AMD "Trinity" APU Models Release Schedule Details Leaked

I can't wait for this, it will DESTROY wimpy ivy bridge!

One of the most optimistic statements of this year!

Pretty impressive. Def see a perfomance boost over Llano.

Hopefully nor another overhyped underacheiving line up... I want to buy something AMD ... but there is nothing out there worth buying from these guys..E350 is probably the only AMD setup that trounces something from Intel.

If the A10 line-up really has the 7660D, and this performs on par with HD6750/HD5750, this would be an awesome-enough-for-casual gaming platform PC... Should wait for the bench soon...

The name will be misleading, like the mobile moniker for their gfx cards.

a 6990m will not resemble anything near a 6990 for example.

I wouldn't like to guess what's coming from AMD, it's been a mixed bag which has been generally disappointing.

Though I would love a graphically capable Llano, as it should be the future if done right.

(graphics core to boost performance where consumers want it, games and video playback, and cutting down on space requirements/power requirements)

I don't hold much hope for anything AMD though.

I was expecting them to release at least a six-core trinity APU, with the A10 name. After all, if they mean to compete with Intel on APUs, the other guys have bad GPU paired with an i5. Maybe the TDP would be too high, but I think it would reach something like 125W, which is acceptable.

Oh, well, let Piledriver come before making judgements...

Bummed to see AMD drop out of the enthusiast desktop race with Intel, but I have to admit it looks like they made a good decision. If those APUs have as much video processing power as it looks like they do it could be an amazing budget choice.

Quad-core 3.8GHz/4.2Ghz Turbo looks nice, especially paired with 7660 graphics.

I will have to wait and see if the graphics are as nice as the model number implies, especially considering that they are way too optimistic with their labeling (I doubt the integrated 7660 would be as nice as a 6950, which by the way is still about $250 for a discrete card). We will have to find out what type of hardware acceleration is included with the APU. At a minimum, I have to assume that better hardware support for transcoding is included. There is simply no reason why Intel's Quick Sync is faster than AMD's solution.

I had hoped that the TDP would be a little lower as well, 100W, though nice, should be more like 85W or even 65W like the A10-5700. I suppose that the graphics component adds to the bottom line, but with the advances in power management and core deactivation, I had hoped for a little bit better.

Time will tell if the revised Piledriver is better than Bulldozer (which is a disappointment to say the least). If the hype is to believed and the graphics performance is on par with a 6950, then methinks that AMD has a winner here. Price may very well determine whether this revised lineup will be a success, or a flop.

Here's hoping for an AMD comeback, so we aren't held captive by Intel in the upper-midrange market. I've been putting off purchasing a new rig because the total cost to move to an Intel 2500K /2600K (including processor, motherboard and a graphics card to replace the pathetic integrated HD graphics) is too much to bear, especially for an AMD fanboi like myself.

This looks quite promesing. I might not get one for a gaming rig but I will defitily get one for a nice SFF HTPC build.

quote: vaughn2k 02/15/2012 1:17 PM
"If the A10 line-up really has the 7660D, and this performs on par with HD6750/HD5750, this would be an awesome-enough-for-casual gaming platform PC... Should wait for the bench soon..."

Well the HD7750 has 512 GCN cores and the same core speed and has 15% more performance than HD6750/HD5750. 7660D has 384 GCN cores, that is 3/4, so the performance will be about 10-15% behind the HD6750/HD5750, that is assuming the memory bandwidth is good enough, so the system should have fast enough memory.

which kinda just makes it as fast as a HD5670 with slow memory and latency taken into account.

Doubtful it will be able to beat Ivy Bridge in the enthusiast market. Unless Piledriver improves Bulldozer by a lot.

But it will be awesome for HTPCs and ultrabooks.

I am an AMD fan, and I even found your comment hilarious!!!!! I agree with both the humour and sarcasm!!! +1

This may sound nooby, but I noticed the package says FM2, while the current APUs are FM1. Anybody know if AMD is basically going to a whole new socket with this, or will FM2 be usable on FM1 boards?

I wanted to build a computer with an A8 3800 - 3820; but have yet to find either chip. Will the A10 5700 actually be available at the consumer level?

Assuming they keep the same hybrid graphics set ups, perhaps it will combine with a 7700 series GPU to get power near that of a 7800 series in Xfire enabled games. That would be a pretty good setup.


Why would it equal a 6950? The first number is the generation, 950 is higher than 660. I read it as "slightly under a 7670" which still might be a pretty large claim considering 384 vs 480 shaders.


Even 100w is impressive and likely a bit underquoted. A DDR3 7570 is quoted at 44w, which combined with a 65w CPU would make something more around 110w. You can't compare the power draw to Intel until you combine it with an equal discrete card. Power management isn't considered when claiming a maximum TDP, it could run much lower at idle. Intel will likely still draw less power though, they have a 22m process.

All in all, looks like a pretty good GPU upgrade, but how will that CPU perform? I'm skeptical here due to use of "cores" when Piledriver uses modules. Is the A10 a 2 module chip then? 2 modules may not perform as well as a true quad core, even if they repair their IPC woes.

Remember AM2 vs AM2+ vs AM3? An AM3 processor could be dropped into an AM2 board, so likely they'll allow an FM2 proc in an FM1 board, but have FM2 boards with better feature sets to fully support the new proc.

As for "core" counts in the slides, I doubt they'd bill something as an AM10, and have it only have 2 modules (4 "cores") when an A8 can be had with 4 modules (8 "cores"). The "core" column is likely equivalent to "modules," just AMD blasting past the marketing hype and semi-accurately going with the 4core/8thread mentality of Intel's marketing. I was hoping for a 6-module/12-core A12 chip myself If you're going to combat Intel, might as well one-up them, even if AMD's official stance is only to match Intel's thread count with "real" cores... Such a loss.

What drugs are you on? Performance numbers for the CPU side of the APU have been abysmal. APU's will be in the value range devices. ULV Ivy CPUs with either integrated Intel graphics or a low end Nvidia GPU will blow any APU out of the water. The APU market was created by AMD because it couldn't compete with intel anymore.

Question?!? This actually sounded like a very interesting product and then I remembered it is bulldozer core. Does the 4 core really mean 2 cores like Bulldozer, or ahem, 4 'integer' cores? In other words, is this 4 cores like Intel's dual hyperthreaded cores or is it truly 4 actual full CPU cores? If it is a true 4 core, I might take a stab at one of these. If it is really a dual core with extra integer cores, then I'll go with Ivy instead.

Why would they do something that would hurt performance? Gaming needs high performance per thread. Bulldozer cores have horrible IPC so they need high clock rates just to get half decent performance. Increasing core count means there's less room for higher clock rates and doesn't help gaming much so it would be rather counter intuitive to cripple their APUs in such a way. Besides that, AMD's APUs are intended for consumer, low budget areas where highly multi-threaded work is a lot less common, so the extra cores wouldn't be much help at all even outside of current gaming.

A 6 core APU would give APU users two nearly useless cores that increase power usage and heat generation and increase the die size, making it more expensive and have poorer yields, making it even more expensive and/or have fewer in stock.

Besides that, APUs aren't high power devices and aren't meant to be. AMD has received enough criticism about their processors using more power than much faster Intel processors use. For example, many of AMD's CPUs use the 125w TDPs, but Intel's Sandy Bridge i7s are faster than anything AMD has on the consumer side in everything yet use equal to or less power than most of AMD's mid range and high end CPUs!

With Ivy Bridge coming out, only AMD's low power, horribly performing processors can keep up in power usage with Intel's fastest Ivy Bridge i7s. Point is that AMD does not want 125w TDPs where they can avoid them.

Moving on, I have to wonder how AMD's increasingly powerful graphics on these APUs is doing with its minute amounts of memory bandwidth. APUs are meant for low budget systems where highly overpriced RAM is much less of an option. As the GPU performance increases, so too does the memory bottleneck. I also have to wonder if the bottleneck affects the CPU cores as well.

In Memory of the Intel Atom, you will not be sadly missed. May you rest in peace. 2008~2012.

7660 sounds like midrange, not value range. I think he was saying the performance would destroy Ivy Bridge on the graphics side. Kinda obvious the CPU side cannot compete with Intel at this stage.

AMD mobile/integrated GPU model numbers have no direct comparison to their discrete counterparts. The 7660 will likely perform on par with the 6470 desktop part IMHO. It won't come close to a 6670 or the like. May very well destroy an IVB IGP, but Intel has promised a serious boost to their IVB IGP, so it will at least be an interesting fight. Intel IGP will likely get a fail simply due to crappy drivers, as always, even if performance ends up being on-par.

There is no Radeon 6470 and desktop APU graphics names do correspond to desktop graphics, the 6550d in the A8 Llano APUs is almost as good as the Radeon 6570.

Ivy Bridge HD 4000 graphics is supposed (as said by Intel) to be 60% faster than HD 3000 which puts it, at best, in the range of the Llano A4s and the Radeon 6450, not even close to the Llano A6s let alone the new A10s. HD 4000 will beat the Radeon 5450 handily, but the 6450 is probably faster than HD 4000 for gaming. 60% is a decent increase in performance, but not even close to being enough to make it worth gaming on. IHowever, I wouldn't game on Llano and probably wouldn't game on Trinity either... If I'm going to game, then I'm getting a 1080p monitor and a Radeon 6950 or GTX 560 Ti from the current generation or similar card from the next generation and playing at 1080p with maxed out settings on every game.

If I can't do that, I'll grab the Radeon 6790 and play at 720p at maxed out settings. If Trinity's A10s can perform at least as good as the 6790 I'll consider one at some point, but I'll definitely overclock it a lot to make up for it's shortcomings. If the price is right, I might do it. Unfortunately, the price isn't right for Llano so I doubt it will be for Trinity. It's close, but I can still get a Sandy Bridge Pentium and a Radeon 5670 and get more performance for less money than AMD's current Llano offerings.

Since an integer core is what we refer to when we say core, yeah... it is a quad core CPU, not a hyper-threaded dual core like an i3. Besides, these are supposed to be piledriver cores, not Bulldozer cores. They should perform better. A module has two cores. The same is true for all other CPUs. Besides, the performance delta between Phenom II cores and Bulldozer cores isn't too big and if Bulldozer had been made better, it would perform better, hence we have piledriver as an improvement. It's still not as good as Intel, but it is better than Bulldozer and at least as good as Phenom, maybe a little better.

A Bulldozer module ins't like two 80% cores, it is two full cores. It performs so badly because of other reasons, not just the architecture. One such reason was automatic transistor paths in performance critical areas that are generally made by hand because computer generated designs tend to perform about 20% worse whilst using 20% more power, pretty much accounts for most of Bulldozer's poor performance and high power usage right there. We've heard many accounts of this problem with the new FX CPUs from several reputable sources, including Tom's if I remember correctly. Then there's the poor performance of AMD's SRAM caches... However, that's been a recurring theme, it's not limited to Bulldozer style architectures either.

Point is that Modular architectures like Bulldozer's aren't bad. We simply had AMD do a crap job of it.

i like to see how the quad core 65w apu working

Hmm ok. My A8 llano has the 6550D with 400 shaders. That the same thing? I've clocked it up to 960mhz, which funny thing, works great on many benchmark suits. A realistic clock speed that actually works in games without causing crashes and restarts is about 930mhz, tied to DDR3 memory at 1600, is a bus width of 128 Bit and bandwidth about 25.6 GB/s, a texture fill rate of 18.0 GTexels/s, and a pixel fill rate of 7.2 GPixel/s.

I can tell you right now stepping up to DDR3 at 1866 isn't much difference at all that I can tell.

In a nutshell, and Im jus pulling this out of where the sun dont shine, from what Ive read or seen, means a difference of playing DiRT 3 on medium settings instead of low, when comparing the current A8 to the future's A10. How far off am I? (to the experts)

Maybe I just want to feel good about adopting the A8 today instead of waiting for the A10. The multitasking experience sounds like it will be well improved, and more drastic a difference than the gpu side of things.

That's cute. I live in the Brazos area of Texas.

VLIW5 shaders (yours) and GCN shaders are pretty different, the 384 will beat the 400 VLIW5.

The Trinity A10s should be significantly faster than Llano A8s in both graphics and CPU performance. We see conflicting reports on by how much, but the lowest I've seen was 20-30%. 1866MHz memory does provide a benefit over 1600MHz, but it will probably provide a larger benefit with Trinity A10s because they will have an even greater memory bottleneck. You definitely get better frame rates even on your A8 from using 1866MHz memory, but it might not be enough of an increase for you to see it.

As for how far off you are? It's hard to say because it depends on several factors, not the least of which the actual performance of these A10s when we see them reviewed. It will also depend on the game being played.

Actually, if you look close at the benchmarks, they compare to the Intel HD 2000, not 3000. It's 60% faster than 2000, which means its only 30% faster than 3,000. NOT TO MENTION that they are using the i7-3770 (Which is around $300). Ivy Bridge is a step up from sandy bridge and will have some nice new features, but nothing close to what AMD will do. Of course, Intel just does what they want since the consumers get herded in their direction anyways.