Samsung S3C2460
- Thread starter Lazy8s
- Start date
Freak'n Big Panda
Regular
ooh looks promising. Is it's power useage suitable (low enough) for use in a cell phone (a smart phone maybe?)? If such functionality could be pakced into a small form factor samsung could really take a lot of the mobile market.
EDIT: after checking out the samsung site I answered my own question. Too big and power hungry for cell phone usage.
EDIT: after checking out the samsung site I answered my own question. Too big and power hungry for cell phone usage.
Last edited by a moderator:
It's made for cellphones. The target markets even mention smartphone specifically.
The performance of the S3C2460 is actually very conservative compared to SoCs which power the successful FOMA 902i cellphones that launched some time ago. Using Texas Instruments' OMAP2420, they employ an ARM11 and a full MBX plus the VGP.
The performance of the S3C2460 is actually very conservative compared to SoCs which power the successful FOMA 902i cellphones that launched some time ago. Using Texas Instruments' OMAP2420, they employ an ARM11 and a full MBX plus the VGP.
Last edited by a moderator:
well the iphone is now out and......
We suspect that it may be Samsung's S3C6400 based on the ARM1176 core, however some readers have written us stating that it's more likely to be the S3C2460; judging by the model numbers on the chip itself, the ARM processor may be a part of a multi-chip package that includes 1Gbit of system memory, for running the iPhone's OS. The K4X1G153PC-XGC3 is a Samsung part number, indicating a 1Gbit memory device, but it is placed on the same package as the ARM processor itself.
http://www.anandtech.com/mac/showdoc.aspx?i=3026&p=3
We suspect that it may be Samsung's S3C6400 based on the ARM1176 core, however some readers have written us stating that it's more likely to be the S3C2460; judging by the model numbers on the chip itself, the ARM processor may be a part of a multi-chip package that includes 1Gbit of system memory, for running the iPhone's OS. The K4X1G153PC-XGC3 is a Samsung part number, indicating a 1Gbit memory device, but it is placed on the same package as the ARM processor itself.
http://www.anandtech.com/mac/showdoc.aspx?i=3026&p=3
I like that idea Arun! I wrongly assumed Anand found the Arm11 through that specific part number. The only thing is could the 2460 have H.264 decode as part of it's MPEG-4 capability? Because that appears to be the only Samsung with 3D, so otherwise there has to be a discrete video chip right?
And if you are going to have a decent discrete video processor why even have an applications processor with adavanced video at all? Or even audio if the discrete processor is better at that? Would it practical to keep switching on and of the dicrete and integrated parts as warranted?
And if you are going to have a decent discrete video processor why even have an applications processor with adavanced video at all? Or even audio if the discrete processor is better at that? Would it practical to keep switching on and of the dicrete and integrated parts as warranted?
Nope, Samsung clearly dissociates MPEG-4 from H.264 in its other products' specifications. So that chip clearly does not support it.
Jen-Hsun did confirm in a CC that the iPhone used OpenGL. So unless Apple is using a software implementation (which seems massively ridiculous to me, given their focus on battery life) it doesn't really matter which of these two Samsung chips they are using: they'd need another chip to do either 3D or video in either case.
My guess? S3C2460 + GoForce 5300, with the active display controller being the one on the S3C2460, so compressed video data would flow from NAND to the ARM to the GoForce, and then the uncompressed data would be moved back from the GoForce to the ARM.
Note that here I am just considering what is, IMO, the most likely configuration based on the premise that the chips in the iPhone have all been announced and are not custom. If you believe that they *are* custom or just haven't been publicly announced yet, then imagination is the only frontier... Which would also make speculating about it completely stupid, boring and meaningless.
Jen-Hsun did confirm in a CC that the iPhone used OpenGL. So unless Apple is using a software implementation (which seems massively ridiculous to me, given their focus on battery life) it doesn't really matter which of these two Samsung chips they are using: they'd need another chip to do either 3D or video in either case.
My guess? S3C2460 + GoForce 5300, with the active display controller being the one on the S3C2460, so compressed video data would flow from NAND to the ARM to the GoForce, and then the uncompressed data would be moved back from the GoForce to the ARM.
Note that here I am just considering what is, IMO, the most likely configuration based on the premise that the chips in the iPhone have all been announced and are not custom. If you believe that they *are* custom or just haven't been publicly announced yet, then imagination is the only frontier... Which would also make speculating about it completely stupid, boring and meaningless.
Well the 5300 is probably very low power. But I wonder whether the 3D in the 2460 is robust enough to run that OS. And obviously we don't know the power characteristics of any of the GoForce chips (or do we?). It may be that the 5500 also excels there. Not to mention that IF the power characteristics of the 5500 are suitable, the iPhone sure has a nice screen for games, and surely that would be a big potential deal for Apple once its software ducks are lined up.
Point is, if it is a discrete 3D chip it might be a Samsung applications processor that does not have the 3D video features - either a lesser ARM9 or the ARM11 chip.
Or something custom or unannounced.
Point is, if it is a discrete 3D chip it might be a Samsung applications processor that does not have the 3D video features - either a lesser ARM9 or the ARM11 chip.
Or something custom or unannounced.
Heh, this speculation is quite attractive too: http://www.investorshub.com/boards/read_msg.asp?message_id=20881648
But then the question becomes where the OpenGL support comes from. Apparently, the PXA-300 does support OpenGL ES 1.1 in software, and it is true that the 3D core would be idling 90% of the time on the iPhone, and when it's used, it might never be at the same time as anything else.
The one thing I like most about the PXA-300 is that it might explain why the iPhone only has a 2MP camera. However, that might also be why whoever is 'leaking' that thought about the chip, if it was actually just speculation and not a real leak.
To answer a few of your points quickly: MBX-Lite would be more than sufficient for the iPhone's OS, 5500 is quite attractive in terms of power consumption but the 5300 would probably be even better (although bus traffic might hurt it a bit against an application processor with integrated audio/video hardware).
One of the key reasons why I like the "S3C2460 + GoForce 5300" theory is that analysts originally said NVIDIA had the contract (for the iPod widescreen and/or the iPhone) then claimed that Samsung had won an integrated ARM/video processor socket, displacing NVIDIA. But if that processor is a S3C2460, Samsung and NVIDIA are actually complementary. I don't think that little bit of wasted MPEG-4 hardware is going to kill anyone either.
EDIT: Although those rumours were related more to the next-gen iPod than the iPhone, but I wouldn't be surprised if the HW for the two was actually very similar, if not identical, except for the wireless parts.
But then the question becomes where the OpenGL support comes from. Apparently, the PXA-300 does support OpenGL ES 1.1 in software, and it is true that the 3D core would be idling 90% of the time on the iPhone, and when it's used, it might never be at the same time as anything else.
The one thing I like most about the PXA-300 is that it might explain why the iPhone only has a 2MP camera. However, that might also be why whoever is 'leaking' that thought about the chip, if it was actually just speculation and not a real leak.
To answer a few of your points quickly: MBX-Lite would be more than sufficient for the iPhone's OS, 5500 is quite attractive in terms of power consumption but the 5300 would probably be even better (although bus traffic might hurt it a bit against an application processor with integrated audio/video hardware).
One of the key reasons why I like the "S3C2460 + GoForce 5300" theory is that analysts originally said NVIDIA had the contract (for the iPod widescreen and/or the iPhone) then claimed that Samsung had won an integrated ARM/video processor socket, displacing NVIDIA. But if that processor is a S3C2460, Samsung and NVIDIA are actually complementary. I don't think that little bit of wasted MPEG-4 hardware is going to kill anyone either.
EDIT: Although those rumours were related more to the next-gen iPod than the iPhone, but I wouldn't be surprised if the HW for the two was actually very similar, if not identical, except for the wireless parts.
There would almost certainly be overlap on a standalone video ipod if there is one.
As for different analyst reports - so many inaccuracies hard to sort wheat from chaff.
I'm not convinced that the MBX in that Samsung chip would truly be sufficient. If 5500 power is so good, why not put strong 3D capabilties in. It's important for the OS and with so many things moving to 3D plus games that would really make the iPhone killer.
As for different analyst reports - so many inaccuracies hard to sort wheat from chaff.
I'm not convinced that the MBX in that Samsung chip would truly be sufficient. If 5500 power is so good, why not put strong 3D capabilties in. It's important for the OS and with so many things moving to 3D plus games that would really make the iPhone killer.
MBX Lite devices like the Dell Axim X50v trail only MBX devices in performance benchmarks such as 3DMarkMobile and GLBenchmark, so it'd be sufficient for the iPhone's requirements if any portable processor would.
Is the GoForce 5500's 3D core clocked near the 150-MHz that other parts of the chip supposedly are? If so, competitive power consumption would be an incredible feat.
Is the GoForce 5500's 3D core clocked near the 150-MHz that other parts of the chip supposedly are? If so, competitive power consumption would be an incredible feat.
That's true, but after all these years, I'd certainly hope that I have greater ability than most dissociating wheat from chaff!
Note that I'm not saying I know for sure which of the two it is, but from my experience, that is definitely the kind of data that sometimes has a very real basis. The Samsung combined application/video processor in analyst reports, that is.MBX Lite is not some kind of borked half-assed core that couldn't run GLQuake to save its life. Quite on the contrary, it is already a few times too powerful for the iPhone's GUI, as far as I can tell, according to the theoretical specs and what I've see of the iPhone.
I would be VERY surprised if the 5500's 3D core was as efficient as the MBX Lite for the iPhone's purposes, and I would also be VERY surprised if the 5300 wasn't more power-efficient than the 5500. Put two and two together, and I'm sure you can see why I think it is a more logical design decision.
The MBX Lite is more than sufficient for basic 3D games at 320x480... Is it as powerful as a GoForce 5500? Maybe not. But it probably also takes less power and die space. I don't want to guess efficiency details because then we'll just enter another debate about that stuff, but let's just say I would be surprised if the 5500's 3D core was more efficient (perf/mm2, perf/watt) than the MBX Lite.
In terms of gaming, and attractive part of that Samsung chip is also that you could do T&L on the DSP, since it'll be idle for audio and video anyway if there is indeed a GoForce too.
The 5300 only does video and audio, it doesn't have a 3D core; the 5500 and 4800 are the ones with that. There would indeed be several duplicated features, however:
- Camera interface.
- MPEG-4 fixed-function blocks.
- DSPs (although the one on the S3C2460 could still be used for T&L, I guess)
- JPEG compression/decompression fixed-function blocks.
However, if the audio/video blocks are better on the 5300 (or on any other 'GPU' from any other company that would reside next to the main processor), I think it might still be a smart choice to go that route.
Source/Link?Lazy8s said:
I'm not even sure how that would work, since lighting happens *after* transformations. But yes, the 5500's 3D core has just been overclocked, I don't think they even added anything at all, but I could be wrong.Errr, what?
http://stream.ifixit.com/Parts/images/stream/images_large/25-21.jpg
Look at the bottom left... This and the main 'ARM' chip are the only ones Apple stamped with their own name, too.
PXA-300 is nearly as probable as the S3C2460 anyway, imo. We'll see though...
EDIT: Of course that might be something else than a coprocessor, but if so, I don't really see what.
nVidia's specification list for GoForce details acceleration for only transformations, and some people, surprised by that fact, sought confirmation on the issue.
http://forum.beyond3d.com/showpost.php?p=332892&postcount=18
http://forum.beyond3d.com/showpost.php?p=332892&postcount=18
Heh, yuck. Quickly googled around a bit, and indeed, it can only do transforms, and quite hackily so at that. I guess they bothered with that because many ARMs didn't/don't have FPUs, so they didn't really have a choice. Here's the patent kinda-sorta describing their transform & setup hardware: http://www.patentstorm.us/patents/7106336-description.html - and yes, before someone like Simon F chimes in, I agree that this is a completely ridiculous and obvious thing to patent.
And there's also a presentation on the SC10 (presumably GoForce 4500?) that includes the AR10, including transistor count figures and a coarse die shot: http://www.hotchips.org/archives/hc16/2_Mon/8_HC16_Sess1_Pres2_bw.pdf
The most terrifying part? As far as I can tell, there is no Early-Z. Only a command to set a kill bit based on a test, that's used for both alpha tests and depth tests. Interetingly, this patent claims to save power that way by disabling the ALUs etc. when this bit is set. The processing is not any faster as without Early-Z, however...
http://www.patentstorm.us/patents/7091982-description.html
Kinda shitty to say the least, not sure how it compares in terms of transistor counts...
And there's also a presentation on the SC10 (presumably GoForce 4500?) that includes the AR10, including transistor count figures and a coarse die shot: http://www.hotchips.org/archives/hc16/2_Mon/8_HC16_Sess1_Pres2_bw.pdf
The most terrifying part? As far as I can tell, there is no Early-Z. Only a command to set a kill bit based on a test, that's used for both alpha tests and depth tests. Interetingly, this patent claims to save power that way by disabling the ALUs etc. when this bit is set. The processing is not any faster as without Early-Z, however...
http://www.patentstorm.us/patents/7091982-description.html
Kinda shitty to say the least, not sure how it compares in terms of transistor counts...
Similar threads
