AMD VP gives ballpark Llano gpu performance [Fudzilla]

[Vesku]

Seems GPGPU is critical to their plans to compete with Llano though. So far it seems that nVidia has been doing just as much with OpenCL support as AMD, not a good position to be in.

 

[Nemesis 1]

[yyyyyyyyyy

 

[GaiaHunter]

So I assume ya seen the demo of aliens. Did ya try to watch it on full screen . It was a very small monitor.

Wasn't that running on a Ontario or Zecate?

Or did I miss AvP on Llano?

 

[Nemesis 1]

[uuuuuuuuuuuu

 

[busydude]

Or did I miss AvP on Llano?

http://www.youtube.com/watch?v=uwhuNcx4BTY

http://www.legitreviews.com/article/1443/1/

 

[Nemesis 1]

[hhhhhhhhhhhhhhhh

 

[Idontcare]

Zacata. Thats not the point resolution is important but screen size is also . Anyone who believes zacata is going into tablets is foolish . 8 watts for the chip is way overboard.
Oaktrail holds a few surprizes for netbooks/notebooks and as far as gaming on tablets Intel oaktrail will own this market.

You got samples of zapata or Ottawa in house yet? What about liano?

 

[extra]

Highly likely Llano will own SB in the GPU performance and SB will own Llano in the CPU performance area, and whichever one makes more sense depends on what you're wanting to do with your computer. Intel supporting opencl will be a great thing for everyone (including AMD). It'll be nice as well because it'll ensure Intel has great reason to keep improving their integrated gpu.

 

[Vesku]

Yes, I think it's up to AMD to really push OpenCL and will be a key to how successful Llano is. Haven't heard anyone optimistic on how much GPU software AMD will be pushing on Llano launch though. I'm open to hearing from someone who thinks otherwise.

 

[Hard Ball]

You could be right . But let me say . When conroe was released early benchmarks. It fall into the exact . I mean exact performance figures as comparred to AMD 64 as my wife gave to this forum months befor those benchmarks . That statement threw replies was turned into a ban for Intelia my wife. Its easy when brother inlaw works for intel . Yet even after the benchmarks people said they were skewed benchies. You know that and I know that . Than add in the fact which team did the SB (Israel). I think SB igp is going to knock many people for a loop. Thats exactly what were being told also . I really want the 2600k retail . Overclock it to 5(H2O) ghz which also speeds up the last level cache to 5ghz which helps the IGP. lower latency and better rendering. I also want to O/C the IGP to 1.5 GHz min. Secretly hoping for 1.6 to 1.7 ghz.

But just stock 2600K I believe will be better than fusion. Why because intel did a better job of fusing the 2 processors into 1 package. That last level cache could be the game changer. only 2 months befor we see the O/Cing reviews . I think many are going to be surprised. Intel has NO discret card . Intel can't shot self in foot. AMD has discret and can shot self in foot.

Also you guys keep complaining about consol games ruining PC gaming . Well this fusion type processors are going to do its share of damage . I make games who do I target . If intels SB is good enough that is my target . period its about $$$$$ at the end of the day.

I already have a very good idea of how both will perform, but it is not really my place to make any comment in that regard.

I will however say, about the highlighted part of your proposition. Last level cache shared between graphics and CPU, in most computer microarchitects' opinion, could turn out to be one of the most performance degrading designs elements for SOCs of the future.

Many people in the community regard it as a mistake for the current set of programming models and graphics APIs. We need to keep in mind, things that might bring some advantage to explicit, ISA/API controlled caching behavior, are not prevalent at all in graphics middleware. Some benefits for OpenCL based compute, yes; but the currently predominent SPMD programming model for graphics shaders, makes such a design at best irrelevant, at worst a detriment for graphics workload. Plus when not exactly optimally implemented, it could impact performance of the general purpose cores as well.

 

[Arkaign]

I know it's all in the name of progress, but CPU-resident GPU is about the least exciting thing I've seen in tech in a while. At least it will make notebook/SFF systems more efficient and easier to get into smaller and smaller packages, and will pave the way towards better tablets/etc.

 

[GaiaHunter]

http://www.youtube.com/watch?v=uwhuNcx4BTY

http://www.legitreviews.com/article/1443/1/

I had seen the first one but not the second one - rather short AvP demo.

To nemesis1:

Zacate has a 80 sp part. Llano is rumoured to have somewhere between 240 to 480 sps - that is quite a difference.

The bandwidth problems seems to be the big question, but it seems that the amount of cache on the GPU die can play an important role.

 

[Scali]

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

A fixed pipeline is not the same as a non-programmable pipeline.
You always need to have a lot of programmability in order to support the unified shaders that Direct3D and OpenGL require.
Thing with Intel was that they used their programmable shaders even for things that most other GPUs do with fixed-function units.
Sandy Bridge has more of these fixed-function units, to improve performance.

But obviously it still has programmable unified shaders. As long as they didn't mess up the design of these shaders too much, they should be able to support DirectCompute and OpenCL on them quite easily (they both have the original GeForce 8800 as minimum requirement. DX10-level hardware with a few special things for GPGPU).

 

[Idontcare]

No I do not. Bob has seen in action I haven't . Bob says for first silly its good.

Bob did give me an AMD product tho . But thats for another day.

cool, thx as always for sharing what you can.

SB looks to have a healthy 6 month head-start over liona, but it may be even longer if folks aren't getting their liana samples yet.

PS - I wonder how many people actually got what you meant by "Bob says for first silly its good." But is it really first silly (A0 step) or is it just first silly they let their partners sample?

 

[GaiaHunter]

cool, thx as always for sharing what you can.

SB looks to have a healthy 6 month head-start over liona, but it may be even longer if folks aren't getting their liana samples yet.

PS - I wonder how many people actually got what you meant by "Bob says for first silly its good." But is it really first silly (A0 step) or is it just first silly they let their partners sample?

What is wrong with your typing IDC - liona, zapata, liana, liano? Been drinking some?

 

[Scali]

Heh, IDC has made the link between Lllano and Top Gear's Reasonably Priced Car (Suzuki Liana).
Perhaps Llano is a Reasonbly Priced APU?
Funny enough Liana actually stands for Life In A New Age.

 

[Idontcare]

What is wrong with your typing IDC - liona, zapata, liana, liano? Been drinking some?

It's intentional for two reasons - first is "situational awareness" and knowing your target audience. When in Rome... My conversation there was with a specific individual who has a certain panache for garbling codenames, and I reciprocate out of respect for his approach. Emulation being a form of flattery and all that.

Second reason is that anybody who is under NDA or confidentiality agreements can't speak about Zacate, etc,...but I'm not asking them to speak about Zacate nor are they speaking about Zacate...we are talking about Zapata, whatever the hell that is

Oh, and yes, yes I am, I'm awake aren't I

 

[Dresdenboy]

Well its going to have to differentiate itself on tessellation than . Intel has Opencl on SB.
Won't that be a kicker. The AMD fanbois are fighting it out with tessellation on GPUs as we speak . NV fanbois saying its the Holy grail the AMD/ATI guys saying not .
Ya can't make this stuff up . Wait till the reviews . AMD fanbois will hold tessallation up as the holy grail of IGPs. But looking at the h2 release time and what Intel said at IDF . It will be a short lived victory dance.

Thomas Piazza, an Intel fellow and director of graphics architecture for the Intel Architecture Group, said that Sandy Bridge-based chips in their current implementation will not support DirectX 11, a Microsoft technology for accelerating multimedia and games. Currently, Sandy Bridge supports DirectX 10.1 and OpenCL 1.1--the latter used on Apple's Mac operating systems, according to Piazza.

http://news.cnet.com/8301-13924_3-20016302-64.html

Does Piazza by "Sandy Bridge" mean the processor as a whole or it's GPU part? Hexus talks about SB's graphics, but as someone else here already wrote, how could SB do OpenCL computing with fixed function units? This is currently not clear for me.

OTOH SB could do OpenCL computing regardless of it's GPU's capabilities:
http://www.khronos.org/developers/l...of_opencl/OpenCL-BOF-Intel-SIGGRAPH-Jul10.pdf
http://blog.jackpien.com/?p=228

 

[Scali]

Hexus talks about SB's graphics, but as someone else here already wrote, how could SB do OpenCL computing with fixed function units? This is currently not clear for me.

Read my post, I already answered that... And can I be honest here for a second?
I mean, no offense, but why are you wondering about such issues in the first place? The last generation of hardware without programmable shaders was the DX7 era.
We haven't had 'fixedfunction' GPUs in ages. They've been programmable since DX8. We have some fixedfunction *units* in these GPUs, but that's different. They are only for graphics-related duties (triangle setup, texture filtering, clipping, etc), and have nothing to do with OpenCL or any other GPGPU framework.
But if you apparently know so little about how shaders and programmable hardware on a GPU work, why are you trying to think about how OpenCL may or may not work on certain hardware?

 

[Martimus]

Since Scali didn't actually elaborate on what post he was talking about, nor is the post visible, I have found it for you if needed: http://forums.anandtech.com/showpost.php?p=30671824&postcount=50

Also, I haven't seen you post on Anandtech before Desdenboy. I would love to see this become more of a regular thing since I enjoy hearing about the possible features of new architectures.

 

[Scali]

Since Scali didn't actually elaborate on what post he was talking about, nor is the post visible, I have found it for you if needed: http://forums.anandtech.com/showpost.php?p=30671824&postcount=50

I think you have an off-by-one error:
http://forums.anandtech.com/showpost.php?p=30671827&postcount=50

 

[jvroig]

Does Piazza by "Sandy Bridge" mean the processor as a whole or it's GPU part? Hexus talks about SB's graphics, but as someone else here already wrote, how could SB do OpenCL computing with fixed function units? This is currently not clear for me.

GPU only, of course.
The problem is that "fixed-function" (fixed-function units) is an obsolete term, but lay people (non-developers) always think of it when "fixed pipeline" is mentioned. Worse, I have actually also seen some presenters slip and say fixed-function units (old habit, probably), when in fact everything (since 2006 for Intel, and you know they're late to that party) is already composed of programmable units (call them shaders, EUs, whatever).

Intel had this published in a white paper years ago, in what I believe was the first time they coined the term EU (execution units) to accompany their "Generalized Programmable" architecture. "EU" terminology supposedly replaced the "fixed-function units", but people never picked it up, it seems (of course, can't blame anyone, who gets excited with Intel graphics after all?). Today, though, I am sure more people should be familiar with "EU"s, because they have been highlighted with the SB previews.

So unless they do something really stupid (and they won't, because why would they even think of, or how could they even screw up, a 5 year old tech they developed), SB's GPU will be no different (they even still use the term EU's, right?)

I'll try to find a link or two, shouldn't be hard.

EDIT: Not just a link, but the old white paper itself, google is awesome: www.intel.com/assets/pdf/whitepaper/313343.pdf Welcome to 2006, everything is programmable now We can rest assured stuff wouldn't regress so bad that 2011 tech is suddenly not programmable.

 

[Scali]

www.intel.com/assets/pdf/whitepaper/313343.pdf

Forum seems to have messed up the link, fixed it for you.

 

[jvroig]

Forum seems to have messed up the link, fixed it for you.

Whoops, you are right, thanks.

 

[slayernine]

Well its going to have to differentiate itself on tessellation than . Intel has Opencl on SB.
Won't that be a kicker. The AMD fanbois are fighting it out with tessellation on GPUs as we speak . NV fanbois saying its the Holy grail the AMD/ATI guys saying not .
Ya can't make this stuff up . Wait till the reviews . AMD fanbois will hold tessallation up as the holy grail of IGPs. But looking at the h2 release time and what Intel said at IDF . It will be a short lived victory dance.

Well the new 6000 series cards do really well at tessellation compared to the 5000 series. Also the leaked benchmark of the 6970 shows it getting 36fps versus the 480 at 30fps on the heaven tessellation benchmark.

So I think AMD "fanbois" will be bragging up tessellation for both integrated and dedicated GPUs.