Lincroft and Pineview die(Take a look at the size of the GPU)

[DavidC]

These are the dieshots of the next generation Atom CPUs. Pineview is the version for Nettops/Netbooks and Lincroft is for Moorestown, the MID/Smartphone version.

Pineview uses the modified GMA 950 IGP named GMA 3150 and Lincroft uses modified GMA 500 IGP aka PowerVR variant. The CPU core is the outlined part in the Lincroft die pic. You guys can probably figure out what that is for Pineview.

They both use Intel's 45nm process.

Pineview:

Lincroft:


Notice how much bigger is the rest of the die compared to the CPU core. What I'm trying to point out is overall, they seem to have similar die size despite vastly different GPU, power, and target market segment.

And more importantly, while the power consumption advantage is definitely on Lincroft's IGP, there's no real die size advantage to be had using the GMA 500/PowerVR. Considering GMA 950 is in average 2x faster, it doesn't look like the Intel IP IGP is inefficient in terms of die size, which seemed to be in contradiction to most at Beyond3D.

 

[Rys]

If you look at Intel's Hotchips '09 presentation for Lincroft, and that graphics block marking is accurate, the graphics block is the hottest part of the chip when the chip is fully clocked and powered.

 

[Psycho]

But isn't GMA500 tile based (I could imagine the big lower block being the tilebuffer), and thus requiring less system memory, lowering the memory chip die size
And then there's probably some difference in what kind of IO Pineview and Lincroft supports.

 

[mczak]

Notice how much bigger is the rest of the die compared to the CPU core. What I'm trying to point out is overall, they seem to have similar die size despite vastly different GPU, power, and target market segment.

Looks to me like Lincroft is indeed smaller (similar width but less height). Hard to tell though the difference might not be that big.

And more importantly, while the power consumption advantage is definitely on Lincroft's IGP, there's no real die size advantage to be had using the GMA 500/PowerVR. Considering GMA 950 is in average 2x faster, it doesn't look like the Intel IP IGP is inefficient in terms of die size, which seemed to be in contradiction to most at Beyond3D.

Don't forget Lincroft also has some block somewhere for video decoding, unfortunately I can't read the labels on that pic but I'd guess it's one of the bigger ones.
I don't know how all that translates to efficiency per die size, but Lincroft should probably have much lower TDP so only looking at die size and performance of the graphic core without considering power seems a bit unfair. Not to mention that old intel igp is quite limited in functionality.
I think though I'd prefer a (downsized) Ironlake IGP to either of them .

 

[CarstenS]

Psycho's right on the money, AFAIK. Lincroft is designed to be a system on a chip, i.e. not requiring external I/O, whereas Pineview requires Intel's NM10 I/O-part externally.

mczak,
I might mistake this for Pinetrail, but wasn't Video decoding to be moved to an external solution with Lincroft? US15W had it integrated, but IIRC Intel wanted to target Lincroft more on the embedded market with custom functionality added only as necessary.

 

[tangey]

mczak,
I might mistake this for Pinetrail, but wasn't Video decoding to be moved to an external solution with Lincroft? US15W had it integrated, but IIRC Intel wanted to target Lincroft more on the embedded market with custom functionality added only as necessary.

Pinetrail doesn't have any video decode capability.

Lincroft has video decode. Also I totally expect it to have video encode capability, so that 3 i/p cores against 1 i/p core. in pinetrail. Lincroft also has most of its I/O on chip too. So perhaps intels solution isn't as space efficient as SGX after all ?

On the die-shot of lincroft, the large shaded area bottom left is labelled "Gfx" which must be the SGX core. Above that the area is labelled "disp2d" so one assumes thats 2d stuff. Above that the area appears to be labeled "Vdc" and the area above that again is labelled "ved", I'd speculate that those two areas are video decode and video encode, which if true would be highly likely to be both from IMG.

So to do a proper GPU size comparision you'd need to figure out the area on the pineview that is the graphics core, I'd conjecture that its the entire purple shaded area next to the cpu, but its just a guess.

 

[compres]

I see this review in Anadtech:

Asus 1005PE
http://anandtech.com/mobile/showdoc.aspx?i=3693

But can't find them in stock in Germany. So they are released int he US but not in Europe? I would probably buy one of these later since watching movies (where an ION would be better) is not what I want such as small device for.

Should I wait to see what they offer with their Lincroft instead? I only care about batter life.

Edit: and linux compatibility ofc .

 

[mboeller]

you can find here two pics of the "new" ATOM CPUs. Single Core and Dual Core. The pics show different layers for the single core and dual core. Also the pics are flipped compared to the pics posted above:

http://www.heise.de/ct/artikel/Atom-1-5-890889.html?view=zoom;zoom=6

 

[mczak]

Psycho's right on the money, AFAIK. Lincroft is designed to be a system on a chip, i.e. not requiring external I/O, whereas Pineview requires Intel's NM10 I/O-part externally.

It's interesting to note imho that those io blocks don't really seem to require much area (for some odd reason pinetrail seems to have a similar amount of io blocks, but maybe I'm misinterpreting things on that die shot).

mczak,
I might mistake this for Pinetrail, but wasn't Video decoding to be moved to an external solution with Lincroft? US15W had it integrated, but IIRC Intel wanted to target Lincroft more on the embedded market with custom functionality added only as necessary.

I don't know for sure, but I can't see why it would have make sense to integrate this into us15w (which was probably more of a tech demo sort of product) and not lincroft.
OTOH it would seem strange if it had separate video encode / decode parts but I've no idea really. One thing's for sure though, if sgx core is only that block on the bottom it's only about one third the size of the gen-3 intel gma core.

 

[Groo The Wanderer]

Psycho's right on the money, AFAIK. Lincroft is designed to be a system on a chip, i.e. not requiring external I/O, whereas Pineview requires Intel's NM10 I/O-part externally.

mczak,
I might mistake this for Pinetrail, but wasn't Video decoding to be moved to an external solution with Lincroft? US15W had it integrated, but IIRC Intel wanted to target Lincroft more on the embedded market with custom functionality added only as necessary.

It is an optional external Broadcom chip last I heard in both cases.

-Charlie

 

[DavidC]

you can find here two pics of the "new" ATOM CPUs. Single Core and Dual Core. The pics show different layers for the single core and dual core. Also the pics are flipped compared to the pics posted above:

http://www.heise.de/ct/artikel/Atom-1-5-890889.html?view=zoom;zoom=6

I flipped it so the comparison is easier, but now I realize the original is better to compare(look at the shape of the L1 cache blocks on the CPU). You can also see there are 3 blocks on the Pineview GPU too.

It's not like everything is 3D on the Intel IP, it can do MPEG decode on hardware. Plus, the GMA is still faster.

And guys, look at the Intel's IDF Fall 2009 presentation for Moorestown. It has hardware video decode on Lincroft. It's not going to require the Broadcom solution, theoretically.

Psycho's right on the money, AFAIK. Lincroft is designed to be a system on a chip, i.e. not requiring external I/O, whereas Pineview requires Intel's NM10 I/O-part externally.

I'm not sure exactly what you meant, but Lincroft pairs with TSMC 65nm crafted Langwell for the I/O chip that is similar in function to NM10.

The die size is 65.9mm2 for Pineview. I've heard that Lincroft is around 70mm2, but without exact numbers, its hard to say what's bigger.

 

[mczak]

It's not like everything is 3D on the Intel IP, it can do MPEG decode on hardware.

Yes, but I'm pretty sure the complexity of this is _much_ lower, it's probably hardly worth mentioning the couple transistors it needs. You're right though certainly there's more there than 3d (for instance display engine which should be very similar between the two chips) but I think all the big blocks on pinetrail belong to 3d (could be wrong though but what else could it be?)

Plus, the GMA is still faster.

Right. I thought though some Lincroft chips will have twice the gpu frequency (400Mhz)? Given the different nature of lincroft/pinetrail gpu though I'd suspect the performance difference to vary wildly between these two...

And guys, look at the Intel's IDF Fall 2009 presentation for Moorestown. It has hardware video decode on Lincroft. It's not going to require the Broadcom solution, theoretically.

In fact it also had video encode. I wonder what it can handle then.

I'm not sure exactly what you meant, but Lincroft pairs with TSMC 65nm crafted Langwell for the I/O chip that is similar in function to NM10.

Oops forgot about that too. Is this using even the same dmi interface?

The die size is 65.9mm2 for Pineview. I've heard that Lincroft is around 70mm2, but without exact numbers, its hard to say what's bigger.

Hmm Lincroft looked smaller to me but maybe it's not after all...

 

[DavidC]

I've seen 400MHz being thrown around for Lincroft's IGP. Now I looked a bit into it, I don't think its Lincroft.

Page 14

There's yet another core: http://download.intel.com/pressroom/kits/events/idffall_2009/pdfs/2009_IDF_Kim.pdf

Sodaville. That's the 400MHz core, with 7-9W TDP, using SGX535 for TV set-top boxes. It can output dual 1080P streams.

 

[Arun]

Uhm... guys. That Lincroft die shot, if you look closely enough at the text, has blocks labeled "VXD", "VXE" and "GFX". Kinda obvious what these are, isn't it? Let's try estimating die size of each block based on the assumption that the Atom block is the same size as on their previous chip, despite the process being slightly different (new SoC variant...)

Atom = 91.9K Pixels = ~13.7mm²
VXD = 32.2K Pixels = ~4.8mm²
VXE = 25.1K Pixels = ~3.7mm²
GFX = 57.6K Pixels = ~8.6mm²

Given that encode is normally more expensive than decode, I can only assume that VXE is the VXE250. If that is correct and I'm reading all the blocks right (big if), the die size of VXD is very nice but VXE and SGX(535) really much bigger than I'd have assumed. I do wonder if there might be more than SGX535 included in that GFX block though; I seem to read another block as 'Display' but maybe not. Who knows. Heck, all my blocks could be wrong! (doubt that though, hehe)

 

[MfA]

I assume encoding is done cooperatively using the USSE2 cores (with VXE handling the bitstream stuff).

 

[Arun]

I assume encoding is done cooperatively using the USSE2 cores (with VXE handling the bitstream stuff).

Since VXE is a PowerVR IP and it's made for handhelds with minimum host offloading to save power, that would require Imaginations/PowerVR to have made a custom core for Intel - not impossible, but rather unlikely. Of course, I might also be misreading the block's label (i.e. it's not VXE) or Intel might also call their own in-house encode block VXE (ugh!) - who knows...

 

[tangey]

Uhm... guys. That Lincroft die shot, if you look closely enough at the text, has blocks labeled "VXD", "VXE" and "GFX". Kinda obvious what these are, isn't it?)

I'm sure I just said all that a few posts ago !

I seem to read another block as 'Display'

I'm definitely sure I said that, in fact it looks to me like "disp2d" i.e. 2D acceleration.

 

[mczak]

I'm definitely sure I said that, in fact it looks to me like "disp2d" i.e. 2D acceleration.

Makes sense though for an old style blitter it seems to be quite big.

 

[Lazy8s]

SGX540 would presumably occupy less than half the size of that graphics block in the same process.

The increase in functionality from OpenGL ES 2.0 to DirectX 9+ is sometimes easy to underestimate, yet extra stuff definitely seems to be bloating Lincroft's graphics block beyond the SGX535.

 

[Ailuros]

SGX540 would presumably occupy less than half the size of that graphics block in the same process.

The increase in functionality from OpenGL ES 2.0 to DirectX 9+ is sometimes easy to underestimate, yet extra stuff definitely seems to be bloating Lincroft's graphics block beyond the SGX535.

It's my understanding that ~SM3.0 is the base level for all cores up to SGX540, with differences mostly in drivers supporting only specific cores. The only exception is SGX545 which is SM4.1/OGL3.0 and has a full embedded OpenCL profile.