Intel Medfield in Nokia N9 unveiled in February MWC?

  • Thread starter Deleted member 13524
  • Start date
D

Deleted member 13524

Guest
It's still a rumour, but it's coming from a "respected" Finnish tech magazine (Finland -> Nokia):
http://www.prosessori.fi/uutiset/uutinen2.asp?id=57146
http://www.engadget.com/2011/01/20/nokia-n9-to-bust-loose-with-meego-on-moorestown-power/


Announcement in February should mean availability in 2H 2011.
It seems that during the summer/fall we'll have quite the hardware war for top-end smartphones.

What do we know about Medfield? Single-core, dual threaded? IGP? 64bit single-lane memory bus?

All I know is it'll be the first Atom on 32nm, fwiw.
 
I'm 99.999% certain that the N9 is based on a 1GHz OMAP3630 (and yes, that's not very exciting). However I can imagine Nokia pre-announcing a Medfield MeeGo phone for 2H11. They will need to pre-announce some products at MWC to keep people excited and that's certainly a strong contender.

Medfield uses a ~2GHz single-core/dual-thread and I heard the GPU is actually a fair bit faster than what you'd find in current-gen ARM application processors (so presumably a very high clocked SGX535, SGX540, or SGX545). I've got a tiny bit more speculative analysis on it in my-article-which-might-be-published-next-week-if-I-can-get-my-hands-on-Rys (TM). Although I might release it in two parts in which case that part would come later, heh.
 
If Moorestown was sacrificed to focus on Medfield, that's OK.

Intel could be pushing a 400 MHz SGX535 and then on to a 554MP in the next revision, or they might have switched to something inbetween considering they've deliberately not disclosed the identity of the GPU yet. The whole API support difference among 535, 540, and 545 makes the answer less obvious than it should be.
 
I'm 99.999% certain that the N9 is based on a 1GHz OMAP3630 (and yes, that's not very exciting). However I can imagine Nokia pre-announcing a Medfield MeeGo phone for 2H11. They will need to pre-announce some products at MWC to keep people excited and that's certainly a strong contender.

Should we then take it that Moorestown is dead, its very late, and to me it looks like Intel re-focused on Oaktrail as they saw that that was an area they could get design wins

Medfield uses a ~2GHz single-core/dual-thread and I heard the GPU is actually a fair bit faster than what you'd find in current-gen ARM application processors (so presumably a very high clocked SGX535, SGX540, or SGX545). I've got a tiny bit more speculative analysis on it in my-article-which-might-be-published-next-week-if-I-can-get-my-hands-on-Rys (TM). Although I might release it in two parts in which case that part would come later, heh.


There was both a pdf at one of the IDF's and an unofficial leak late last year, that confirmed that Medfield would have x2 the graphics performance of moorestown, and moorestown has/had options of 400MHZ SGX535.

To me that means medfield will either be 400MHz SGX545, or 543MP2@200Mhz. Intel's prior policy was to use a single core across multiple Socs, so SGX545 would allow the I/P block to be used in other Socs where DX compliance was important. I suppose it depends whether the DX10.1 and full profile OpenCL compliant parts of the SGX545, which would be redundant in a Meego phone, are significant power draws.
 
If Moorestown was sacrificed to focus on Medfield, that's OK.

Intel could be pushing a 400 MHz SGX535 and then on to a 554MP in the next revision, or they might have switched to something inbetween considering they've deliberately not disclosed the identity of the GPU yet.

400MHz SGX535 was already the target for Moorestown.
I heard something like 4x current IGP performance (200MHz SGX535), which could mean something like a 600MHz SGX540.
Or something completely different, like getting the 6EU IGP from sandybridge (that would save them a lot of money from IP (not that Intel needs money right now)).
 
I find it extremely odd that after months and months of waiting without a Moorestown release a customer would suddenly not only demonstrate a Medfield product but release it in a scant few months. Medfield is something Intel hasn't said an awful lot about.. surely if they were this far along they'd be the first to demo and present it.

There's little chance that the 6EU variant Sandy Bridge IGP will be low enough power to work in an Atom Z-series SoC, much less be competitive in power draw with SGX. I don't have power numbers or any real great metric for them, but we know it performs pretty similarly to the GPU on Zacate which appears to take at least a few W in even its smaller incarnation. And I certainly won't so easily believe that Intel is beating AMD in GPU perf/W, although they do have a process advantage here. But it's still a design made for desktop/laptop CPUs.
 
To me that means medfield will either be 400MHz SGX545, or 543MP2@200Mhz. Intel's prior policy was to use a single core across multiple Socs, so SGX545 would allow the I/P block to be used in other Socs where DX compliance was important. I suppose it depends whether the DX10.1 and full profile OpenCL compliant parts of the SGX545, which would be redundant in a Meego phone, are significant power draws.

Hmmm under the presupposition that Intel didn't ask for DX9/L3 compliance in 544/554 (which I can't figure out who else would) a SGX545 makes sense. However if Intel was one of the incentives for the L3 above I can't imagine a roadmap that contains a DX10.1 compliant core followed by a DX9.0/L3/MP config later on.

400MHz SGX535 was already the target for Moorestown.
I heard something like 4x current IGP performance (200MHz SGX535), which could mean something like a 600MHz SGX540.

I don't know what Intel is cooking but given their process technology advantage I'd suggest that a SGX544 2MP@>300MHz would be the smartest solution of them all.

Or something completely different, like getting the 6EU IGP from sandybridge (that would save them a lot of money from IP (not that Intel needs money right now)).
Sandybridge is perf/W and not perf/mW. As for saving them a LOT of money, how much do you think the royalty amount is exactly per core? You're probably looking at cents values, while for a high end console GPU IP the royalty is more in the $4-6 ballpark per unit.
 
I find it extremely odd that after months and months of waiting without a Moorestown release a customer would suddenly not only demonstrate a Medfield product but release it in a scant few months. Medfield is something Intel hasn't said an awful lot about.. surely if they were this far along they'd be the first to demo and present it.

It seems to me that Moorestown was scratched in order to speed up Medfield.
Moorestown devices were supposed to come during 2010.

Nonetheless during 2009, Intel made several official claims that Medfield devices would be on sale during 2011, and this just seems to be a part of that.

Given AMD's sudden "change of direction" to attack the handheld market (firing their CEO over that and all), I'd say that may have put some pressure in Intel to finally get Atom into handsets.



There's little chance that the 6EU variant Sandy Bridge IGP will be low enough power to work in an Atom Z-series SoC, much less be competitive in power draw with SGX. I don't have power numbers or any real great metric for them, but we know it performs pretty similarly to the GPU on Zacate which appears to take at least a few W in even its smaller incarnation. And I certainly won't so easily believe that Intel is beating AMD in GPU perf/W, although they do have a process advantage here. But it's still a design made for desktop/laptop CPUs.

We also know all the variants are clocked at 850MHz minimum, 1.1GHz maximum. Plus, all the mobile variants are carrying the 12EU part, which certainly says how confident Intel is about its power consumption (I wonder if the UM sandybridges will also carry the 12EU GPU).
And what if they make a ~100MHz min/~350MHz max 6EU part?

Plus, it's a part that's already been made and sold in 32nm, which could make a difference.

I'm not saying I'm betting my life it'll be there, just leaving the option open for discussion.
 
It seems to me that Moorestown was scratched in order to speed up Medfield.
Moorestown devices were supposed to come during 2010.

Other devices were supposed to come during 2010 as well, like OMAP4. nVidia said Tegra 2 phones would launch in Q2 2010. The industry isn't moving as fast as SoC vendors would like. And we don't know that anyone

Also, how would "scratching" Moorestown speed up Medfield deployment for Intel? Moorestown is finished, it has been ready and available for a good long while. Maybe vendors would want Medfield instead, but that won't make it come out sooner. Intel needs to get 32nm low-power working first.

Nonetheless during 2009, Intel made several official claims that Medfield devices would be on sale during 2011, and this just seems to be a part of that.

There's a really big difference between "2011" and "H1 2011." Granted, I'd like to see some of these claims.

Given AMD's sudden "change of direction" to attack the handheld market (firing their CEO over that and all), I'd say that may have put some pressure in Intel to finally get Atom into handsets.

What sudden change in direction? Is firing their CEO all you have to go on? In terms of actual design decisions AMD doesn't do anything suddenly, and thinking that firing a CEO is pressuring Intel is kind of out there.. AMD is in absolutely no position to release anything for handhelds any time in the near future, say this year.

We also know all the variants are clocked at 850MHz minimum, 1.1GHz maximum. Plus, all the mobile variants are carrying the 12EU part, which certainly says how confident Intel is about its power consumption (I wonder if the UM sandybridges will also carry the 12EU GPU).

Laptops can be a good couple orders of magnitude away from handhelds in power consumption, being good for one doesn't make it good for the other, not even close.

And what if they make a ~100MHz min/~350MHz max 6EU part?

Plus, it's a part that's already been made and sold in 32nm, which could make a difference.

I'm not saying I'm betting my life it'll be there, just leaving the option open for discussion.

There's a reason why companies like Intel make SB and Atom, or why AMD is making Brazos and Bulldozer. Single designs don't tend to scale the entire low power to high performance spectrum. Both in design and process choice. Intel having deployed 32nm for Nehalem and SB probably doesn't help them as much as you think for 32nm Atom because it's a high performance vs low power process. This is evidenced by there being like 2 years between the two. I'm sure if Intel could have had a 32nm Atom out by now they certainly would have.
 
Other devices were supposed to come during 2010 as well, like OMAP4. nVidia said Tegra 2 phones would launch in Q2 2010.
NVIDIA never ever (publicly) said that. They said tablets in Q2 2010 and phones in Q4 2010. The tablet delay was mostly (but not only) because of Android's immaturity for tablets, and I'd say a single quarter delay for phones is pretty good!

It's certainly expected in this industry for everyone to be overly optimistic about timeframes and this is not just true of application processors. I was amused when some people claimed Verizon launching LTE in Q4 2010 was on or even ahead of schedule. When I attended Mobile World Congress 2009, Verizon clearly said they planned a commercial roll-out in Q4 2009 or Q1 2010 iirc. The idea is that if you claim things will happen earlier than they realistically ever could, your partners will push themselves harder than they otherwise would because they'll fear their competitors will beat themselves to it.

Also, how would "scratching" Moorestown speed up Medfield deployment for Intel?
Remember that LG Moorestown phone? It was supposed to be released a long time ago, but was cancelled. If all of Intel's top-tier potential customers tell them they're not interested, they're not going to continue developing the necessary software enhancements for Moorestown so even the second tier OEMs won't bother (to take an extreme example, would you really buy an Android 1.6 phone today?)

This is evidenced by there being like 2 years between the two. I'm sure if Intel could have had a 32nm Atom out by now they certainly would have.
Don't forget phones have a much longer delay between tape-out and end-products. Even if it taped-out later than Westmere, Medfield has still been sampling for quite some time (it has been prioritised over the other 32nm Atom chips).
 
I was amused when some people claimed Verizon launching in Q4 2010 was on or even ahead of schedule. When I attended Mobile World Congress 2009, Verizon clearly said they planned a commercial roll-out in Q4 2009 or Q1 2010 iirc.
Verizon launching what?

Also, I'd be curious when is Intel's 22nm SoC process and 22nm SoC's coming online. IIRC, (not sure where I heard it) they said ithe process will come up alongside their high perf logic process. If that's true, and they are successful, I'd expect them to beat everyone to pulp next year. The 32/28 nm transition has barely begun for the rest of the vendors.
 
Verizon launching what?
Oops, I meant Verizon launching LTE. I'm not sure they launched anything big besides that in Q4 2010, but heh ;)
Also, I'd be curious when is Intel's 22nm SoC process and 22nm SoC's coming online. IIRC, (not sure where I heard it) they said ithe process will come up alongside their high perf logic process. If that's true, and they are successful, I'd expect them to beat everyone to pulp next year. The 32/28 nm transition has barely begun for the rest of the vendors.
I think Otellini implied end-products would be available in Q1 2012. And yes that's certainly much earlier than TSMC which only plans to have their 20nm process available (i.e. ready for customer tape-outs) in 4Q12 for High Performance and 1Q13 for SoCs (and that's assuming they're even still on schedule) - that's nearly 2 years behind Intel.
 
Don't forget phones have a much longer delay between tape-out and end-products. Even if it taped-out later than Westmere, Medfield has still been sampling for quite some time (it has been prioritised over the other 32nm Atom chips).

Though realistically those delays need to and should lesson over time. There is more and more pressure on the SOCs to not radically change the system interfaces and structures with each iteration. A lot of the time delay isn't necessarily in the hardware but in re-porting the low level software to the new SOC.
 
NVIDIA never ever (publicly) said that. They said tablets in Q2 2010 and phones in Q4 2010. The tablet delay was mostly (but not only) because of Android's immaturity for tablets, and I'd say a single quarter delay for phones is pretty good!

Okay, tablets - the point doesn't really change. It doesn't matter why they're delayed, the point is that delays have been happening all around. Although IMO it's not just software since companies were still eager to launch Tegra 2 tablets with pretty poor software.

It's certainly expected in this industry for everyone to be overly optimistic about timeframes and this is not just true of application processors. I was amused when some people claimed Verizon launching LTE in Q4 2010 was on or even ahead of schedule. When I attended Mobile World Congress 2009, Verizon clearly said they planned a commercial roll-out in Q4 2009 or Q1 2010 iirc. The idea is that if you claim things will happen earlier than they realistically ever could, your partners will push themselves harder than they otherwise would because they'll fear their competitors will beat themselves to it.

Yeah, exactly.. which is why I'd be really shocked if we saw Medfield phones in a few short months.

Remember that LG Moorestown phone? It was supposed to be released a long time ago, but was cancelled. If all of Intel's top-tier potential customers tell them they're not interested, they're not going to continue developing the necessary software enhancements for Moorestown so even the second tier OEMs won't bother (to take an extreme example, would you really buy an Android 1.6 phone today?)

However, I don't see what software development towards Moorestown wouldn't be directly applicable to Medfield, which is mainly a consolidation and a shrink. So software delays would effect both, certainly not leading to a accelerated Medfield release instead.

Don't forget phones have a much longer delay between tape-out and end-products. Even if it taped-out later than Westmere, Medfield has still been sampling for quite some time (it has been prioritised over the other 32nm Atom chips).

This I didn't know, thanks. Still, if it were really this close to release I would have expected showing at CES.
 
I'm not saying I'm betting my life it'll be there, just leaving the option open for discussion.

Intel doesn't hold only a SGX535 license from the entire SGX family and I'd be very surprised if Intel hasn't licensed already IMG's next generation S6. Intel stated itself that they're well armed with IMG's IP in the embedded space and don't intend to change anything any time soon.

Besides I'm afraid you're missing the entire point with Intel's own GenX stuff; if you clock those EUs as low as you suggest it's not only questionable that it'll burn less power/mm2 than a SGX variant, but since the GenX stuff relies more on frequency 6 EUs@350MHz would perform like crap. As I said IMG has on its website a demo from Q3A running on an Intel STB where I'd suggest that the SGX535 runs at 400MHz resulting into ~60fps@1080p. How much are you suggesting they'd ridicule their own stuff exactly? I wouldn't personally exchange a full TBDR with a "TilerThatSucks"(tm) but that's just me.

Current GenX isn't designed anywhere near to be able to scale to perf/mW heights. It'll cost Intel more to develop such a thing then a simple license from IMG and a few dozen cents per IP core royalties to IMG.

***edit: by the way xbit labs also takes reference for speculation on that Finnish website: http://www.xbitlabs.com/news/mobile..._N9_Expected_to_Feature_Intel_Atom_MeeGo.html
 
I find it a bit funny how everyone quickly dismisses Intel's own current graphics tech, comparing to the SGX family, while we've seen how weak the GMA500 turned out to be in a x86 environment.
Even the dirt-old GMA950 @ 220MHz found in Calistoga's N2xx series was ~3x faster than SGX535 @ 200MHz in the Z500 series, even with matching CPU clocks.
Z600 was scratched for all we know and that was the only product we knew for sure it would come with a SGX IGP.

Even more when no one appears to know exactly how much the HD2000/HD3000 consume and how it would scale with lower clocks.
 
Last edited by a moderator:
I find it a bit funny how everyone quickly dismisses Intel's own current graphics tech, comparing to the SGX family, while we've seen how weak the GMA500 turned out to be in a x86 environment.
Even the dirt-old GMA950 @ 220MHz found in Calistoga's N2xx series was ~3x faster than SGX535 @ 200MHz in the Z500 series, even with matching CPU clocks.

Not my fault that you missed that the display driver from Tungsten Intel had ordered weren't doing the underlying hw any justice. Of course is IMG in that case also responsible. Here's some old reference: http://www.mitrax.de/?cont=artikel&aid=36&page=7

Z600 was scratched for all we know and that was the only product we knew for sure it would come with a SGX IGP.
GMA600 hasn't been scratched for all I know, it merely won't see as much integration as Intel might have hoped probably due to manufacturers wanting to concentrate more (and they're hard to blame) on other sw platforms. Intel never had any intention afaik of integrating any SGX above netbook level.

Even more when no one appears to know exactly how much the HD2000/HD3000 consume and how it would scale with lower clocks.
Intel isn't going down to mobile/embedded heights with GenX hw. Ask Intel why. They'd be mighty stupid if they could scale down their own technology and would license over several hw generations third party graphics IP.

Here's the SGX535@400MHz (GMA600) demo at IMG's site: http://www.imgtec.com/demo_room/viewdemo.asp?DemoID=63&DemoTech=POWERVR%20Graphics&DemoDev=Imagination&#ViewPort

Considering 1080p is times higher than 1024*768, there's absolutely nothing wrong with GMA500 yielding only ~20fps in that resolution right? http://www.mitrax.de/?cont=artikel&aid=36&page=7 IMG's own OGL driver delivering nearly twice as much isn't any sort of indication either....
 
Not my fault that you missed that the display driver from Tungsten Intel had ordered weren't doing the underlying hw any justice. Of course is IMG in that case also responsible. Here's some old reference: http://www.mitrax.de/?cont=artikel&aid=36&page=7



The Z500+Poulsbo were launched in mid-2008. By then, the driver performance was terrible.
Almost 3 years later, the drivers are still bad and neither Intel, Tungsten or IMG did nothing about it.

What makes you think that will change for the next iteration?
 
The Z500+Poulsbo were launched in mid-2008. By then, the driver performance was terrible.
Almost 3 years later, the drivers are still bad and neither Intel, Tungsten or IMG did nothing about it.

What makes you think that will change for the next iteration?

I don't know what this situation was with Tungsten but if competent SGX53x drivers can be done for iOS, Android, Linux and probably others and Intel has demonstrated competent performance themselves then obviously it can be done and it shouldn't be a huge leap to make it done in a real product. Probably has sucked all this time because there wasn't enough market demand to make it not suck, but now Intel is hoping that'll change. Maybe in 2008 or whenever Intel signed with Tungsten IMG's reference drivers were immature. That wouldn't be the case now.

Wouldn't be very surprised at this point if Intel dropped Tungsten and moved to IMG's offerings. Given that MeeGo has a Linux-derived kernel it shouldn't be very much work.
 
Back
Top