New Amd "Llano" Slides

[itsmydamnation]

No its not. To put it in layman's terms:

DDR3: Frequency x 8
GDDR5: Frequency x 8 x 2

Double pump the double pump and you get GDDR5.

doesn't GDDR also sacrifics latency for extra thoughput?

 

[busydude]

The real question is if people are willing to buy AMD. This was the main hurdle when they had a CPU that was superior in almost every way than Intel. They couldn't get the average person to choose AMD over Intel.

Blog

Back when I just started my undergrad back in India, Without having any hardware knowledge whatsoever, I bought a P4 1.8GHz desktop from HP at an insane price. At that time I was told wthat AMD processors run hot and are slower when compared to Intel, when infact the opposite was true.

People take advantage if the customers are not savvy enough, the best example is the way B&M hardware stores mislead customers into buying POS.. PSU's that explode if you add an additional HDD.

I know a lot of my friends who were misinformed, and were led into buying Intel based desktops. Don't get me wrong I have nothing against them.. I would have never bought a Macbook and also recommend SB builds for my friends. It was a [successful] hype they managed to create back in those days.

/Blog

On topic, totally agree with you. I read somewhere (possibly a leak I guess).. that AMD are expecting Llano based products to take up 60-70% of its revenue in consumer market.. the rest being Bobcat and BD based products.

Most of the enthusiasts do not care about these products, but they are the ones that make up most of AMD's/Intel's revenue in consumer space.

Llano, atleast on paper, seems to strike a good balance between CPU and GPU which Intel is severely lacking in until now.. although SB is changing that.

 

[ElFenix]

doesn't GDDR also sacrifics latency for extra thoughput?

usually latency goes up as frequency increases, so i would guess so.

latency is usually expressed as a number of clocks, so is there a real latency difference between 1 ghz ram with a latency of 10 and 500 mhz ram with a latency of 5? doesn't look like it to me. more realistic numbers may be 12 and 5, though, which would be more latency.

 

[wahdangun]

i just hope some oem will be start bundling bf3 with this llano based system and promote pc gaming.

 

[ydnas7]

AMD's entire strategy is centered around creating niches where they can outperform their competition, then exploiting those niches to drive unit prices up to a profitable and sustainable level. For the most part, these niches are not niches that people reading hardware forums are most interested in, however, in some cases they do represent a pretty significant portion of the market.

With LLano being 30%-50% larger in die size than its Intel equivalent, it will be pretty easy for Intel to set a price that allows Intel to make a profit, but deny AMD a profit.

 

[podspi]

With LLano being 30%-50% larger in die size than its Intel equivalent, it will be pretty easy for Intel to set a price that allows Intel to make a profit, but deny AMD a profit.

What is Intel's equivalent? At this point in time, Intel has no CPU or IGP that can match Llano (in terms of GPU performance), so Intel HAS no equivalent. Has there been any information on design wins of Llano yet?

 

[jones377]

What is Intel's equivalent? At this point in time, Intel has no CPU or IGP that can match Llano (in terms of GPU performance), so Intel HAS no equivalent. Has there been any information on design wins of Llano yet?

Which begs the question whether the CPU or GPU performance will set the prices for Llano. Will AMD dare price it the same or higher than SB even though it's CPU performance will likely be significally lower and GPU significally higher ?

 

[IntelUser2000]

Which begs the question whether the CPU or GPU performance will set the prices for Llano. Will AMD dare price it the same or higher than SB even though it's CPU performance will likely be significally lower and GPU significally higher ?

It wouldn't make sense if they price it too high, because then previous AMD processors + equivalent level graphics at a lower price point might make more sense.

 

[podspi]

It wouldn't make sense if they price it too high, because then previous AMD processors + equivalent level graphics at a lower price point might make more sense.

For us, yes. For OEMs, it saves them inventory and production costs, which could be important.

 

[jones377]

It wouldn't make sense if they price it too high, because then previous AMD processors + equivalent level graphics at a lower price point might make more sense.

Oh I firmly believe AMD will continue to give away their processors, I'm just not sure that's the best strategy for them because it just continues the perception that their processors are slower in all regards compared with SB. Llano should have a good power advantage vs their previous gen/discrete even if performance is roughly the same. That should be worth some extra dollars.

 

[Dribble]

"Not sure where company will use the video part of this chip though."

jeez do people still not understand what fusion is about?

The GPU isnt just for graphics... its a powerfull processor on its own (the GPU), thanks to things like CUDA (nvidia only), DirectCompute,APP,OpenCL. A GPU doesnt "just" do graphics, it can do more. (which is the point of fusion, with APUs, to have strong GPGPU)

We ll see alot of software with GPGPU come in the future, because everyone supports it now, and wants it... right from IMB,INTEL,AMD,NVIDIA,MS.

You could do all the above without the gpu being part of the cpu? Infact the ones that do it best are nvidia and obviously their gpu's aren't part of the cpu. What fusion is hinted at meaning is something better where gpu and cpu work more closely, but that's not the reality. Right now all fusion means is the AMD integrated graphics have been moved from the motherboard directly onto the cpu. It performs no differently, it just saves a few $ in manufacturing.

Intel have done exactly the same thing. With that in mind llano is likely to do as well as AMD did with mobile A64's and ati integrated graphics vs mobile core2's with intel integrated graphics.

 

[nonameo]

I believe that intel's marketing and general clout is enough to get them by AMD's IGP. They don't need to be better, they're more popular.

 

[Arkadrel]

You could do all the above without the gpu being part of the cpu? Infact the ones that do it best are nvidia and obviously their gpu's aren't part of the cpu. What fusion is hinted at meaning is something better where gpu and cpu work more closely, but that's not the reality. Right now all fusion means is the AMD integrated graphics have been moved from the motherboard directly onto the cpu. It performs no differently, it just saves a few $ in manufacturing.

Intel have done exactly the same thing. With that in mind llano is likely to do as well as AMD did with mobile A64's and ati integrated graphics vs mobile core2's with intel integrated graphics.

Yes but theres a bottleneck or two, that you avoid by putting the GPU on the CPU the way AMD did. That means this small little APU might have better gpgpu than a big card from nvidia.

We just need to see people start makeing software to make use of it.

Right now all fusion means is the AMD integrated graphics have been moved from the motherboard directly onto the cpu.

They removed some bottlenecks, the GPGPU power of their APUs will be decent for the size of the gpu they put in them. THAT is the reason why they did it in part.

It performs no differently, it just saves a few $ in manufacturing.

Thats part of the reason, and good enough in itself, cheap product with same/equal performance is a good reason for doing something. And while its true that in games, the gpu is just the same as a equal sized discrete gpu that you can buy sepperatly, its not true for GPGPU stuff.

AMD are besically gambling on GPGPU to take off, thats the reason behinde the APUs, to do that they had to buy ATI so they could make a fusion of CPU and GPU that could do more than just a CPU+GPU on 1 peice of silicone.

Intel have done exactly the same thing.

afaik they havnt, Intel has 2 sepperated parts on 1 silicone... they still have the same bottlenecks that a discrete card would have. Can Sandy Bridge even do OpenCL? DirectCompute? CUDA? APP? Because Im not sure it can., while the APUs that AMD makes are made for GPGPU power (thus they are very differnt, and not exactly the same).

 

[IntelUser2000]

afaik they havnt, Intel has 2 sepperated parts on 1 silicone... they still have the same bottlenecks that a discrete card would have. Can Sandy Bridge even do OpenCL? DirectCompute? CUDA? APP? Because Im not sure it can., while the APUs that AMD makes are made for GPGPU power (thus they are very differnt, and not exactly the same).

Not sure about all of them, but the GPU supports enough to accelerate Adobe CS4(partial DX11 feature just for GPGPU) using the Compute Shader function.

Regarding whether is better fit for integration, Llano, or Sandy Bridge, than dedicated graphics cards, I'd say they are about equal. They both share the router to pass communications between the CPU and GPU, and neither advances GPGPU programming beyond the level of what's available in discrete cards already. Both use shared memory model too.

On the hardware level, they are equal.

doesn't GDDR also sacrifics latency for extra thoughput?

Latency can be quantified in two ways. One is the absolute latency measured in uS or nS and other is relative one which is cycles.

Let's compare the two. The numbers are arbitrary.

DDR3 at 1600MT/s CAS4
GDDR3 at 3200MT/s CAS8

Cycles-wise, the latency is twice as great on the GDDR3. But on an absolute basis, its the same, because GDDR3 achieves twice the effective frequency. The latter achieves same latency with twice the bandwidth.

 

[Arkadrel]

Not sure about all of them, but the GPU supports enough to accelerate Adobe CS4(partial DX11 feature just for GPGPU).

Regarding whether is better fit for integration, Llano, or Sandy Bridge, than dedicated graphics cards, I'd say they are about equal. They both share the router to pass communications between the CPU and GPU, and neither advances GPGPU programming beyond the level of what's available in discrete cards already. Both use shared memory model too.

On the hardware level, they are equal.

Latency can be quantified in two ways. One is the absolute latency measured in uS or nS and other is relative one which is cycles.

Let's compare the two. The numbers are arbitrary.

DDR3 at 1600MT/s CAS4
GDDR3 at 3200MT/s CAS8

Cycles-wise, the latency is twice as great on the GDDR3. But on an absolute basis, its the same, because GDDR3 achieves twice the effective frequency. The latter achieves same latency with twice the bandwidth.

Source:
http://www.amd.com/us/press-releases/Pages/amd-fusion-apus-acc-mr-50-app-2011mar07.aspx?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+amdpressreleases+%28Press+Releases%29

Based on performance per watt comparisons between AMD Fusion APUs and the AMD Athlon™ II P320 CPU combined with the ATI Mobility Radeon™ HD 4250 GPU. In testing conducted by AMD performance labs, AMD Fusion APUs demonstrated the following: A-Series—up to approximately 500 GFLOPS; E-Series/C-Series—up to approximately 90 GFLOPS at 18/9 W. In comparison, the AMD Athlon™ II P320 CPU and ATI Mobility Radeon HD 4250 GPU deliver a combined total of 74 GLOPS at 38 W.

500 GFLOPs for Llano... (this is probably CPU+GPU, not just GPU alone, as those below are)

I might be wrong but I believe (just from quick googling):
5870 ~ 544 GFLOPS of Double-precision FP.
480 ~ 168 GFLOPS Double-precision FP.


So this small GPU thats part of the next APU the Llano, will have ALOT of GPGPU power.

Somehow I doubt Intels GPU has that amount of GPGPU level...

 

[IntelUser2000]

That doesn't make it better than a discrete GPU system with similar GFlops.

 

[Arkadrel]

That doesn't make it better than a discrete GPU system with similar GFlops.

thats true..

My point was just it has very good GPGPU abilities, more so than the small gpu in the APU might indicate, its a more effective way of getting good GPGPU abilties vs a discrete card (doing GPGPU stuff).

performance/$ will rise, in gpgpu. (im guessing the Llano will be cheaper than a 5870)
performance/watt will rise, in gpgpu.

That is why AMD put the gpu together with the cpu.
Their gambling that the future of computeing power, will come from these GPUs (in their APUs).

This seems differnt than why Intel put a IGP it on their cpus.

 

[podspi]

That doesn't make it better than a discrete GPU system with similar GFlops.

No, but it should be a heck of a lot cheaper! Interesting that it has so much more compute performance than graphic performance (everything I've heard so far puts Llano between a 5570 and 5650 in terms of gaming performance).

 

[itsmydamnation]

No, but it should be a heck of a lot cheaper! Interesting that it has so much more compute performance than graphic performance (everything I've heard so far puts Llano between a 5570 and 5650 in terms of gaming performance).

memory bandwidth, if AMD are really going to push fusion i would almost expect tri channel to be on a mainstream board on there next mainstream socket.

 

[piesquared]

intel can't compete with Llano. Developers are throwing all their weight behind fusion, and looking at sandy bridge and laughing. their graphics are a joke, and all the wreckless irresponsible intel emplyees running around forums (who?) bragging and sucking in naive consumers that their graphics will rule the roost have lost ALL credibility. it has the same render problems as every other attempt intel has made at making something that renders 3D. and reviewers like Anand lol Shimpi should be ashamed of themselves for not doing their job and informing consumers of just how poor they actually are. did anand even mention that intel can't support DX11, or did he at least make a token mention of it? you can bet reviewers make a huge deal out of any graphics glitch for NV and AMD. these reviewers are pathetic, is there no integrity or is it ALL about the pay off. do they think they are doing consumers a favor by such shameful ass kissing? dr who? was bloviating about how much intel relished competition, yet they sponsor crap reviews when their hardware can't stand on it's own 2 feet? what a joke. when the going get's tough, intel runs away and bribes reviewers and consumers alike. lol

 

[Concillian]

It wouldn't make sense if they price it too high, because then previous AMD processors + equivalent level graphics at a lower price point might make more sense.

The target market for Llano is one in which you literally cannot put in "equivalent level graphics" after the fact.

Small form factor
Notebooks
All-in-ones

This is where the world is moving.
2 kinds of people need desktops with full size slots that would even accommodate a video card:
- gamers for video cards
- scientists & engineers for workstations / GPGPU / specialty tool cards.

Llano is not targeted at those markets. It might "fill in" to the gamer / discrete video card market, but you can tell by how they designed it that it's not aimed there.

AMD knows it needs to raise ASPs. It will try. I'm not saying that I expect them to cost double what they do now, but I expect them to try for a little bit of margin increase since they have something that Intel can't match in certain aspects of performance (and conversely Intel has something AMD can't match in certain aspects of performance). The result is that both can coexist and demand price premiums because they aren't 100% direct competitors.

It may not work out this way in the end, and AMD may eventually lower prices if people don't catch on, but you can be damn sure AMD will try to drive some value out of the efforts they've put into Fusion. You can tell by the marketing.

 

[Nemesis 1]

intel can't compete with Llano. Developers are throwing all their weight behind fusion, and looking at sandy bridge and laughing. their graphics are a joke, and all the wreckless irresponsible intel emplyees running around forums (who?) bragging and sucking in naive consumers that their graphics will rule the roost have lost ALL credibility. it has the same render problems as every other attempt intel has made at making something that renders 3D. and reviewers like Anand lol Shimpi should be ashamed of themselves for not doing their job and informing consumers of just how poor they actually are. did anand even mention that intel can't support DX11, or did he at least make a token mention of it? you can bet reviewers make a huge deal out of any graphics glitch for NV and AMD. these reviewers are pathetic, is there no integrity or is it ALL about the pay off. do they think they are doing consumers a favor by such shameful ass kissing? dr who? was bloviating about how much intel relished competition, yet they sponsor crap reviews when their hardware can't stand on it's own 2 feet? what a joke. when the going get's tough, intel runs away and bribes reviewers and consumers alike. lol

Oh lord The ATI/NV wars have made it to the cpu/APU level . Man I gonna buy 2 llanos. NOT! I am sure the llano APU will battle the 2core SB in a good battle . But thats pretty lowend .

 

[ShadowVVL]

For quad desktops amd will probably price it around the same price as sb i3 2120
and mobile could range from 699 to 1k and up

So my guess on desktop is @ retail it might be 129-149

@ the high end
cpu $149
regular mobo $94
ram $120
psu 400w $39

Thats $402 before shipping and without hdd/ssd, which isnt bad.

But I dont see why they dont just put 3 memory channels in like 1366 then you could run 3x 1600.



I think llano fx would be fun.We could see apu crossfire,yay for Cake!!!

 

[Vette73]

Maybe this has been gone over but will they have a way to crossfire the on-chip video and a add on card?
I know the old AMD onboard had a hybrid system.

 

[Dribble]

Yes but theres a bottleneck or two, that you avoid by putting the GPU on the CPU the way AMD did.
....

And there are bottlenecks from moving the gpu onto the cpu - you are now heavily memory bound - where as before even the weakest integrated gpu's used to have their own separate memory which was generally faster then main memory, now they have to share the relatively low bandwidth main memory with the cpu.

I agree one day fusion might do more, but what AMD has released so far isn't it. Its a cost and power saving exercise, nothing more.

AMD knows it needs to raise ASPs. It will try. I'm not saying that I expect them to cost double what they do now, but I expect them to try for a little bit of margin increase since they have something that Intel can't match in certain aspects of performance (and conversely Intel has something AMD can't match in certain aspects of performance). The result is that both can coexist and demand price premiums because they aren't 100% direct competitors.

It may not work out this way in the end, and AMD may eventually lower prices if people don't catch on, but you can be damn sure AMD will try to drive some value out of the efforts they've put into Fusion. You can tell by the marketing.

They've got to - before fusion AMD made money from selling a gpu with every cpu, now they have effectively stopped selling the gpu's. Hence you need a larger margin on the cpu to stay at the rate you were at.