Implications of SGX543 in iPhone/Pod/Pad?

[Exophase]

In certain ALU limited cases an SGX543 core can be 4 times faster than an SGX535. They could be claiming a 9x performance improvement with only an SGX543MP2 and a modest clock improvement. To try to balance real world results we might be looking at more than a modest clock improvement. But I wouldn't bet on an MP4 just yet.

The CPU quote is actually pretty surprising, I'd say this goes beyond conservative. Surely the clock speed won't be lower and they're not undercutting the ability for anything to scale close to perfectly for 2 cores. I hope there isn't some other bottleneck (L2 latency?)

 

[RDGoodla]

In certain ALU limited cases an SGX543 core can be 4 times faster than an SGX535. They could be claiming a 9x performance improvement with only an SGX543MP2 and a modest clock improvement. To try to balance real world results we might be looking at more than a modest clock improvement. But I wouldn't bet on an MP4 just yet.

The CPU quote is actually pretty surprising, I'd say this goes beyond conservative. Surely the clock speed won't be lower and they're not undercutting the ability for anything to scale close to perfectly for 2 cores. I hope there isn't some other bottleneck (L2 latency?)

Maybe dual-core A8 is used ? Since there is no official statement about dual-core A9.

 

[Exophase]

ARM doesn't offer multi-core for Cortex-A8. You can't just slap two on an SoC to make it work either, since they wouldn't be cache coherent (and you probably wouldn't want to use independent L2 caches on a dual-core ARM).

Otherwise, I can't think of any good reason why they would use A8 over A9. In perf/W or perf/size there's just no benefit, and I doubt that A9 licenses cost enough less to matter.

Is there actually an official statement that it's dual core? It'd be funny if they were using a single core A9 with a much higher clock speed. And weird, pretty weird.

 

[Entropy]

In certain ALU limited cases an SGX543 core can be 4 times faster than an SGX535. They could be claiming a 9x performance improvement with only an SGX543MP2 and a modest clock improvement. To try to balance real world results we might be looking at more than a modest clock improvement. But I wouldn't bet on an MP4 just yet.

The CPU quote is actually pretty surprising, I'd say this goes beyond conservative. Surely the clock speed won't be lower and they're not undercutting the ability for anything to scale close to perfectly for 2 cores. I hope there isn't some other bottleneck (L2 latency?)

Actually, if we assume that they are using a similar process to the A4, somewhat lowered clock speeds are quite possible, since they say that power draw and battery lives are unchanged. It would also fit the 2x CPU statement. And with lowered clock speeds, the graphics performance improvement could refer to SGXMP4.
(Or not.)
Yes, the slide that says "up to 2x faster CPU" also says "dual core processors" on the top line. So if you're extremely optimistic, you could interpret that as 2x2 CPU performance.

 

[Lazy8s]

No, definitely not dual A8s. That would be both awkward and would require more design work. Dual A9s.

Remember, the 3GS was also billed as being up to twice as fast as the previous generations, and the upgrade from a Lite MBX+VGP + ARM11 to an SGX535 + A8 was also quite a bit more.

 

[Ailuros]

In certain ALU limited cases an SGX543 core can be 4 times faster than an SGX535. They could be claiming a 9x performance improvement with only an SGX543MP2 and a modest clock improvement. To try to balance real world results we might be looking at more than a modest clock improvement. But I wouldn't bet on an MP4 just yet.

Same clock each 543 is 4.5x the arithmetic throughput than a 535. The whole thing times 2 = 9x. If 535 is clocked at 200MHz in the original iPad, it might be an indication of the frequency of the MP2 in A5.

The CPU quote is actually pretty surprising, I'd say this goes beyond conservative. Surely the clock speed won't be lower and they're not undercutting the ability for anything to scale close to perfectly for 2 cores. I hope there isn't some other bottleneck (L2 latency?)

Marketing ramble; above it's way exaggerated and here it's too humble.

 

[metafor]

Actually, if we assume that they are using a similar process to the A4, somewhat lowered clock speeds are quite possible, since they say that power draw and battery lives are unchanged. It would also fit the 2x CPU statement. And with lowered clock speeds, the graphics performance improvement could refer to SGXMP4.
(Or not.)
Yes, the slide that says "up to 2x faster CPU" also says "dual core processors" on the top line. So if you're extremely optimistic, you could interpret that as 2x2 CPU performance.

Well, one thing to keep in mind is that all that "same power draw" means is that the device as a whole consumes as much power (or close enough) to the original iPad. That says little other than that the AP power didn't balloon to 2x or something.

I'd wager on dual-A9's at 1GHz although there's always a chance the PA Semi team finished whatever they were working on.

 

[Deleted member 13524]

Maybe a dual A5 @ ~1.4GHz w\ NEON?
Or dual A9 @ 800MHz?
Both could do 2x A4 CPU performance.

Regarding the GPU, no way it's a SGX543MP4. That would be "at least" some 10x faster, not "up to 9x".
"Up to" 9x performance can be achieved with a MP2, with just a slight clock increase.


It'll still wipe the floor with OMAP4, Orion and Tegra 2 in gaming, nonetheless.

 

[Ailuros]

Regarding the GPU, no way it's a SGX543MP4. That would be "at least" some 10x faster, not "up to 9x".
"Up to" 9x performance can be achieved with a MP2, with just a slight clock increase.

A MP4 would have 18x times more GFLOPs than a single 535 at the same frequency. The 9x times is correct (as they're obviously quoting FLOPs) at the same frequency.

It'll still wipe the floor with OMAP4, Orion and Tegra 2 in gaming, nonetheless.

Depends on which OMAP4 you actually mean; the bigger the 4xxx number the smaller the wipe.

 

[Deleted member 13524]

So it's safe to say the A5 is half NGP.

 

[Entropy]

A MP4 would have 18x times more GFLOPs than a single 535 at the same frequency. The 9x times is correct (as they're obviously quoting FLOPs) at the same frequency.

So it seems the Beyond3D consensus base data for the A5 would be dual A9 and SGX543MP2, confirming the rumors.

Well, one thing to keep in mind is that all that "same power draw" means is that the device as a whole consumes as much power (or close enough) to the original iPad.

I don't think this is correct.
The slide presenting the A5 chip explicitly states "same low power as A4". And Steve Jobs says the same thing: "We go all out on the graphics performance. Up to 9 times faster. Same low power as A4. This will be the first dual core tablet to ship in volume."

Edit: 1GHz confirmed by Apple in their comparison material vs. the previous iPad.

 

[metafor]

I don't think this is correct.
The slide presenting the A5 chip explicitly states "same low power as A4". And Steve Jobs says the same thing: "We go all out on the graphics performance. Up to 9 times faster. Same low power as A4. This will be the first dual core tablet to ship in volume."

Edit: 1GHz confirmed by Apple in their comparison material vs. the previous iPad.

In market speak, as long as it isn't 2x the A4's power draw, it would be "the same low power". I've reiterated my thoughts on how much impact AP power has on total device power, especially on a tablet. I don't think having the same battery life as the first iPad really indicates anything about the SoC power other than it isn't something absurd like 4W.

 

[Entropy]

In market speak, as long as it isn't 2x the A4's power draw, it would be "the same low power". I've reiterated my thoughts on how much impact AP power has on total device power, especially on a tablet. I don't think having the same battery life as the first iPad really indicates anything about the SoC power other than it isn't something absurd like 4W.

Well, you're obviously right about the battery life and total device power, but as I said the slide referred to the chip specifically. I doubt we'll ever see accurate independent power draw figures for the chip, so either we take Apples word for it or not. I don't see why they wouldn't have phrased it like "the iPad2 draws the same power as its predecessor" if that was indeed what they intended. In practical terms it makes no difference whatsoever to the user, so specifically referring to the chip brings no benefit from a marketing point of view, and that Apple and Steve Jobs personally would lie under such circumstances seems unlikely to me.

 

[Pressure]

It is dual-core at 1 Ghz, clearly says so on the iPad specifications page

1GHz dual-core Apple A5 custom-designed, high-performance, low-power system-on-a-chip

 

[wco81]

So whatever GPU the A5 has, it would be able to drive the Retina Display (4X the current iPad resolution)?

Speaking of which, Apple has once again shown a bias for thinner form factors and supposedly the type of screen allowed them to shave some depth from the iPad 2.

Would a higher-resolution display necessarily require a bigger form factor?

 

[metafor]

Well, you're obviously right about the battery life and total device power, but as I said the slide referred to the chip specifically. I doubt we'll ever see accurate independent power draw figures for the chip, so either we take Apples word for it or not. I don't see why they wouldn't have phrased it like "the iPad2 draws the same power as its predecessor" if that was indeed what they intended. In practical terms it makes no difference whatsoever to the user, so specifically referring to the chip brings no benefit from a marketing point of view, and that Apple and Steve Jobs personally would lie under such circumstances seems unlikely to me.

Apple likes to point out anything they make. I mean, the A4 wasn't exactly a performance beast; most people wouldn't even notice it compared to any other Cortex A8 device (including the iPhone 3Gs) and yet they announced it with the first iPad. I believe the words were "it screams!".

 

[Xmas]

Would a higher-resolution display necessarily require a bigger form factor?

No.

 

[Ailuros]

So it's safe to say the A5 is half NGP.

Not really and that from many aspects.

 

[Deleted member 13524]

Not really and that from many aspects.

Why not?

NGP: 4*A9@1GHz + SGX543MP4
A5: 2*A9@1GHz + SGX543MP2

We're only missing memory bandwidth here, but everything seems pretty much doubled-up.

 

[Ailuros]

Why not?

NGP: 4*A9@1GHz + SGX543MP4
A5: 2*A9@1GHz + SGX543MP2

We're only missing memory bandwidth here, but everything seems pretty much doubled-up.

SONY states MP4+ for one and on the other there's still the low level API advantage of a "closed environment" of a hand-held console amongst others.