Samsung Exynos 5250 - production starting in Q2 2012

Cortex-A8 did actually hit products in 2008, I remember seeing an Archos MID with it in fall 2008 before Pandora was supposed to ship, making it about 3 years. But yeah, Cortex-A9 pulled that in to about 2.5 years (announce December 2007, first device summer 2010), and Cortex-A15 2 years (announce fall 2010, first device shipment fall 2012). A7 is already in shipping products as of last month so that counts as barely over 1 year. At this rate I wonder what the chances are of saying Cortex-A57 and Cortex-A53 by the start of 2014.

But more impressive is the SoCs themselves. It used to be normal to wait for years for TI to come out with their new SoCs.. no wonder TI decided they can't keep up. Same with Freescale. Now nVidia sends out demo units with its Tegra 4 the same day they announce it and Chinese and Taiwanese SoC makers like Allwinner, Rockchip, and Mediatek have their SoC shipping in tablets at virtually the same time they're announced.

Exciting times. Good for me because I hate waiting for things to show up in something

 

http://www.imgtec.com/corporate/pre..._Half_Year_Results_Presentation_12dec2012.pdf

Page22; not saying it is but what else could an important design win be?

 

You're much more on the pulse than me then. I had forgotten about the shipping A7 stuff.

The most exciting thing to me is to have multiple vendors doing custom ISA implementations again. The Intel and AMD rivalry has been stale ever since Conroe.

Notice how Apple never gets mentioned

 

I'd say getting back into any Samsung phone/tablet soc could be considered an important design win for IMG

I don't know enough about power requirements, but anandtech recent article seem to suggest that the dual A15 samsung soc wouldn't be a good choice for a phone.

Even allowing for the small reduction in process size, it would appear to me that the only way this octa core chip can be feasible in a phone is if the 4xA15s are rarely used?

 

All four A15s can probably scale to a low enough clock speed/voltage to be active simultaneously even in a phone. I'm sure the minimum consumption for that is under 2W. The question is if they'll still provide more performance than the four A7s at this point.

 

I tipped off Brian Klug to investigate it: https://twitter.com/nerdtalker/stat...&iid=am-60662501713578500319581592&nid=27+234

What are the advantages for Samsung to even go back to IMG unless it is a Rogue GPU? ARM has been touting their CCI-400 interconnect and how well it plays with the new Mali T6xx series also especially in compute scenarios.

I can imagine power consumption to be the main reason, but performance? The T624 would hold up to it I reckon.

 

Amazing, your most important investor and customer and they never appear in the presentation ?!!??

BTW, wasn't aware that Nethra got acquired by IMG. AFAIK Nethra has some camera processing and video encoding IP. From what I know, they supplied IP to some high-def 4K camera makers in the US (Red ??). Interesting acquisition.

 

ImgTec is on an acquisition tear between that, MIPS and Caustic. All the while Apple is a 10% stakeholder.

 

Apple is known to be extremely strict with its providers: they don't have the right to disclose anything unless they want to break their contract

 

I think the rate RAM sizes are increasing, whether due to actual need or simply to market big GBs, is going to push for the presence of 64-bit ARM in 2014. We'll probably see 4GB RAM at least in tablets this year and greater than 4GB in 2014. Cortex A15 has PAE support, but ideally we'll have 64-bit support to better utilize that RAM.

It'll be interesting to see what the timing of release of custom 64-bit designs from Apple, nVidia, and Qualcomm will be to reference Cortex A5x SoC.

Despite the occasional talk of whether Apple will ever switch to x86 for the mobile platforms, what might be more interesting is whether possible tight integration between PowerVR and MIPS ever makes it worthwhile to switch away from ARM.

 

Samsung took a 5XT license a while back, about the right time ago for something to be popping up shortly. They don't have an ANNOUNCED rogue license.

The most obvious reason for me would be to make it easier to get into the next round of windows RT hardware. SGX is already qualified via omap. I imagine using the same graphics hardware means one less set of graphics drivers for microsoft to qualify/support, also remember that Mali has no previous history of windows as their mali400 doesnt have dx compliance.

 

Yes, IMG bought them some months ago. They have created a new division,PowerVR vision, and have indicated they'll have some camera IP for licensing early 2014.

 

I'd exept a T624 to beat a 544MP3@533MHz in terms of performance (at the same frequency). Since you can imagine power consumption (which I'm not so convinced about at the moment) then think of smartphone SoC possibilities. I'd still think that without lowering frequencies and without any possible further SoC revisions dumping even a 5250 at the moment in a smartphone isn't an easy task. If the latter should be a headache, then how far would they have to go down in frequencies and how would performance then look like?

 

DX compliance on a big.LITTLE chip is basically meaningless as Windows won't be able to use the A7 cores. It's really a sub-optimal solution.

They also announced an Exynos 5 Quad during CES which was running on their Smart TVs, who knows what's in that one either.

 

According to Reuters, Samsung will also have more Exynos design wins among the chinese vendors this year. So far Meizu has been one of the few but apparently this is going to change

I

 

They need to diversify once Apple drops them as a fab too.

 

I hope MS fixes this...

 

It pretty ironic that ARM fixing a limitation in their CPU's is marketed as a reason for using T6xx. The reality is that IMG GPU's have supported the level of cohenency, that ARM have only just enabled, since the late 90's!

 

The big.LITTLE white paper details how processes migrate from A15 cores to A7.

Colour me skeptical. It takes 20.000 cycles to transfer a process from a A15 core to a A7 core, register-files, interrupt state and data cache needs to be transferred. On top of that the L2 cache for the A15 can't be shut down because it might hold dirty data. So you need to power two L2 caches, but can only allocate in one !

IMO, the most (only) realistic use of heterogenous performance cores is to statically allocate a process to run either on a A15 cluster or A7 cluster. OS services and background tasks can easily run on A7s with high performance apps, games and (most importantly) benchmarks run on the A15

Cheers

 

I'm going to have a blast in kernel development and modding on this chip