AMD VP gives ballpark Llano gpu performance [Fudzilla]

[Vesku]

http://www.fudzilla.com/notebooks/item/20614-amd-vp-hints-at-llano-graphics-performance

Initial thoughts: The suggested Llano performance *should be better than* Anandtech's, probably top end IGP, Sandybridge preview. We already know that with the Llano CPU cores they are going with "good enough".

Big question in my mind is whether AMD is going to be able to dedicate the software and driver resources to make Llano a compelling buy. Anybody know of AMD gpu software integration I have missed out on?

*Edit: Looked at the Sandybridge preview and it was actually Radeon 5450 now*

 

[Nemesis 1]

oooooooooooo

 

[Vesku]

Do you think Intel will push OpenCL or will AMD have to pony up with a few apps before Intel feels the need?

 

[cbn]

http://www.fudzilla.com/notebooks/item/20614-amd-vp-hints-at-llano-graphics-performance

Initial thoughts: The suggested Llano performance seems similar to Anandtech's, probably top end IGP, Sandybridge preview. Meaning it will have to differentiate itself on DX11, OpenCL and other feature enhancements not to mention pricing. We already know that with the Llano CPU cores they are going with "good enough".

Why do you think the Sandy Bridge GPU will be similar to Redwood? According to the Anandtech SB preview Intel's top end iGPU was more in line with Cedar.

 

[cbn]

Do you think Intel will push OpenCL or will AMD have to pony up with a few apps before Intel feels the need?

Hopefully someone more knowledge will chime in, but I believe Sandy Bridge iGPU is fixed pipeline and therefore not programmable.

AMD is developing Open CL through its fusion fund.

 

[Vesku]

I was going by Nemesis saying Intel will be supporting OpenCL on their IGP. Went back to the Sandy Bridge preview and it does seem to be 5450 and not 5570 performance. Still pretty amazed at how much performance gap they closed in just two IGP revisions, maybe they can squeeze some more clocks out of it before release.

Core question still remains, will AMD have apps and frameworks available to push Llano properly? They've had good products languish before due to lack of follow through.

 

[cbn]

I was going by Nemesis saying Intel will be supporting OpenCL on their IGP.

Nemesis is right. For some reason I missed that comment in his post.

http://www.hexus.net/content/item.php?item=26452

Sandy Bridge's graphics are DX10.1, OpenGL 2.1 and Open CL 1.1 compliant.

P.S. I should really keep my mouth closed instead throwing around terms like "fixed pipeline" and "programmable". (I'm not an IT guy). My apologies for the bad information.

 

[cbn]

Well first off. SB that AT benched was a very early sample that had no Turbo operating . That would likely mean that the IGP was running at 850Mgz.The highend IGP runs at 1.35Ghz. Thats alot of performance we haven't seen . Than the drivers were very very early. So none including AT could print anything other than what they experianced. Than add in the fact that intel final silly of SB 1155 is now in production . So what reality is is still an unknown . My sample is gimped also, so I don't know . But based on what Bob has given me lately I would say we haven't seen anywere near whats coming in early jan.

Thanks for the information. It will be really interesting to see how Intel pushes into this direction. According to Wikipedia they will even include DX11 on the 22nm Die shrink.

 

[alexruiz]

Am I the only one reading it differently?

A 500 Gflops, it has > 4x times the processing power of a 5450, and game performance is at least 3x the 5450.... So why are you saying that SB and Llano will be even in graphics power? Yes, highend SB will top a 5450.... but overcome the 3x differential between a 5450 and a 5570?

 

[StrangerGuy]

Am I the only one reading it differently?

A 500 Gflops, it has > 4x times the processing power of a 5450, and game performance is at least 3x the 5450.... So why are you saying that SB and Llano will be even in graphics power? Yes, highend SB will top a 5450.... but overcome the 3x differential between a 5450 and a 5570?

SB IGP performance = 400 SP Radeon? Don't make me laugh. I won't dispute SB CPU will own Llano's CPU pretty hard but everything points the IGP = 5470m (80 SP) tops. 2-3x over current Arrandale IGP isn't hard to do when it sucks in the first place...current i3 GPUs can't even play a CS: S high settings at constant 60 fps @ 1366x768 and that's a six year old game.

 

[OCGuy]

^

I don't think you can just compare flops and know where a processor will stand compared to others.

 

[Vesku]

Giving a large numbers of PCs base performance of low tier dedicated cards, instead of an eighth or quarter of the performance, will most likely help PC gaming. AMD will really need to leverage the GPU on Llano though, hoping they provide some good libraries and support some sort of OpenCL push.

 

[Idontcare]

That statement threw replies was turned into a ban for Intelia my wife.

I wasn't active in the forums at the time, I just searched for Intelia posts and holy cow did she ever get unjustly lynched by the mob here at the AT CPU forum.

That's a black-eye on AT. Can't believe they caved to mob rule just because people didn't like the message. And she was right, of course she was because you guys had the hardware in-house already at the time.

 

[alexruiz]

SB IGP performance = 400 SP Radeon? Don't make me laugh. I won't dispute SB CPU will own Llano's CPU pretty hard but everything points the IGP = 5470m (80 SP) tops. 2-3x over current Arrandale IGP isn't hard to do when it sucks in the first place...current i3 GPUs can't even play a CS: S high settings at constant 60 fps @ 1366x768 and that's a six year old game.

That is what I meant, Llano will be a 400 SP Radeon, so we cannot expect an IGP form a manufacturer not exactly know for graphics expertise to match it

 

[Nemesis 1]

8888888888888

 

[formulav8]

I remember your 'wife'. And well is not part of the topic anyways....

Anyways, hopefully Llano will end up better than just simply beating SB. If they can't topple SB then its a big failure as far as AMD's IGP's are concerned. Decent IGP's was supposed to be one of the main reasons for buying out ATI IIRC....



Jason

 

[Vette73]

I remember your 'wife'. And well is not part of the topic anyways....

Anyways, hopefully Llano will end up better than just simply beating SB. If they can't topple SB then its a big failure as far as AMD's IGP's are concerned. Decent IGP's was supposed to be one of the main reasons for buying out ATI IIRC....



Jason

Llano is not supposed to replace a graphics card just be the next gen of IGP. Its also supposed to be made cheap and make AMD some money.
Why would AMD come out with a chip that makes oems not want/have to buy a seperate video card?

Intel can put out the best they have onchip as their best is IGP. AMD can;t afford that. If I could get 56xx level of performance out of IGP I would not have a seperate video card right now.

 

[Martimus]

This is totally new stuff . and intel has been working on it longer than AMD/ATI.
Its kinda like that statement AMD made above. About the memory controller.
Which I find very very dishonest on AMds part . They should have gone into comparring their controller to SBs ringbus with stop control and latency control . Instead of a ding dang empty statement they made.

I don't see the statement above about a memory controller. What is the statement?

Also, the Llano bus should be better than the Intel Ring bus for the specific circuit. It will have lower power, lower latency, plus is designed with specific bandwidth in mind for the circuit they are building. The advantages to the Ring-bus are far more for the ease of design and the flexibility it gives you in altering the design than it is for actual performance. It won't really affect the end user so much as it will allow Intel to make adjustments to the architecture quicker at the expense of power consumption and latency.

 

[formulav8]

NV is already screwed and died its just a delayed reaction.

That part is true. Theres a reason they are pushing non-gaming markets so much the past few years...

 

[GaiaHunter]

First AMD has good reasons to kill the low end market:

- NVIDIA competes there with them for every discrete low end card - if AMD can sell a compelling CPU+GPU package there is no competition between AMD and NVIDIA as no low end card is needed.

- An APU will be reviewed based on CPU and GPU power as opposed to today CPUs that are only reviewed on CPU power, so AMD can balance their CPU deficit with a strong GPU and create a market for Laptops and budget desktops for regular day to day applications + gaming on par with consoles.

About the memory controller - the key for AMD APU seems to be the cache on the GPU die.

 

[busydude]

Ideally AMDs first Fusion part should have equaled the 6870 . Than AMD pushes out CPU/GPU into the $400 dollar range and put a world of hurt on intel .

I hope you are being sarcastic.

If I could get 56xx level of performance out of IGP I would not have a seperate video card right now.

All the rumors point out that Llano is ~5570 performance wise.

IMO, to compete with SB AMD should not just rely on its IGP's performance, instead it should put more resources into the software aspect of technology at which it has failed many times.

 

[Martimus]

For martimus


Therefore, AMD believes its approach of linking the CPU and GPU on the Fusion processors through the memory controller, not cache, is more appropriate as it prevents the GPU from thrashing the CPU caches due to the very different access patterns between the two. AMD staff said that Intel most likely didn't know how to design a versatile memory controller that would handle both types of access efficiently, so it ended up using the L3 cache as a buffer of a sort.


Just to backup what I think VS. What AMD says about Intel not being able to do a memory controller . SO thoughtless it was . Whats Intel got on it present i5 cpu/gpu. A memory controller for both cpu/gpu in the Gpu part of chip . AMD spends way to much time in the dirt.

I don't know who in the company said that, but I agree it isn't very thoughtful to assume that Intel doesn't know how to design a versatile memory controller. I haven't read the quote, so I don't know who said it or really even if it was actually said, but I would like to know what the rationale was behind such a statement.

I am not sure what you mean about spending time in the dirt. We must be from different parts of the world, as that isn't an expression that I am familiar with.

 

[fastclock]

You don't want a huge on-die IGP for at least the following reasons:
1) High cost.
2) Low yield.
3) High power consumption, thus more heat, more noise, low battery life...
4) Shared memory bandwidth/performance limitation.

There's no good reason for AMD to go with anything larger than a 400SP IGP for 2011. Most users/gamers that want more GPU power will prefer dedicated/discrete GPU anyway.

 

[Vesku]

I hope you are being sarcastic.



All the rumors point out that Llano is ~5570 performance wise.

IMO, to compete with SB AMD should not just rely on its IGP's performance, instead it should put more resources into the software aspect of technology at which it has failed many times.

That's what I'm thinking too, will AMD commit the software resources necessary to give Llano legs?

 

[GaiaHunter]

That's what I'm thinking too, will AMD commit the software resources necessary to give Llano legs?

I'm not sure about the GPGPU side of it but you have to admit the MLAA seems quite a useful thing to improve gaming IQ at very low performance cost.