? IIRC fm2 will use the old a75 chipset as well as the new a85 chipset. Also what new features?[/citation]

The chipset has nothing to do with the new display output. Trinity supports up to three outputs normally whereas Llano only supports up to two (both numbers true when there is no discrete card installed). There's also a new chipset in addition to current chipset support, A85, but the chipsets have nothing to do with the new socket change.

 

[oxford373]

amd APUs are successful processors ,a laptop with AMD APU is the best choice for many people who want cheap efficient gamming laptop, because laptops with discrete gpu are expensive and less efficient . also trinity is good choice for entry level gamming desktop.
in spite of pentium g630,g860,i3-2100 are slightly faster on games than athlon II x3,athlon II x4,phenome II x4 but those cpus are faster on multithreaded programs and more suitable for multitasking ,but what is the better than quad is the best quad core like i5-3570 ,because all games and most programs don't use more than 4 cores, and because of that i5-2500 or i5-3570 out performs ( quad core i7s Nehalem and phenome II x6 and fx8120) on most programs and all games ,in spite of those cpus are slightly faster than i5 on highly multithreaded programs which most of us don't use this kind of programs , trinity is 10% faster than liano but I am sure if amd remake liano with the same instruction sets of trinity like (sse 4.1,4.2,avx,AES,FMA4....)this revised liano will be much faster and much more efficient .
trinity is just 15%faster IPC(instruction per cycle) than bulldozer, trinity still have 10% slower IPC than phenome II ,maybe fx pile driver have 10% more ipc than trinity even if they would do that it's still 50% slower than ivy-bridge, so i5-2500or i5-3550 as fast as 7 cores of(fx8120) bulldozer and 6 cores piledriver which most users won't use more than 4cores.
i still have question for AMD, does bulldozer deserve to lose for intel for year and 9 months?, as i know fx piledriver won't be available until September ,and if amd don't sell server processors would they ever think to make bulldozer architecture? ,it doesn't matter what kind of quad core would you have if you want a cheap efficient gaming laptop or desktop, and maybe bulldozer is good architecture for servers , but highend desktop cpus must be native cores , actually there isn't any benefit of bulldozer architecture(cores in pairs) at all for laptop or desktop even for servers as you know 8150 cant out performs i7-2600 even at highly multithreaded programs, bulldozer only make a sense for someone looking for a very cheap server cpu .if I am mistaken, at 2010 amd was selling x6 1090priced 300$ which out performs(i7-920,i7-860) quad core Nehalems and many people were happy to spend that much of money on CPU faster than nehalem i7, and if amd remake penome II x6 with the same instruction sets of bulldozer like (sse 4.1,4.2,avx,AES,FMA4....) i am sure that revised x6 will be faster than i7-2600andi7-3770 and fx8150, and I don't know why amd insist to make their cpus using bulldozer architecture when they can sell native 6 cores at 300$ and everyone will be happy to get one of them, which they are right now selling 4 muddles(8cores) bulldozer at 170$ like 8120 and no one want to buy any of them,some people wrote comments about amd cant make good highend CPUs as intel , I am sure amd can, but we have to wait until amd stop fixing what is unfixable, and remake native cores for highend desktop like phenome II , 6 native cores are much faster than( 4mudles- 8cores)and much more suitable for highend desktop,i had read many articles about bulldozer architecture woud have slower IPC because of sharing resources a year before it released so bulldozer was doomed since the first day.

 

[silverblue]

People keep calling for an octo-core Phenom II. The issue is, even with extra instruction sets, the architecture would need a severe boost to perform close to even SB. Remember, SB is massively quicker at integer calculations for a start...

 

[army_ant7]

[citation][nom]oxford373[/nom]but what is the better than quad is the best quad core like i5 ,because all games and most programs don't use more than 4 cores, and because of that i5-2500 or i5-3570 out performs ( quad core i7s Nehalem and phenome II x6 and fx8120) on most programs and all games ,in spite of those cpus are slightly faster than i5 on highly multithreaded programs which most of us don't use this kind of programs ,in spite of trinity is 10% faster than liano but I am sure if amd remake liano with the same instruction sets of trinity like (sse 4.1,4.2,avx,AES,FMA4....)this revised liano will be much faster and much more efficient ,trinity is just 15%faster IPC(instruction per cycle) than bulldozer, trinity still have 10% slower IPC than phenome II ,maybe fx pile driver have 10% more ipc than trinity.
if amd remake penome II x6 with the same instruction sets of bulldozer like (sse 4.1,4.2,avx,AES,FMA4....) i am sure that revised x6 will be much faster than i7-2600ori7-3770 or fx8150, and I don't know why amd insist to make their cpus using bulldozer architecture when they can sell native 6 cores at 300$ and everyone will be happy to get one of them, which they are right now selling 4 muddles(8cores) bulldozer at 170$ like 8120 and no one want to buy any of them,some people wrote comments about amd cant make good highend CPUs as intel , I am sure amd can, but we have to wait until amd stop fixing what is unfixable, and remake native cores for highend desktop like phenome II , 6 native cores are much faster than( 4mudles- 8cores)and much more suitable for highend desktop,i had read many articles talking about bulldozer architecture will have slower IPC because of sharing resources a year before it released so bulldozer was doomed since the first day.[/citation]


Just a friendly piece of advise. Try not saying "all games" since some, or at least one game does use more than 4 threads, supposedly Multi-player Battlefield 3 for example. (Thanks again blaz for that info.).

Those are nice ideas I would think about adding those instructions to PhenomII and Llano or maybe even just Husky cores (32nm, upgraded(?) PhenomII's found in Llano) without IGP's that are available for the AM3(+) socket, though I'm not sure if changing their sockets would still make them Huskies. But I think AMD's interests are just to stray away from that old architecture and try to reach for more integer performance from the CPU and floating-point performance from the GPU through GPGPU. I remember reading this before, though their plans may have changed/been put off for now due to the fact that not much software support GPGPU, and for some that do at least not to a mainstream practical kind of way.

Are FX (CPU-only) Piledrivers expected to yield more IPC compared to Trinity? I mean, Trinity does use Piledriver cores already. Maybe the L3 cache and other stuff then would do it since Trinity doesn't have those.
Would a revised PhenomII x6 really generally beat SB/IB i7's in applications that can, let's say, utilize 8-threads nicely, if they were improved by adding more instructions?

I don't think the module-styled architecture is an "unfixable" thing. blaz convinced me here.

 

[bccorrupt]

I still don't understand who these are marketed toward other than low priced laptops and desktops($350-$500ish). Even the A10 has unusable frame rates on most games on the two higher resolutions, and at the same price you could buy a sandy bridge celeron and a 7750 or GT640. The only thing the A4 and A6 would be usable for(HD video playback, bluray playback, internet browsing) a celeron dual core or G620 would do better. It's tough to admit(being a 10 year AMD only purchaser until this year), but these are as unimpressive as the FX processors.

 

[abitoms]

i was going thru the a10-5800k and the mobile trinity reviews on this site and decided to do some mental math-jugglery with the TDPs of the mobile and the desktop APUs.

now the desktop APUs are 3.8 GHz with 4.2 GHz turbo. So we can take it as 3.9 GHz average assuming most of the time it wont stray too far from the base clocks. And the GPUs is 384 cores at 800MHz.

Doing the same for the top mobile part, i estimate freq~=2.5GHz (1.4GHz below 5800k) and graphics clock around 520 MHz (around 280 MHz below the 5800k)

So looks like adding another 280 MHz (800 - 520) for the iGPU and 1.4 GHz for the CPU side upped the TDP to 100 W from just 35W?

Is this is a wonderful insight from me or just plain tomfoolery?

 

[oxford373]

I hope this post can help to understand my last one .
as I wrote before bulldozer is almost 80% slower ipc than IVB and phenome II is 45% slower than IVB so i5-3570 as fast as almost 7cores bulldozer and 6cores phenom II , and the question why core i5 2500 or i5 3570 out performs i7s nehalem and phenome II x6 and fx 8120 on most benchmarks? ,because 95% of programs don't use more than 4cores ,4.9% of programs 6threaded and 0.1% of programs use more than 6 cores(highly multithreaded) so i5-3570 is unbeatable at 95% of programs and match the performance for 4.9% of programs and slightly slower for 0.1% of programs. so cores wars are over .
and about how amd can make a processor that out performs quad SB\IB i7s, maybe what happened at 2010 can help you, phenome II was 23% slower ipc than nehalem so 23% higher clocks can match the performance of Nehalem's so phenome II x4-975 match i5-750 and x6-1090 slightly faster than i7-920 and i7-860,and they made the memory bandwidth slower than intel to keep them efficient( phenome II have 4000 hyper transport and intel 4800QPI),many of us know higher memory bandwidth don't change anything since CPUs come with integrated memory controller with DDR3 . and if they remake phenome II with the same instruction sets of bulldozer and out of order execution points and a little more ipc they can boost phenome II ipc at least 25% , that revised phenome II will be just 20% slower ipc than ivy bridge ,so a 20% higher clocked revised quad phenome II will match quad IB i5-3570 and hexa will match IB i7-3770,and don't forget i7 is( 4cores /8threads)so the quad Sandy bridge with HT match hexa SNB with Hyper threading disabled so the program must be 8threaded to make i7-2600with HT match i7-3960 with HT disabled ,and because of that x6 1090 is faster than quad i7 nehalems(the program must be 8 threaded so (i7-920/i7-860)can match the performance of x6 1090)but x6 1090 outperformed i7s nehalem in all 6threaded programes and there are more 6 threaded programs than 8 threaded programs,so this revised hexa phenome II will outperform i7-3770 like how x6 1090 outperformed i7-920/i7860 ,and I am not calling for 8 cores revised phenome II I am calling for 6 cores revised phenome II so they can easily increase clocks,and when amd can make architecture have the same ipc as intel architecture it will be the right time for amd to make 500$ 8 native cores CPU ,so it must be native cores for desktop and bulldozer for servers, I don't know if bulldozer really that much better than native cores for servers , if amd cant make a cpu for desktop and another for servers ,so they have to sacrifice a little bit server market to make a powerful desktop processor where they have much more customers .

 

[oxford373]

Just a friendly piece of advise. Try not saying "all games" since some, or at least one game does use more than 4 threads, supposedly Multi-player Battlefield 3 for example. (Thanks again blaz for that info.)

I don't think the module-styled architecture is an "unfixable" thing. blaz convinced me here.

if all games bother you ,so lets just say 99%of games don't use more than 4 threads ,one of the games that use more than 4 cores is metro 2033 but higher clocked quad cores or quad core with higher IPC still faster,keep in your mind that phenome II x4 980 is the best gaming cpu that amd have ever made.
this article can help understand this (Picking A Sub-$200 Gaming CPU: FX, An APU, Or A Pentium? )
http://www.tomshardware.com/reviews/gaming-fx-pentium-apu-benchmark,3120.html
at first year we waited 9 months for B2 stepping of bulldozer and they told us there is a hotfix for bulldozer coming with windows 8 and now we have to wait until drivers and bios get fixed for discrete graphics card benchmarks .
just how beautiful are native cores like phenome II they don't need any kind of software optimization.

 

[army_ant7]

if all games bother you ,so lets just say 99%of games don't use more than 4 threads ,one of the games that use more than 4 cores is metro 2033 but higher clocked quad cores or quad core with higher IPC still faster,keep in your mind that phenome II x4 980 is the best gaming cpu that amd have ever made. .

I just wanted to clarify in case people who don't know might not get the idea that all games are like that in an absolute kind of way. I didn't know that about Metro 2033, but I won't argue with you there a quad-core may still be faster, especially since some of those threads may just be highly or substantially unbalanced, which I would think means that when a core/thread is done with a lighter thread, it can go on to take on another light thread while the others are still working on heavier threads.

at first year we waited 9 months for B3 stepping of bulldozer and they told us there is a hotfix for bulldozer coming with windows 8 and now we have to wait until drivers and bios get fixed for discrete graphics card benchmarks .
just how beautiful are native cores like phenome II they don't need any kind of software optimization.

I find sophistication in the use of software along with hardware to create a harmonious end-product. If AMD does in fact come to something extremely awesome with this in the future, then that would be worth the effort. I'm not counting them out quite yet. Though they could use a lot of improvement with this still though as it seems.

 

[oxford373]

I just wanted to clarify in case people who don't know might not get the idea that all games are like that in an absolute kind of way. I didn't know that about Metro 2033, but I won't argue with you there a quad-core may still be faster, especially since some of those threads may just be highly or substantially unbalanced, which I would think means that when a core/thread is done with a lighter thread, it can go on to take on another light thread while the others are still working on heavier threads.

i didn't write bulldozer is unfixable becuase of these small bugs, i wrote that because amd architecture have much slower IPC than intel's and cores wars are over,i explained that enough,so its all about IPC and nothing else ,I only believe in benchmarks and nothing else, and what i wrote is the conclusion of CPU benchmarks.

 

[army_ant7]

i didn't write bulldozer is unfixable becuase of these small bugs, i wrote that because amd architecture have much slower IPC than intel's and cores wars are over , i explained that enough.

I wouldn't say it's over 'til the fat lady sings (Hehe...), or rather until AMD gives up on improving this architecture. A 15% percent increase per generation might not be able to help though at the rate Intel's going, assuming Haswell and the following will yield higher performance gains. If AMD kicks it up though, they should hopefully get competitive, though I've heard that their management claims that they aren't planning to compete with Intel on that field anymore. If HSA becomes very prominent and effective, their architecture might be more suited for it (more integer units), in tandem with a GPU, that its performance becomes competitive. Just a bunch of ideas that may say that may prove in the future that it isn't in fact, unfixable.

Here's a question/idea for anyone. Isn't Bulldozer amd Pildriver's ability to be highly clocked easily a feature of those architectures? It's just a thought.

 

[zulutech]

Well done Chris, good review.
I think trinity will have a great impact on mobile computing. However, as it stand now, I believe Intel's APU solution which consists of the Intel HDxxxx which are built into Intel's CPUs are better for desktop and laptop uses.

 

[Guest]

So where is the comparison with i3s? and discrete graphics performance?

 

[ashinms]

I don't know where these numbers are coming from about how much software is multithreaded, but the programs I use most on a day to day basis are all coded for more than 4 cores. Battlefield 3, blender, luxrender, an assload of video editors, a few custom-written python scripts that I use to manage my filesystem, metro 2033; they all use at least half of my 8150, and the software that isn't multithreaded doesn't need any more speed than my proc gives (itunes, photobounce, opera- which is actually partially multithreaded and multiprocessed). Not only that, but opera, luxrender, blender(cycles),BF3,metro 2033, and a ton of other software I use employs some form of GPGPU. It's also nice to be able to play a game with a screen recorder running in the background while luxrender finishes the opening animation for a video I'm also rendering that was made from different clips that I transcoded with handbrake taken from the videos that my scripts located for me from my videos that are scattered all over my computer all while photo bounce scours my pictures for pictures of my family and friends.

 

[bah_humbug76]

[citation][nom]oxford373[/nom]amd APUs are successful processors ,a laptop with AMD APU is the best choice for many people who want cheap efficient gamming laptop, because laptops with discrete gpu are expensive and less efficient . also trinity is good choice for entry level gamming desktop.in spite of pentium g630,g860,i3-2100 are slightly faster on games than athlon II x3,athlon II x4,phenome II x4 but those cpus are faster on multithreaded programs and more suitable for multitasking ,but what is the better than quad is the best quad core like i5-3570 ,because all games and most programs don't use more than 4 cores, and because of that i5-2500 or i5-3570 out performs ( quad core i7s Nehalem and phenome II x6 and fx8120) on most programs and all games ,in spite of those cpus are slightly faster than i5 on highly multithreaded programs which most of us don't use this kind of programs , trinity is 10% faster than liano but I am sure if amd remake liano with the same instruction sets of trinity like (sse 4.1,4.2,avx,AES,FMA4....)this revised liano will be much faster and much more efficient .trinity is just 15%faster IPC(instruction per cycle) than bulldozer, trinity still have 10% slower IPC than phenome II ,maybe fx pile driver have 10% more ipc than trinity even if they would do that it's still 50% slower than ivy-bridge, so i5-2500or i5-3550 as fast as 7 cores of(fx8120) bulldozer and 6 cores piledriver which most users won't use more than 4cores.i still have question for AMD, does bulldozer deserve to lose for intel for year and 9 months?, as i know fx piledriver won't be available until September ,and if amd don't sell server processors would they ever think to make bulldozer architecture? ,it doesn't matter what kind of quad core would you have if you want a cheap efficient gaming laptop or desktop, and maybe bulldozer is good architecture for servers , but highend desktop cpus must be native cores , actually there isn't any benefit of bulldozer architecture(cores in pairs) at all for laptop or desktop even for servers as you know 8150 cant out performs i7-2600 even at highly multithreaded programs, bulldozer only make a sense for someone looking for a very cheap server cpu .if I am mistaken, at 2010 amd was selling x6 1090priced 300$ which out performs(i7-920,i7-860) quad core Nehalems and many people were happy to spend that much of money on CPU faster than nehalem i7, and if amd remake penome II x6 with the same instruction sets of bulldozer like (sse 4.1,4.2,avx,AES,FMA4....) i am sure that revised x6 will be faster than i7-2600andi7-3770 and fx8150, and I don't know why amd insist to make their cpus using bulldozer architecture when they can sell native 6 cores at 300$ and everyone will be happy to get one of them, which they are right now selling 4 muddles(8cores) bulldozer at 170$ like 8120 and no one want to buy any of them,some people wrote comments about amd cant make good highend CPUs as intel , I am sure amd can, but we have to wait until amd stop fixing what is unfixable, and remake native cores for highend desktop like phenome II , 6 native cores are much faster than( 4mudles- 8cores)and much more suitable for highend desktop,i had read many articles about bulldozer architecture woud have slower IPC because of sharing resources a year before it released so bulldozer was doomed since the first day.[/citation]


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.