[suryad]

I highly doubt the specs will be anywhere near as what you guys are talking about especially with the A15 processor.

[ToTTenTranz]

Fact:
- If this Galaxy S III does come with a quad Cortex A9 Exynos 4412, it won't be the fastest SoC of this generation, unless it clocks up to something like 2GHz to counter Krait's higher IPC in Snapdragon S4.


BTW, is LPDDR3 a possibility? In a dual-channel configuration, it'd give them the same bandwidth as A5X but for the whole system (not just GPU).

[french toast]

Originally Posted by ToTTenTranz

Fact:
- If this Galaxy S III does come with a quad Cortex A9 Exynos 4412, it won't be the fastest SoC of this generation, unless it clocks up to something like 2GHz to counter Krait's higher IPC in Snapdragon S4.


BTW, is LPDDR3 a possibility? In a dual-channel configuration, it'd give them the same bandwidth as A5X but for the whole system (not just GPU).

I don't think the controller supports LPDDR3 on Exynos 4 series..it does on 5 series, besides i still think its early for that is it?

Wrong, overall the 4412 will likey destroy S4, both in graphics and cpu, not on everything of course, vellamo will still likely be a snapdragon winner, as will Linpack...everything else will be dominated by Exynos.

Dont forget Exynos will have some 50% better graphics..
Exynos will likely be clocked higher than Tegra 3, it will also have dual channel memory, maybe better nand performance as well (samsung does make that)

EDIT; How do you work out bandwidth? can't find the calculation on the net.
would like to find out the bandwidth of LPDDR2 1066 x2
I thought A5X did allow all bandwidth for system, just the GPU had priority?

[ToTTenTranz]

Originally Posted by french toast

EDIT; How do you work out bandwidth? can't find the calculation on the net.
would like to find out the bandwidth of LPDDR2 1066 x2
I thought A5X did allow all bandwidth for system, just the GPU had priority?

Max theoretical bandwidth in MB/s is:

DR x ChW x nCh x (1/8)

DR - Data Rate, which, in the case of DDR (double data rate) memory it's 2 x clockspeed (MHz), so it's usually measured in "Mega-Transfers per second" (MT/s). For example, LPDDR2 1066 does 1066MT/s, although it runs at 533MHz.

ChW - Channel Width. For x86 solutions, it's usually 64 bit. For ARM SoCs it's usually 32bit.

nCh - Number of Channels. For example, dual-channel in OMAP4, single-channel in Tegra 2 and Tegra 3.

(1/8) - To turn bit into byte. 1 byte = 8 bit


For a dual channel LPDDR2 1066 ARM SoC implementation we have:
1066MT/s x 32 x 2 x (1/8) = 8528MB/s

The A5X has four 32bit channels at 800MT/s, so it's:
800 x 32 x 4 x (1/8) = 12800MB/s

But it's speculated that only the 543MP4 GPU has access to the four channels, and the CPUs can only access two channels, so while the GPU gets 12,8GB/s max theoretical bandwidth, the CPU only gets 6.4GB/s max. Bandwidth for the ipad 3's graphics subsystem is of the most importance because of the screen's very high resolution.

However, I often see that having many channels of slower memory results in higher latencies and lower performance than fewer channels of faster memory. It's probably the reason why Tegra 3 (single-channel 1066MT/s) does better in synthetic memory benchmarks than OMAP4 (dual-channel 800MT/s)

If Samsung implements a dual-channel LPDDR3 controller at 1600MHz, it would get the same theoretical bandwidth as the A5X has for the GPU, with the difference that it could get better results for using less channels for the same max theoretical bandwidth.

[french toast]

Thanks i got all of that except the last byte=bit? so would that mean depending on the amount of ram ie 512 megabyte or 1gb would affect the bandwidth? so 512MB/8= (1/8)bit to calculate bandwidth?

EDIT; Also, what is the main difference between LPDDR,LPDDR2,LPDDR3? (low power double data rate) x2 x3? or not multiplication anything, just signifies newer version?

[ToTTenTranz]

The bit is the smallest digital form of data. It can be either "0" or "1.
A byte is a set of 8 bits.

Bits are more often used for hardware architecture descriptions like addressing, channel width and others.
Bytes are more often used to describe the size of a digital storage.

But it's equally correct to say that 512MB of RAM is the same as having 512*8 = 4Gbits of RAM or the other way around.

The difference between using bits/bytes for a memory channel width or memory storage (volatile or not) is.. erm..
Imagine you have a building with 1024 people inside. You want to take those people to another building, and all you have is a bus /where you can fit 32 people at a time. The people you fit inside the bus is your memory channel width spec. The people you have inside the building is the memory storage spec.
Furthermore, you could say that the speed at which the bus goes and returns is the "data rate".
So for emptying the building you'd need 1024/32 = 32 travels on that single bus, and how fast you'd do it depends on how fast the bus is.

So the speed at which you can move people around depends on:
- How fast your bus is (data rate, clock speeds)
- How many people you can fit inside your bus (channel width)
- How many buses you have (number of channels)

The amount of RAM per se isn't an indicator of number of channels.. You could have 1GB RAM through a single 32bit channel or 256MB RAM through four 32bit channels, for example.

[french toast]

Thanks thats a very good description! right, the conversion either into byte or inversely bit is by a factor of 8 either way..now that explians what everyone was on about 4Gbit (512mb) memory sticks

I now clicked on you only have to divide by 8 the whole score to find the badwidth, no need to divide the amount of memory by 8 in a seperate equation..cool.

So when Apple moves to A6X in theory they could have..using LPDDR3 1500MT/S *32*4/8= 24000MB/S thats massive, however i must have got the ram speed wrong as LPDDR3 dual channel you said 12.8, 12.8*2=25.6GB/S??

EDIT; Its 1600MT/S.cool.

[ltcommander.data]

Originally Posted by ToTTenTranz

However, I often see that having many channels of slower memory results in higher latencies and lower performance than fewer channels of faster memory. It's probably the reason why Tegra 3 (single-channel 1066MT/s) does better in synthetic memory benchmarks than OMAP4 (dual-channel 800MT/s)

If Samsung implements a dual-channel LPDDR3 controller at 1600MHz, it would get the same theoretical bandwidth as the A5X has for the GPU, with the difference that it could get better results for using less channels for the same max theoretical bandwidth.

Don't successive memory generations tend to have higher latencies at the same frequencies? Perhaps the increase in latency going from LPDDR2 to LPDDR3 will offset the decrease in latency going from 4 channels to 2 channels?

Tegra 3's low single channel memory latency may also more be a consequence of nVidia's better design compared to the presumably stock ARM memory controller in OMAP4 than an inherent disadvantage of 2+ memory channels. I believe Intel also criticized the memory controllers found in ARM SoCs. Anand claimed Apple designs their own memory controllers, but it's hard to tell how good it is. I don't believe Apple's memory controller designs from the PowerPC days inspire confidence, but I guess those designers were let go in the Intel transition and they have a different design team now from their acquisitions. I wonder if Samsung does their own memory controller design as well?

[french toast]

Right, some good news; we have a video of the Galaxy S3 in the flesh!
http://blog.gadgethelpline.com/samsu...ideo-pictures/

The bad news is that it confirms our worst fears, that it has a Exynos 4412 clocked at a lowly 1.4ghz, with only 1gb ram and a Mali 400mp4..

It also contains like i said it would a 4.65 720P display, but doesn't confirm whether it is a RGB or a Pentile screen, it also has NFC which you would expect, i have to say the overall speed of the UI doesn't seem to indicate any extra speed than the Galaxy note, which also carries a 1.4ghz clock speed.

The plus point is that with 4 cores and dual channel memory this beast will multitask as good or better than any other phone currently out there.
It carries a large 2000mah battery along with 4 cores to spread the load around and lower voltage, as well as being made on that advanced 32nm gate first HKMG process....2 days battery life is going to be guarenteed.

As for the design of the handset it looks completely ugly as hell, like a big brick, so ugly in fact that it must be a fake generic shell to hide that real casing which i expect to be Ceramic/alloy unibody, and with that battery come in around 8mm thick.

I was kinda hoping that we would see clock speeds hit 1.6ghz and get another 1gb ram, but twas not to be, along with the YOUM bezzle less display

I expect the build quality of this thing to be awesome, thin as hell best batterylife of anymajor smartphone outside of the Razr maxx, as well as best gaming performance and best multitasking.

Single threaded performance is going to really lag a Snapdragon S4, you can see in the video applications take a second to load, could be early software, likely to be the real deal though.

If that is not a new type of AMOLED with Gorrilla glass 2 and RGB then i wont be spending money on it me thinks, will wait till WP8.

[french toast]

It also carries an 8mp camera, its all a bit underwhelming too be honest, all this secrecy for what appears to be the expected update, i still think they have something up their sleeve, Samsung has access to many technologies this year, it would seem a bit odd if none of them found a way into this device, even the camera is the same one from the Galalxy S2 bizzare to say the least.

[Mariner]

When a sensor and lens of the camera is as small as those found in a standard smartphone, I can't really see the point in having anything higher than 8 mpix. The megapixel race in normal point and shoot digital cameras hasn't improved image quality in recent years - some older models with lower resolution sensors are still much in demand as they have better low-light performance than their successors.

I'd even go as far as saying anything higher than 5 mpix is mostly unnecessary - such a resolution is good enough for A3 prints and how often does anybody print images this large from a smartphone?

[french toast]

Originally Posted by Mariner

When a sensor and lens of the camera is as small as those found in a standard smartphone, I can't really see the point in having anything higher than 8 mpix. The megapixel race in normal point and shoot digital cameras hasn't improved image quality in recent years - some older models with lower resolution sensors are still much in demand as they have better low-light performance than their successors.

I'd even go as far as saying anything higher than 5 mpix is mostly unnecessary - such a resolution is good enough for A3 prints and how often does anybody print images this large from a smartphone?

When you get to 8 mega pixel yes, the most most important thing is the optics..but having a very sensor with a high quality lens has its advantages..did you check out the Nokia pure view camera? amazing.

[Mariner]

Yeah, I know the PureView is a different concept and this is why I mentioned the sensor and lens of a standard smartphone!

I see that the latest info is that the SGS III has a lens with a F2.6 aperture, so decent, but little improvement over the SGS II.

That said, although the HTC One X has a lens with an aperture of F2.0 (which is a lot bigger than F2.6), reports are that the picture quality itself isn't that great.

Just a pity that, by design, the PureView camera mechanism has to be quite so thick in comparison to modern smartphones.

[french toast]

Yea i agree, i still the best camera quality on a smartphone to date is the Nokia N8..that 12mp snapper still has the biggest sensor fitted to a phone i think.

Pureview is really going to destroy the competition, the best of the high end handsets seems to be the 4s 8mp snapper, this Samsung one doesn't look anything ground breaking, in fact im quite dissapointed by the leaks, it seems the only real innovation might be the yet to be shown casing, everything else is not going to be a big improvement day to day, if that screen turns out to be the very same Pentile diplay off the Nexus then this phone will be a massive let down IMHO.

[french toast]

Rumour has it that the New Samsung phone will contain eye tracking technology, and even a icloud copy called S cloud..offering the exact same features 5gb, it is sure to annoy Apple even more.

Other rumours point to a complete renaming of the phone...maybe Samsung imobile, or igalaxy.

[Nebuchadnezzar]

Everybody's going insane calling bullshit on the leak and how fake it is and how underwhelming it is; yet nobody is asking the obvious question: Is this even a Galaxy S3? My answer: no. The Samsung source code is full of "Midas" (Project M?) and S2Plus references, and that is what I think this leak was. The blogosphere and news sites all curious about how it still got a Mali400 instead of a T604 as rumoured, even though Samsung confirmed that themselves back in February. Journalism at its worst.

[french toast]

Maybe, but i dont think so, i think this is the GS3 with a fake outer shell to make it a suprise.
We clocked this phone in testing on GL benchmark a few weeks back with the exact same processor and screen resolution.

This backs up earlier rumours about the fake casing and the Cemaric/alloy unibody it will eventually come with.
That GTi9300 can't be a midrange phone, because of the high end specs, it doesnt look right, and they wouldn't go to that trouble of concealing a Galaxy s2 plus or HD, also having a GS2 hd with a quad core processor when the GS3 launches with a 5250 dual core would confuse the market.

I want to believe this is some kind of hoax, however this looks to be the real deal IMO.

[ToTTenTranz]

From the pictures, that seems to be the GTi9300, for which it's been said it's not the Galaxy S3.

Remember, on par with the regular Galaxy S2 they also came out with the Galaxy R with Tegra 2, a GTi9100G GS2 with an OMAP4, the t-mobile version with a Snapdragon S3..
If I'm not mistaken, there are also Galaxy S1 variants with Snapdragons and even OMAP3630..


Looking at that picture, I'd see no reason at all for Samsung to deliver testing devices with fake cases inside steel safes. It's as bland as it could be.

[Nuker]

Originally Posted by french toast

Right, some good news; we have a video of the Galaxy S3 in the flesh!
http://blog.gadgethelpline.com/samsu...ideo-pictures/

The bad news is that it confirms our worst fears, that it has a Exynos 4412 clocked at a lowly 1.4ghz, with only 1gb ram and a Mali 400mp4..

It also contains like i said it would a 4.65 720P display, but doesn't confirm whether it is a RGB or a Pentile screen, it also has NFC which you would expect, i have to say the overall speed of the UI doesn't seem to indicate any extra speed than the Galaxy note, which also carries a 1.4ghz clock speed.

The plus point is that with 4 cores and dual channel memory this beast will multitask as good or better than any other phone currently out there.
It carries a large 2000mah battery along with 4 cores to spread the load around and lower voltage, as well as being made on that advanced 32nm gate first HKMG process....2 days battery life is going to be guarenteed.

As for the design of the handset it looks completely ugly as hell, like a big brick, so ugly in fact that it must be a fake generic shell to hide that real casing which i expect to be Ceramic/alloy unibody, and with that battery come in around 8mm thick.

I was kinda hoping that we would see clock speeds hit 1.6ghz and get another 1gb ram, but twas not to be, along with the YOUM bezzle less display

I expect the build quality of this thing to be awesome, thin as hell best batterylife of anymajor smartphone outside of the Razr maxx, as well as best gaming performance and best multitasking.

Single threaded performance is going to really lag a Snapdragon S4, you can see in the video applications take a second to load, could be early software, likely to be the real deal though.

If that is not a new type of AMOLED with Gorrilla glass 2 and RGB then i wont be spending money on it me thinks, will wait till WP8.

But this is only an old prototype from december, and it is one of the earlier leaked images, to which Samsung responded that it would not resemble the S3. It may have been considerably upgraded since then (which I hope, I would not want a lowly Mali 400 in my next phone; the T604 is certainly prefered as it will blast the competition away for some 6 months possibly). I really hope it will have Exynos 5250, then everything will be set for me.
http://www.phonearena.com/news/Samsu...nboard_id29305
A further possibility is that device seen in the video is a counterfeit one, made from a fake Samsung in China.

[Exophase]

Originally Posted by ltcommander.data

Don't successive memory generations tend to have higher latencies at the same frequencies? Perhaps the increase in latency going from LPDDR2 to LPDDR3 will offset the decrease in latency going from 4 channels to 2 channels?

Latency stays about the same, improving slightly with successive process shrinks. Why would the latency get worse? I think people are just confused by the latency numbers, that are measured in clock cycles, going up as the clocks go up.

Originally Posted by ltcommander.data

Tegra 3's low single channel memory latency may also more be a consequence of nVidia's better design compared to the presumably stock ARM memory controller in OMAP4 than an inherent disadvantage of 2+ memory channels.

I think so too. Samsung also has much lower latencies than TI or Freescale.

Originally Posted by ltcommander.data

I believe Intel also criticized the memory controllers found in ARM SoCs.

What Intel criticized was the number of in-flight L2 transactions in Cortex-A9 SoCs vs Medfield, as the sole explanation for the much higher Sunspider score. I don't buy this for a minute, I think there are much more blatant explanations that have nothing to do with hardware design. I really don't think Intel's comments on competing hardware can be trusted.

[RaySoft]

My guess is that we will se a Quad Cortex A9 + Quad Mali-T604 SOC with 2GB RAM and a 4.7" 720p Super AMOLED HD display.

If you check out this roadmap from ARM, you will se that between 2012 and 2013 (around when S3 launches) they have a T604 with A9 SOC:


Source: http://www.anandtech.com/show/5077/a...r-than-mali400

[Dr Evil]

http://www.anandtech.com/show/5786/e...axy-smartphone

Just a week before its Samsung Mobile Unpacked 2012 event where it will announce "the next galaxy smartphone," Samsung has officially announced what SoC will be inside. The answer is Exynos 4 Quad, which is no doubt the Exynos 4412 we've heard about before. Exynos 4 Quad is built on Samsung's 32nm HKMG process and consists of 4 ARM Cortex A9s running at up to 1.4 GHz.

[wishiknew]

Thats kind of a downer in a way even if I had hope for something else that wasn't ready.

[wco81]

Does the Galaxy brand depend on pushing the specs to generate interest?

Or is the brand strong enough without having the latest and greatest?

[kyetech]

french toast better get me your paypal details ready. Looks like youve got 3 quid on the way.

lol. But obviously we will wait till may 3rd. Fat lady sings and all that