Samsung Orion SoC - dual-core A9 + "5 times the 3D graphics performance"

M

Mike11

Regular
Well, doesn't look good for a 32nm A4 successor in H1/2011 (iPad in Q1, iPhone in Q2) if Samsung can't even use 32nm for their own H1/2011 High-end SoC. Not that big of a deal for the iPad, but could make things harder for a >800MHz dual-core Cortex-A9 in the next iPhone.

Samsung Orion, 45nm 1GHz dual-core ARM Cortex-A9, mass production in H1/2011
http://www.engadget.com/2010/09/07/samsungs-orion-is-the-1ghz-dual-core-arm-cortex-a9-weve-all-be/
Samsung Introduces High Performance, Low Power Dual CORTEXTM - A9 Application Processor for Mobile Devices

TAIPEI, Taiwan--(BUSINESS WIRE)--Samsung Electronics Co., Ltd., a world leader in advanced semiconductor solutions, today introduced its new 1GHz ARM® CORTEXTM A9-based dual-core application processor, codenamed Orion, for advanced mobile applications. Device OEM developers now have a powerful dual processor chip platform designed specifically to meet the needs of high-performance, low-power mobile applications including tablets, netbooks and smartphones. Samsung's new processor will be demonstrated at the seventh annual Samsung Mobile Solutions Forum held here in Taiwan at the Westin Taipei Hotel.

"Consumers are demanding the full web experience without compromise while on the go," said Dojun Rhee, vice president of Marketing, System LSI Division, Samsung Electronics. "Given this trend, mobile device designers need an application processor platform that delivers superb multimedia performance, fast CPU processing speed, and abundant memory bandwidth. Samsung's newest dual core application processor chip is designed specifically to fulfill such stringent performance requirements while maintaining long battery life."

Designed using Samsung's 45 nanometer low-power process technology, Orion features a pair of 1GHz ARM Cortex A9 cores, each comes with a 32KB data cache and a 32KB instruction cache. Samsung also included a 1MB L2 cache to optimize CPU processing performance and provide fast context switching in a multi-tasking environment. In addition, the memory interface and bus architecture of Orion supports data intensive multimedia applications including full HD video playback and high speed 3D action games.

Samsung's new application processor incorporates a rich portfolio of advanced multimedia features implemented by hardware accelerators, such as video encoder/decoder that supports 30fps video playback and recording at 1080P full HD resolution. Using an enhanced graphics processing unit (GPU), the new processors are capable of delivering 5 times the 3D graphics performance over the previous processor generation from Samsung.

For design flexibility and system BOM cost reduction, Orion integrates a set of interfaces commonly used in mobile devices to configure various peripheral functionalities. For example, with this processor, customers have the choice to use different types of storage including NAND flash, moviNANDTM, SSD or HDD providing both SATA, and eMMC interfaces. Customers can also choose their appropriate memory options including low power LPDDR2 or DDR3, which is commonly used for high performance. In addition, a global positioning system (GPS) receiver baseband processor is embedded in the processor to seamlessly support location based services (LBS), which is critical in many emerging mobile applications.

Orion features an onboard native triple display controller architecture that compliments multi-tasking operations in a multiple display environment. A mobile device using the Orion processor can simultaneously support two on-device display screens, while driving a third external display such as a TV or a monitor, via an on-chip HDMI 1.3a interface.

Orion is designed to support package-on-package (POP) with memory stacking to reduce the footprint. A derivative of Orion, which is housed in a standalone package with a 0.8mm ball pitch, is also available.

Samsung's new dual-core application processor, Orion, will be available to select customers in the fourth quarter of 2010 and is scheduled for mass production in the first half of 2011.
Click to expand...

"5 times the 3D graphics performance over the previous processor generation from Samsung"
5x SGX540? What SGX GPU might that be? SGX543MP2?

"Samsung's new application processor incorporates a rich portfolio of advanced multimedia features implemented by hardware accelerators, such as video encoder/decoder that supports 30fps video playback and recording at 1080P full HD resolution."
Dedicated VXE and VXD? I wonder which one Samsung used. Is it too early for VXE380 and VXD390 implementations?
 
Mike11 said:
"5 times the 3D graphics performance over the previous processor generation from Samsung"
5x SGX540? What SGX GPU might that be? SGX543MP2?
Click to expand...

Sounds like a typical marketing exaggeration. Additional CPU core most likely included in the equation or better for the entire SoC; long story short you won't have it easy to find that many case scenarios with up to 5x times performance difference. On the GPU front however if it should have a 2MP at the same frequency of a SGX540 you can easily claim twice the real throughput ;) On a more humorous note it could be an ARM Mali multi-core variant and the 5x times claim comes from master Ed :devilish:

I'd dare to speculate that they'll "serve" Apple first and then release their own stuff.
 
Given that Intel probably nearly certainly never licensed the SGX543MP and waited for the SGX544MP, I don't see the problem with assuming Samsung was one of the secret licensees. But yes, it could be Mali-based too.

Honestly Orion is not very impressive as the quad-core Tegra3 on 28nm (and likely its Qualcomm equivalent) will sample one quarter later at most. I can see Samsung very successfully using its own APs in many of its own phones along with a slim baseband from Infineon/Icera/Qualcomm, but I can't see them being very successful at third parties.
 
Arun said:
Given that Intel probably nearly certainly never licensed the SGX543MP and waited for the SGX544MP, I don't see the problem with assuming Samsung was one of the secret licensees. But yes, it could be Mali-based too.
Click to expand...

Not unlikely at all, but it sounds more like Samsung isn't going to concentrate on one graphics technology source either (which is a positive aspect in my book).

Honestly Orion is not very impressive as the quad-core Tegra3 on 28nm (and likely its Qualcomm equivalent) will sample one quarter later at most. I can see Samsung very successfully using its own APs in many of its own phones along with a slim baseband from Infineon/Icera/Qualcomm, but I can't see them being very successful at third parties.
Click to expand...

Anything so far Tegra has had a stronger (inevitably) focus for tablets/netbooks than for the smart-phone segment. We'll see how their 3rd generation will pan out and of course if Samsung has more ambitious plans for tablets in the future.

If I look today over the Galaxy S the overall impression for Samsung's current high end smart-phones is far better than what NOKIA for example has on shelves at the moment (and yes that's completely irrelevant what each SoC actually contains). Could very well be that Samsung (see also for reference parts of the AMD Fusion project) intends to focus more on the graphics than on the general purpose processing side for their next generation and I don't think I can decide yet whether it would be a wise or dumb decision.

Another question that bothers me is whether Samsung will continue to manufacture SoCs for Apple for the foreseeable future or not. If yes I can see that fact influencing their design decisions also.

In any case you can see the advantages and disadvantages of any SoC more than anything when you hold a final device in your hands and can judge upon its responsiveness and efficiency amongst others in real time. That's exactly the point where fancy on paper specs lose their importance after all.
 
Ailuros said:
Anything so far Tegra has had a stronger (inevitably) focus for tablets/netbooks than for the smart-phone segment. We'll see how their 3rd generation will pan out and of course if Samsung has more ambitious plans for tablets in the future.
Click to expand...
Well there's also a question of timeframes, the lead times for phones are slightly longer. I do expect a Tegra2 phone from both Motorola and Palm in Q1 2011, so that's three fairly major OEMs that would probably make at least two phones each. Certainly not enough to amortise the R&D, but could be worse.

If I look today over the Galaxy S the overall impression for Samsung's current high end smart-phones is far better than what NOKIA for example has on shelves at the moment (and yes that's completely irrelevant what each SoC actually contains).
Click to expand...
Not entirely; from both a cost and a performance standpoint, Nokia's approach with the N8 (same 45nm ARM11 SoC as in everything else + Broadcom multimedia coprocessor) is very dubious. Nearly all of their S60 phones are still based on single-chip TI baseband+application processors, and the features of those few chips were decided by Nokia a long long time ago. Their decisions, IMO, turned out very wrong because that chip is cost-efficient but not good enough in the upper mid-range, and certainly not in the high-end.

This has the further consequence of OS development being centered on a fairly weak processor, which means that unlike Apple they couldn't design the user experience around 3D acceleration early enough among other things, and too many changes were incremental. If Nokia had decided to integrate a MBX Lite in that 45nm chip (which started shipping in early 2009 Nokia phones), things already would have been different at very little cost. Not because of the chip itself, but because of the feedback effect on the OS.

Could very well be that Samsung (see also for reference parts of the AMD Fusion project) intends to focus more on the graphics than on the general purpose processing side for their next generation and I don't think I can decide yet whether it would be a wise or dumb decision.
Click to expand...
That's possible. Their latest SoCs are also strong video-wise. Certainly the difference between 1.5GHz 4xA9 and 1GHz 2xA9 isn't day and night in practice, whereas the difference between 720p Baseline video encoding and 1080p High Profile encoding arguably is with a decent cameraphone sensor.

Another question that bothers me is whether Samsung will continue to manufacture SoCs for Apple for the foreseeable future or not. If yes I can see that fact influencing their design decisions also.
Click to expand...
They're still designed by Apple. They bought Intrinsity. I don't think Apple's roadmap changes anything for Samsung except in terms of being more or less aggressive process-wise.

In any case you can see the advantages and disadvantages of any SoC more than anything when you hold a final device in your hands and can judge upon its responsiveness and efficiency amongst others in real time. That's exactly the point where fancy on paper specs lose their importance after all.
Click to expand...
Sure, but that's often much more a software thing - and not just drivers, but in many cases the implementation of the OEM's custom UI (think of the crap some manufacturers have done with decent chips).
 
rpg.314 said:
A quad core in a smartphone, now that would be something. :)
Click to expand...
Yeah, hopefully we'll see a TurboBoost-like feature where for single-threaded workloads you can go up to, say, 1.5GHz at a higher voltage and for quad-core workloads you can go up to 1GHz at a lower voltage. Otherwise it seems like a waste. Keep in mind also we're looking at a triple gate oxide SiON process.
 
I can't really talk about the chips, but I can talk about this.
Arun said:
Sure, but that's often much more a software thing - and not just drivers, but in many cases the implementation of the OEM's custom UI (think of the crap some manufacturers have done with decent chips).
Click to expand...
Yes a thousand times over, and not just limited to the OEM either. Google's default scrolling application list UI for the Nexus One is really poor, and might cause people to reflect poorly on the Snapdragon, when in fact it's just badly programmed. If Google struggle....

The software is the reason iOS has done so well (although we'll take a bit of credit when it comes to the games!). Everyone else building mid-range and upwards mobile phones need to pay attention to the basics, and the reason it's good on the Galaxy S phones is because Samsung have.

Although looking at the comments on Anandtech's Epic 4G review, it's sometimes understandable as to why Android vendors don't do much there. If the users of the phones are more interested in crap UIs just because they're different to Apple's, rather than the usability, why spend the time making them slick too :rolleyes:

Five times the graphics performance of the last gen might might them slick by brute force though :LOL:
 
Yeah, responsiveness is given top priority by Apple, whereas Android in general feels more about the features at this point. Both approaches definitely lead to their own tangible product advantages.

TSMC closing the gap that much on Intel's timetable for process transition is suspect. I'm not buying the nVidia roadmap for volume production, so I don't see Samsung's disadvantage in this off-year to be so much.

Is VXD/VXE the IP Samsung uses for video?
 
On Orion: the old Samsung roadmap is still available here: http://images.intomobile.com/wp-content/uploads/2010/04/samsungapforcomputing.pdf - apparently there will be a 32nm shrink of Orion next year (codename: Hercules) and a single-core 32nm chip codenamed Pegasus. We'll have to see if it'll have the same GPU or not.

This same Samsung presentation lists Orion as having first samples in Q3 2010 (it's the only Samsung chip with that first sample date), and the Mali press release lists the first Mali-based AP as having first samples in Q3 2010. I'm not sure if the latest PR's Q4 claim indicates a delay or just means that first samples refers to engineering and not customer samples), but I think there's only one possible conclusion here: it is based on the Mali-400, probably a quad-core implementation to claim that level of performance.

It's certainly possible that SGX543MP wouldn't have been ready in time for Orion anyway (who knows what Samsung plans for Hercules). According to http://www.imgtec.com/corporate/newsdetail.asp?NewsID=497, the final IP would be available in December 2009, and Orion was planned to have first samples in Q3 2010 - which would be a very tight schedule but not really impossible (not that it matters since the sample timeframe makes it very clear it's Mali anyway).

Rys said:
Five times the graphics performance of the last gen might might them slick by brute force though :LOL:
Click to expand...
I wonder... in your experience, do you genuinely believe that to be true? I cannot be certain, but in many cases it seems to me the bottleneck isn't so much the GPU but either the CPU or, more dramatically, the I/O. One fairly extreme case in iOS4.0 (I think it has been fixed by now?) was the integrated search engine (Google/Bing/Yahoo) suggestion feature.

It literally waited for a reply back from the search engine's servers before displaying the characters you were typing! It could just have done an asynchronous request and keep using the slightly outdated suggestions for one more second, but no, it had to wait and the result was embarrassingly choppy on my 3GS until (I assume) they fixed it.

The same problem can exist with NAND I/O rather than TCP I/O, or because of either large 'all at a time' computations that make you wait between screens or badly implemented 'on demand' computation that happens exactly when it's needed and isn't instant (rather than predicting what will be needed a bit in advance or using an empty placeholder until it's computed).

I also wonder how cheap 4-bit MLC would affect this in low-end smartphones going forward... Anyway you can't brute force your way through if you're not addressing the bottleneck and I suspect this is a more frequent problem than most people (and probably even devs) realise. The only real solution, of course, is a more intelligent (but also more complex) user interface implementation.
 
Lazy8s said:
TSMC closing the gap that much on Intel's timetable for process transition is suspect. I'm not buying the nVidia roadmap for volume production, so I don't see Samsung's disadvantage in this off-year to be so much.
Click to expand...
Not at all. The 28LPT process is SiON (no High-K) and, unlike practically everybody else's process, does NOT use double patterning immersion litography. It is unambiguously the cheapest 28nm process out there, but it's also not extremely dense (smallest contacted gate pitch is worse than Intel's 32nm process, whereas TSMC had an advantage there on 40nm versus Intel's 45nm - SRAM density is still very good though) and not very fast (but this is nicely compensated by most customers using TPG/triple gate oxide which makes good sense for the target markets of that specific process, including application processors - the best is obviously High-K+TPG which is TSMC will offer as the 28HPM process and Intel will use in its 32nm process for Medfield AFAIK).

EDIT: Oh yeah, where the hell does the GPS IP came from? Fastrax Accelerated Software GPS? Fastrax pure-software? Another IP provider? In-house? It's a frankly unexpected and slightly bizarre design decision, but why not...
 
Where are all these ARM based netbooks everyone is touting (and apparently pinning so many hopes on)? Selling in China?
 
Arun said:
I wonder... in your experience, do you genuinely believe that to be true? I cannot be certain, but in many cases it seems to me the bottleneck isn't so much the GPU but either the CPU or, more dramatically, the I/O
Click to expand...
Just in the broad case of Android doing more with the GPU as time goes by (and in the specific case of the Nexus One's scrolling list, using the GPU to help that is so obvious it hurts), yes. Obviously you can bottlenecked by different things at any given time, but the GPU is becoming a place where more of the work is done, and so just brute-forcing a level of performance simply by using it is doable for some things.
 
Rys said:
Just in the broad case of Android doing more with the GPU as time goes by (and in the specific case of the Nexus One's scrolling list, using the GPU to help that is so obvious it hurts), yes. Obviously you can bottlenecked by different things at any given time, but the GPU is becoming a place where more of the work is done, and so just brute-forcing a level of performance simply by using it is doable for some things.
Click to expand...
Well I could be wrong, but it seems to me that even on a 110MHz SGX530 (~OMAP3630) you'd have to do a gigantic effort to be significantly bottlenecked by the GPU (if below the screen fps) when offloading such things to it.

So then you're not really brute forcing your way through; you're offloading things to the GPU instead of brute forcing them on the CPU. As more and more gets offloaded to the GPU as you say, and as the CPUs become more powerful at the same time, it simply seems to me I/O can only become a bigger problem for implementations that haven't been perfectly thought through.

Or is it your experience that even a 200MHz SGX540 could not be fast enough to achieve the screen's framerate for UI processing (i.e. it would be the bottleneck)? I know some programmers are pretty awful in those fields and a single one is enough to ruin everything, but come on! :)
 
Mike11 said:
They are currently undergoing surgery, keyboard removal to be precise ;)
Click to expand...

So perhaps we'll see a raft of android/chrome os based tablets next.

For all the work into UI, I hope somebody takes a serious (and wide ranging, including overall industrial design as well if need be) look at making typing easier on these xPads. For all the intuitiveness of touch based UI, typing is essential for many tasks.
 
Arun said:
Well I could be wrong, but it seems to me that even on a 110MHz SGX530 (~OMAP3630) you'd have to do a gigantic effort to be significantly bottlenecked by the GPU (if below the screen fps) when offloading such things to it.
Click to expand...
Yes, but you'd also need to use it as more than just a blitter in the first place :D

Arun said:
So then you're not really brute forcing your way through; you're offloading things to the GPU instead of brute forcing them on the CPU. As more and more gets offloaded to the GPU as you say, and as the CPUs become more powerful at the same time, it simply seems to me I/O can only become a bigger problem for implementations that haven't been perfectly thought through.
Click to expand...
Yes, and it's already quite a big problem in some cases (not just I/O throughput, but consumption of the memory space too).

Arun said:
Or is it your experience that even a 200MHz SGX540 could not be fast enough to achieve the screen's framerate for UI processing (i.e. it would be the bottleneck)? I know some programmers are pretty awful in those fields and a single one is enough to ruin everything, but come on! :)
Click to expand...
SGX wouldn't ever be the real bottleneck when it comes to drawing and compositing pixels, if fed properly, but what feeds it can be abused in these devices. We're bandwidth optimised for a reason, because there's not that much of it and we share it with the CPU. Consuming what little there is by being stupid in software is sometimes surprisingly easy, especially as the resolution goes up and up.
 
Rys said:
Yes, and it's already quite a big problem in some cases (not just I/O throughput, but consumption of the memory space too).
Click to expand...
Ah yes, memory space... It's such a shame LPDDR2 has such a premium over DDR3. I really can't blame Apple for sticking to LPDDR1 but using a 64-bit memory bus in the A4, and even that is hardly cheap. And good point about bandwidth obviously, it's easy to forget how easy that stuff goes 'poof!' - I wonder how many UIs used anything else than full 32-bit RGBA textures for components. Something like the new 8-bit 'higher quality' DXTC format in DX11 would come in handy here, but it's nearly certainly too expensive to implement in the handheld world for now.

roninja said:
Now if Orion is software compatible with S5VP210 which we know is SGX540 - does this narrow down the GPU vendor options?
Click to expand...
Not really. It can use a different GPU but expose it through the exact same APIs. Furthermore, 'software compatible' is a frequently overused notion. I still think it's pretty clear based on the 'first samples' date that it's actually
 
You must log in or register to reply here.

Similar threads

I
Samsung Exynos 9810
2
Replies
25
Views
13K
Nebuchadnezzar
N
P
Portable console for all PC games - PGS
2
Replies
30
Views
12K
StakanPortveina
S
P
ARM announces Cortex-A50 series
2
Replies
22
Views
15K
silent_guy
S
O
Inertial Sensors: DualShock 4 vs Sixaxis/DS3 accelerometer/gyroscope chips
Replies
0
Views
10K
onQ
O
D
"Obscure" 800MHz dual-core SoC in Samsung Galaxy Ace 2?
Replies
9
Views
5K
Exophase
E
Back
Top