iPad 2

[ltcommander.data]

http://www.engadget.com/2011/03/09/ipad-2-review/
http://www.macworld.com/article/158439/2011/03/ipad2.html

Well Geekbench is reporting the iPad 2 has 512MB of RAM. And according to Macworld, the A5's system bus has doubled to 200MHz. Interestingly, Engadget said Geekbench is reporting the A5 at 800MHz. Double core Cortex A9 at 800MHz vs. single core Cortex A8 at 1GHz would be more consistent with Apple's 2x faster CPU claim. Although it does contradict the iPad 2 spec page which said the A5 is 1GHz. Perhaps the A5 is just downclocking or Geekbench needs to be updated.

 

[Ailuros]

Anand's take on iOS4.3 (first impressions): http://www.anandtech.com/show/4213/ios-43-review/1

 

[Ailuros]

I don't think they're giving up anything as far as margins. Probably higher margins on iPad2 than the Xoom enjoys at a higher price.

An analyst claimed before the iPad2 announcement that Apple may have a 7.5 to 9.75% cost advantage over other manufacturers because they're able to secure components on a much larger scale than any other single manufacturer:

http://www.marketwatch.com/story/apples-ipad-price-advantage-the-breakdown-2011-03-02

I think the more interesting way that iPad and other tablets become like consoles is if they are able to render games in 1080p and output via HDMI to HDTVs.

Sure you still have the touch screen controls problem but if people start hooking these things up regularly to their HDTVs, someone is going to get the idea to allow game controllers to pair by Bluetooth to actually control games.

That or eventually, tablets will get screens to display 1080p.

While consoles and handhelds remain stagnant in capabilities for years, mobile devices are going to improve every year (or more frequently).

They won't have the power of consoles but if they put out sharp 1080p graphics, tablets could start to take business away from home consoles, the way smart phones has intruded on the handheld market.

Did you actually read the article I linked to, or did you just halt at the article title? The author makes an attempt to analyze Apple's current business model and he makes it perfectly clear that Apple's current business model has quite a few similarities to console manufacturers and NOT that the iPad or any other i-device is in the strict sense a "gaming console" of any sorts.

ltcommander just posted a link in the other thread from eeTimes: http://www.eetimes.com/electronics-news/4213873/Apple--TSMC-to-expand-foundry-ties

''Our contacts expect 45 million iPads produced in 2011. Our contacts continue to expect 45 million iPad builds in calendar 2011, meaningfully ahead of the Street. When looking at the linearity of 1H11 (13 million units) and 2H11 (32 million units) iPad production, it seems Apple is planning on some very large iPad 2 sales volumes in 2H11 in order to achieve its annual target, with even cheaper price points likely necessary in order to achieve these 2H11 volumes,'' according to the report.

I'm still not sure if Apple's and analysts' forecasts are too optimistic or not. Time will tell eventually. However if those kind of volumes are true, it sounds like a reasonable assumption that prices will drop. Good news for the consumer and most likely bad news for Apple's competitors.

How do you think an AppleTV2 with A5 (maybe 30% higher clocked for 720p compared to the 1024x768 on the iPad 2) might compare to a Nintendo Wii? Apple could just bring the App Store to the AppleTV2 and introduce a reference Bluetooth input device/controller (3rd parties would be responsible for more specialized "hardcore" gaming controllers) . Add universal apps across all iOS plattforms (iPhone/iPad/AppleTV) and save games synced via GameCenter and you have a quite attractive gaming console/extender, at least for the casual user.

I haven't the vaguest idea what the GPU block is clocked at in A5, but assuming it's clocked somewhere in the 200MHz ballpark I don't see any absolute necessity to further increase the frequency for something like 720p. Apple considering to expand in that direction future AppleTV variants are at this time still just unverified rumors, yet it would be definitely interesting if they'd go in that direction. The major interest though in such a case would fall rather in the application store revenues direction than anything else.

 

[eastmen]

Sounds like a worth while purchase for someone without an ipad that wants an ipad , certianly worth the premium over the current ipad 1 prices.

For those with an ipad it doesn't seem like its worth putting money out for

 

[Pressure]

Sounds like a worth while purchase for someone without an ipad that wants an ipad , certianly worth the premium over the current ipad 1 prices.

For those with an ipad it doesn't seem like its worth putting money out for

Overall faster performance is actually a godsend. Let's be honest, the original iPad is quite slow and I would pay money to get the faster performance and better user experience (less waiting around for websites to load and less caching issues with 512 MB of RAM).

Side-by-side comparisons will probably show a clearer picture, like loading a heavy website (e.g. Engadget or similar).

Also I am very excited about the increased graphics performance.

 

[max-pain]

Geekbench

iPad2,3 - 743
http://browse.geekbench.ca/geekbench2/view/374870
Processor integer performance - 672
Processor floating point performance - 901
Memory performance - 808
Memory bandwidth performance - 317
Processor Frequency - 894 MHz
L2 Cache - 1.00 MB
Memory - 502 MB
FSB - 250 MHz

iPad1,1 - 449
http://browse.geekbench.ca/geekbench2/view/375950
Processor integer performance - 366
Processor floating point performance - 457
Memory performance - 645
Memory bandwidth performance - 320
Processor Frequency - 1.00 GHz
L2 Cache - 512 KB
Memory - 246 MB
FSB - 100.0 MHz

 

[Arwin]

That's a very modest improvement - no wonder they've been able to keep the base model at 499 (euro).

Does this test use multiple cores properly? Or is this basically just comparing a single core?

 

[Mize]

That's a very modest improvement - no wonder they've been able to keep the base model at 499 (euro).

Does this test use multiple cores properly? Or is this basically just comparing a single core?

??

Given the clock speeds it seems to scale about right...or am I missing something?

 

[Arwin]

??

Given the clock speeds it seems to scale about right...or am I missing something?

I could have just clicked the link. It's multi-threaded alright, and if I'm looking correct then from the looks of things single threaded performance is lower than the original iPad, right?

 

[Mize]

I could have just clicked the link. It's multi-threaded alright, and if I'm looking correct then from the looks of things single threaded performance is lower than the original iPad, right?

But it should be as the clock of the iPad2 is 10% lower.

 

[Arwin]

But it should be as the clock of the iPad2 is 10% lower.

Yeah, but that's not the point - you see on PCs that single-threaded applications on PC run slower on multi-core processors with lower clocks or core performance too. So it would be interesting to know what the typical behaviour is of applications on the iPad 2 vs the original iPad and see what circumstances give actual performance improvements, and/or if there are circumstances where the iPad 2 actually performs worse.

 

[Entropy]

There is some strangeness with the Geekbench results.
The FSB difference 250MHz vs. 100MHz doesn't seem reflected in the memory scores at all, so I'd assume that either or both has issues.
Looking at the memory results, there are instances where the iPad2 is is twice as fast (stdlib write) and instances where it is slower (stdlib copy, even though read speeds are the same). Weird.

Looking at FP, in Mandelbrot (single threaded) the iPad2 is 3-4 faster, primality is over twice as fast and it seems clear that the FPU has changed substantially. Is A9 VFP pipelining enough to explain the difference?

Both run iOS 4.3 so library calls should be equivalent.

But again, the FSB and memory benchmark data look odd.
(Also, the actual bandwidth numbers should help explain why I'm always grumbling about SoC system wide limitations).

 

[loekf]

There is some strangeness with the Geekbench results.
The FSB difference 250MHz vs. 100MHz doesn't seem reflected in the memory scores at all, so I'd assume that either or both has issues.
Looking at the memory results, there are instances where the iPad2 is is twice as fast (stdlib write) and instances where it is slower (stdlib copy, even though read speeds are the same). Weird.

Looking at FP, in Mandelbrot (single threaded) the iPad2 is 3-4 faster, primality is over twice as fast and it seems clear that the FPU has changed substantially. Is A9 VFP pipelining enough to explain the difference?

Both run iOS 4.3 so library calls should be equivalent.

But again, the FSB and memory benchmark data look odd.
(Also, the actual bandwidth numbers should help explain why I'm always grumbling about SoC system wide limitations).

Maybe an Cortex-A8 without NEON extensions in the A4 and a dual A9 with NEON extensions (and VFP) in the A5 ?

 

[rpg.314]

There is some strangeness with the Geekbench results.
The FSB difference 250MHz vs. 100MHz doesn't seem reflected in the memory scores at all, so I'd assume that either or both has issues.
Looking at the memory results, there are instances where the iPad2 is is twice as fast (stdlib write) and instances where it is slower (stdlib copy, even though read speeds are the same). Weird.

Looking at FP, in Mandelbrot (single threaded) the iPad2 is 3-4 faster, primality is over twice as fast and it seems clear that the FPU has changed substantially. Is A9 VFP pipelining enough to explain the difference?

Both run iOS 4.3 so library calls should be equivalent.

But again, the FSB and memory benchmark data look odd.
(Also, the actual bandwidth numbers should help explain why I'm always grumbling about SoC system wide limitations).

Does the notion of FSB even make sense for SoC like A5?

 

[Pressure]

Daring Fireball aka John Gruber did perhaps the most interesting review.

He asked a developer to code a program to quickly determine the performance of the 3D subsystem and the results are very nice:

For example, on my original iPad, with 200 on-screen sprites, the framerate dropped to 45 fps. On the iPad 2, with 400 on-screen sprites, the framerate remained at 65 fps. On the iPad 1, Guy’s demo app dropped below 60 fps with about 100 animated sprites; on the iPad 2, it didn’t drop below 60 fps until there were over 750 animated sprites.

After I showed him the results, Guy told me, “The results show that the iPad 2 is easily about twice as powerful as the original and that this speed gain is a freebie — you don’t need to change your code structure in order to see significant gains. The differences in the amount of time spent rendering indicates that the GPU is really much faster than the original. The original iPad had a comparatively weak fill-rate and it was an issue for the device. The second generation really leaves that behind and it looks like it’ll be able to do some really incredible things graphically. My demo code is workman-like, competent code — meant to measure the relative strengths of the parts of the system. Taking some time to get the most out of that GPU and CPU will pay off with some really remarkable games and graphics apps.”

 

[Mike11]

John Gruber ran a custom benchmark:

Looking for a better benchmark, I asked my friend Guy English, an iOS developer who has worked on games like Tap Tap Revenge (as a contractor for Tapulous and Disney), to write a custom test app to measure an iPad’s graphics capabilities from the perspective of a game developer. It’s a simple app that renders hundreds (or even a few thousand) sprites moving around on screen, with gravity, and tracking up to three touch points. The results show that the iPad 2’s graphics improvements far outshine its straightforward CPU improvements — exactly as Apple has advertised.

For example, on my original iPad, with 200 on-screen sprites, the framerate dropped to 45 fps. On the iPad 2, with 400 on-screen sprites, the framerate remained at 65 fps. On the iPad 1, Guy’s demo app dropped below 60 fps with about 100 animated sprites; on the iPad 2, it didn’t drop below 60 fps until there were over 750 animated sprites.

After I showed him the results, Guy told me, “The results show that the iPad 2 is easily about twice as powerful as the original and that this speed gain is a freebie — you don’t need to change your code structure in order to see significant gains. The differences in the amount of time spent rendering indicates that the GPU is really much faster than the original. The original iPad had a comparatively weak fill-rate and it was an issue for the device. The second generation really leaves that behind and it looks like it’ll be able to do some really incredible things graphically. My demo code is workman-like, competent code — meant to measure the relative strengths of the parts of the system. Taking some time to get the most out of that GPU and CPU will pay off with some really remarkable games and graphics apps.”

http://daringfireball.net/2011/03/the_ipad_2

 

[metafor]

There is some strangeness with the Geekbench results.
The FSB difference 250MHz vs. 100MHz doesn't seem reflected in the memory scores at all, so I'd assume that either or both has issues.
Looking at the memory results, there are instances where the iPad2 is is twice as fast (stdlib write) and instances where it is slower (stdlib copy, even though read speeds are the same). Weird.

What are the details of the test? It could be you're just running into caching issues.

Looking at FP, in Mandelbrot (single threaded) the iPad2 is 3-4 faster, primality is over twice as fast and it seems clear that the FPU has changed substantially. Is A9 VFP pipelining enough to explain the difference?

It depends on the particular instruction as well as load-use dependencies. A9 latencies (along with being pipelined) are less than half that of the A8 (FADD is 4 cycles vs 9 for SP). That can make a big difference. A9 also (IIRC) doesn't suffer the same penalties for subnormal numbers as the A8 does.

 

[metafor]

Does the notion of FSB even make sense for SoC like A5?

There is still a system bus (assuming they're like most other mobile SoC's) that connects the high-performance subsystems. That's usually fairly close if not synchronous to the memory speed. In a stock ARM system, I believe this bus also hosts the L2 cache as well.

 

[Gubbi]

I could have just clicked the link. It's multi-threaded alright, and if I'm looking correct then from the looks of things single threaded performance is lower than the original iPad, right?

The fact that two instances of a benchmark on the iPad 2 runs less than twice the speed of a single instance on the iPad just means that the iPad 2 is memory bottlenecked.

A 800MHz Cortex A9 would beat a 1GHz Cortex A8 hands down on real code (as in real iPad application code)

Cheers

 

[Exophase]

Maybe an Cortex-A8 without NEON extensions in the A4 and a dual A9 with NEON extensions (and VFP) in the A5 ?

A4 has NEON. You won't find a Cortex-A8 without NEON anyway, but I know for a fact A4 does - 3GS as well.

I could have just clicked the link. It's multi-threaded alright, and if I'm looking correct then from the looks of things single threaded performance is lower than the original iPad, right?

Look at more than just the very first benchmark. All the other integer numbers are decently to substantially higher, and this is with the iPad 2 allegedly at 900MHz vs 1GHz for the iPad. As indicated, the FPU results are a dead giveaway of it not being on a Cortex-A8.

There is some strangeness with the Geekbench results.
The FSB difference 250MHz vs. 100MHz doesn't seem reflected in the memory scores at all, so I'd assume that either or both has issues.
Looking at the memory results, there are instances where the iPad2 is is twice as fast (stdlib write) and instances where it is slower (stdlib copy, even though read speeds are the same). Weird.

Bus speed doesn't say anything about bus width: I'm pretty sure it's 64-bit in iPad, but could very well be 32-bit in iPad 2. nVidia seems to think 32-bit is enough for dual core Cortex-A9, afterall (or even quad-core)