Rumour: Trinity 50% Faster than Llano

[Phynaz]

intel is at the border of being split.. and they should be allowed to buy their only competitor in relation with motherboard chips, a competitor for their integrated grahpics and a competitor with Atom?

Intel being split up - No.

Chipsets and integrated graphics - Nvidia is not in that business anymore.

ARM - Intel already has an ARM license, they don't need Nvidia for that.

Intel would not be acquiring a competitor if they were to buy Nvidia.

 

[tweakboy]

CPU with GPU in it is new technology and slower then a 6800GT Im guessing. That is why you disable onboard video and use a dedicated video card. Try the 460 1GB ,, such a good price. gl

 

[nonameo]

Personally, I think it would make more sense for them to buy up imagination tech, they're already working with them. Not to mention that they already have a solution for multi gpu / gpu switching/etc in lucid.

 

[formulav8]

Your first point has never been proven in a court.

That said, you ignored my second point. The market is moving to GPGPU - an area where Intel has no presence. How could a government agency tell (in the US at least) tell Intel they are not allowed to make a technology acquisition that could possibly mean their their existence in the next decade?

Intel doesn't pay 1.5 billion to their competitor to be nice. That action alone shows probable guilt.

NVidia still has a huge presence in their main market, gpus. It should be viewed in that aspect before the others. The government should view nvidias current markets and not possible future markets when deciding on the merger.

Anyways, I was only stating my opinion. Nothing more, nothing less. I just think its bad if Intel is allowed to buy nvidia.

 

[GaiaHunter]

CPU with GPU in it is new technology and slower then a 6800GT Im guessing. That is why you disable onboard video and use a dedicated video card. Try the 460 1GB ,, such a good price. gl

The faster Llano GPU is around X1900XTX so twice as fast as 6800GT.

 

[tweakboy]

Why someone would use the onboard video is beyond me.

Its way slow use dedicated GPU unless you dont have a video card or money for a dedicated video card. thx

 

[Soundmanred]

Why someone would use the onboard video is beyond me.

Its way slow use dedicated GPU unless you dont have a video card or money for a dedicated video card. thx

Such a small minded response.
Do you somehow think everybody that owns a PC plays graphically intensive games?

 

[Tuna-Fish]

Is the market going DDR4? It seems like video cards dabbled in it, but DDR5 was where the real improvements were. I could see a Llano type device (Llano or next gen APU) really benefiting from DDR5.

DDR4, GDDR4 and GDDR5 are not technologies, they are standards. The name DDR4 doesn't mean anything, it just standardizes a bunch of independent interface technologies. The GDDR and DDR standard with the same number are not necessarily in any way related. In fact, GDDR3 was based on DDR2, and GDDR4 was based on DDR3.

Up to GDDR4 the GDDR standards followed the DDR standards, simply re-optimizing them for larger transfer sizes and better latency tolerance, and adding a few operations that GPU's commonly use. GDDR5 was a break from this, and added a couple of novel technologies to double the data rate per pin.

Desktop DDR4 integrates a few of those improvements added first to GDDR5 -- so, to recap DDR4 has nothing to do with GDDR4, and is in the same class with, and borrows some tech from, GDDR5.

 

[DeathReborn]

Intel being split up - No.

Chipsets and integrated graphics - Nvidia is not in that business anymore.

ARM - Intel already has an ARM license, they don't need Nvidia for that.

Intel would not be acquiring a competitor if they were to buy Nvidia.

With Intel desperate to dominate the Phone & Tablet market nVidia would be considered a competitor then. Just because they don't compete in most segments does not mean they aren't competitors. Considering nVidia are investing in technologies capable of pushing into the Notebook, Server & Desktop markets (GPGPU, Project Denver as examples) they can probably be considered a competitor in some markets now, more markets in the future.

 

[podspi]

imho, I think both Intel and AMD are going to be forced to use some sort of on-die or on-package memory.

My personal bet is on-package through the IMC, at least for Trinity. For Intel, perhaps on-die on the ringbus?

 

[Tuna-Fish]

imho, I think both Intel and AMD are going to be forced to use some sort of on-die or on-package memory.

My personal bet is on-package through the IMC, at least for Trinity. For Intel, perhaps on-die on the ringbus?

The only memory Intel can put on-die for Ivy Bridge timeline is SRAM, and that is just plain too expensive to provide in the amounts that would add material speed advantages.

There were some rumors before about stacked memory on IVB: http://realworldtech.com/forums/index.cfm?action=detail&id=115339&threadid=115339&roomid=2

 

[podspi]

The only memory Intel can put on-die for Ivy Bridge timeline is SRAM, and that is just plain too expensive to provide in the amounts that would add material speed advantages.

There were some rumors before about stacked memory on IVB: http://realworldtech.com/forums/index.cfm?action=detail&id=115339&threadid=115339&roomid=2

Stacking is what I meant, I think I may be confused. Isn't stacking on-die? Or is it just a fancy way to produce a MCM?

Either way, it looks to me like Llano is bandwidth bottlenecked already. Not sure how they're going to cost-effectively improve performance by next year, unless they are just planning/hoping faster RAM becomes cheaper...

 

[Tuna-Fish]

Stacking is what I meant, I think I may be confused. Isn't stacking on-die? Or is it just a fancy way to produce a MCM?

It's kind of neither. Die is the thing that got cut out of the wafer -- the memory chips are on their own die. It's just soldered into the main processor die. It differs from MCM by not having data paths that leave the silicon -- this allows much faster, wider and cheaper interconnects.

To me, on-die means "in the masks used for making the processor".

 

[Zap]

Future CPUs are faster than current CPUs? Who would have thought?

 

[ShadowVVL]

What about moving the ram closer to the cpu or would that not change anything.
Is a 3 channel mobo with 2 ddr3 slots and 1 gddr5 slot possible?

 

[Idontcare]

Stacking is what I meant, I think I may be confused. Isn't stacking on-die? Or is it just a fancy way to produce a MCM?

Chip stacking, what people refer to as "3D packaging", is essentially a fancier way to do "MCM".

MCM is intentionally "side by side" though, an engineering benefit when it comes to thermal management and getting the heat out of the chips.

Stacking is more economical (cheaper) but the heat has to transfer through the whole stack to get out of the package.

NAND flash is typically stacked, sometimes as many as 8 die are stacked to create a single NAND Flash "chip" at the packaged level. But NAND flash dissipates so little heat that the stacking of the individual die does not represent a technological conundrum when it comes to extracting the heat.

 

[Nemesis 1]

Kinda funny to already see them talking about this when they just released Llano for Mainstream. For now, I just want to see Bulldozer being launched and reviewed since every day feels like a year.

Actually Axel its not . Hype is really all AMD has. You are aware I am sure that IB is in the wild being debugged. So Its not surprising at all . Intel has likely doubled the the performance of of IB igp over SB igp.+ than IB IGP is DX11 with open cl . Amd likely has a good idea of the IGP performance of IB IGP. So they have to talk about a IGP already on a BD core.Hype giving out performance figures. Intel has shown 3 IB running allready on 22nm tri-gate. server desk and mobile. They made know performance claims about the microarch. No real hype other than the 3D tri-gate. AMD on the other hand has given performance claims. Which if you think about it. Is strange as the BD core is stuggling to make it to market because it isn't to good right now. Hype Hype Hype

 

[Nemesis 1]

Maybe in their internal testing the igp really is 50% faster than Llano in some benchmark. They do have working silicon of the next gen apu.

Or do they, they are in a bad spot. Since PH I I can't and won't except anything AMD says . Intel pulled the same crap with Oak trail which is way late. The way BD is performing I find it hard to believe they would go threw all that work and than discover BD was a good idea on paper if you have the experts to pull it off . But still maybe, just maybe.

 

[ydnas7]

Intel doesn't pay 1.5 billion to competitor to be nice. That action alone shows probable guilt.

No that doesn't show probable guilt, it shows that Nvidia out demonstrated a stronger hand in the settlement.

Primarily, companies really don't want the risk involed in court cases, Nvidia appearing more prone to Mutually Assured Destruction, had the stronger hand in negotiating.

I've been involved in the inside on settlements, just because a party gets money does not demonstrate justise or fairness, it only demonstrates a stronger position to negotiate from.

 

[Tuna-Fish]

What about moving the ram closer to the cpu or would that not change anything.
Is a 3 channel mobo with 2 ddr3 slots and 1 gddr5 slot possible?

The GDDR5 interface is designed for the kind of low tolerances and wire capacitance you can achieve when you solder everything to a board -- I doubt that it would like going trough a socket (or even two sockets, if the ram was pluggable too) very much.

The very least, you'd lose enough speed that it would make using GDDR5 as opposed to more DDR3 questionable.

If there is going to be GDDR5, it would make most sense to integrate it into the CPU package.

 

[Tuna-Fish]

Stacking is more economical (cheaper).

Specifically, it's much cheaper to have wide interfaces between the chips. If all you needed was a single-bit interface between them, MCM would probably be the cheaper alternative. But stacking allows you to make extremely wide interfaces between them at little additional cost. This allows the individual flash chips to clock slower and still maintain performance, and here it would allow you to use relatively slow memory with a ridiculously wide interface to gain ram bandwidth.

 

[podspi]

I wonder how much memory bandwidth would actually be needed? I'm sure the IMC could stripe between on-board/system memory to gain a lot of performance.

 

[Riek]

Actually Axel its not . Hype is really all AMD has. You are aware I am sure that IB is in the wild being debugged. So Its not surprising at all . Intel has likely doubled the the performance of of IB igp over SB igp.+ than IB IGP is DX11 with open cl . Amd likely has a good idea of the IGP performance of IB IGP. So they have to talk about a IGP already on a BD core.Hype giving out performance figures. Intel has shown 3 IB running allready on 22nm tri-gate. server desk and mobile. They made know performance claims about the microarch. No real hype other than the 3D tri-gate. AMD on the other hand has given performance claims. Which if you think about it. Is strange as the BD core is stuggling to make it to market because it isn't to good right now. Hype Hype Hype

great if ivy would double the gpu performance. Would at least bring them almost on par with AMD desktop offering that will be released soon. However if i remember correctly the first estimate was 30% performance increase for ivy.

 

[Blitzvogel]

What about moving the ram closer to the cpu or would that not change anything.
Is a 3 channel mobo with 2 ddr3 slots and 1 gddr5 slot possible?

I think Trinity will have triple memory channels. If AMD expects to expand the APU graphics performance, they will need to expand the memory bandwidth accordingly and increase DDR3 speed support. Needless to say, Llano OCing will be popular I'm sure, as the laptop version should've had DDR3-1866 support and the desktop DDR3-2133 support. And yes, laptop DDR3-1866 exists. Kingston has it available in 2 x 2 GB and 2 x 4 GB stick packages.

 

[ShadowVVL]

I think Trinity will have triple memory channels

I hope they do, I was expecting quad channel with llano but nope,
But maybe it had somthing to do with costs and the controller.