IPad 2: GPU Benchmarked with Impressive Gains | Hacker News Reader

IPad 2: GPU Benchmarked with Impressive Gains

macrumors.com

45 pointstopgeek23 comments

To be fair, the Xoom has 1.3x as many pixels as the iPad. The iPad 2 is still faster taking this into account, but the graph is misleading.

True, but those are empty pixels.

Misleading how? It's an FPS graph between devices. Do you want to scale the Xoom score? It will still be more than doubled by the iPad 2.

Check out this graph with more devices:

http://images.anandtech.com/graphs/graph4191/35571.png

Tegra2 phones do better than Tegra2 tablets and the iPhone4 does better than the iPad 1. Look how well the iPhone 3GS does due to its lower resolution.

In short, higher resolution screens are penalised by this benchmark.

Uhm, yeah. But how is that relevant? The Xoom actually has that many pixels. The whole package matters.

So your takeaway from this is that iPhone 3GS has a better GPU than the iPad? Or that it's a better "package" than the iPhone 4?

It's a benchmark. If you don't know what it's actually measuring then you might as well not bother.

My takeaway is that you can expect better graphics performance on an iPhone 3GS than on an iPad 1 or iPhone 4, yes. If you are not designing tablets and deciding what GPU to put in there, comparisons between GPUs are completely meaningless. You have to look at the whole device.

Looking at GPUs in isolation in tablets is about as meaningless as looking at their batteries in isolation. How many watt-hours they have is only a poor proxy for their runtime.

Well, I have to admit I'm surprised that you're saying that the iPhone 4 has worse "graphics performance" than such classic phones as the Nokia N900, the famously GPU-poor Nexus One, the venerable Motorola Droid and the cheapo Optimus One.

And I'm not even bothering to mention any device in the top half of the table that all beat even the amazing 3GS. But if you're that determined to argue that the iPhone 4 and iPad as a package suck at "graphics performance" to the extent shown on that link I provided then who am I to argue with such a revelatory benchmark. Those Android guys must really have their act together when it comes to holistic design.

The only alternative that I see is that the benchmark is rubbish.

Isn't that a good example of "less is more"?

I wonder if people would detect more the difference in framerate it brings, compared to detecting the difference in resolution on this kind of size of screen.

While this is one important aspect, I think what's interesting is not so much the GPU and what it can perform, but what it brings.

The iPad2 may very well have a better, faster and more available "general purpose" GPU and I'm willing to accept that.

However, when looking at tablet demos, the iPad2 seems flat and dull and the UI entirely 2d. Honeycomb on the Xoom and Galaxy Tab 10.1 looks awesome, spacious and full of action.

If the Honeycomb experience looks better, more exciting, more advanced and visually appealing, do I honestly care about the GPU specs themselves when they bring nothing to the platform?

Note: I do understand that a more powerful GPU allows more sophisticated 3D games graphics, but I just don't care. It's not that important to me, just like I don't care about 3d GPUs on PCs.

As long as its powerful enough to drive the general user-experience I'm more than happy. Everyone may not be like me, but I suspect most are.

It appears that Tegra 2 is missing the ARM NEON SIMD engine (http://www.anandtech.com/show/4144/lg-optimus-2x-nvidia-tegr...) which the iPad 2's A5 chip supposedly has (someone please confirm this report: http://pdadb.net/index.php?m=cpu&id=a50000&c=samsung...). This would explain the benchmark discrepancy. There is another benchmark (http://www.pcmag.com/article2/0,2817,2381767,00.asp) that shows that A5 is 3x faster at matrix operations than Tegra 2, while being 1.5x slower executing code that likely does not involve matrix/vector operations.

Yes, Tegra 2 is missing NEON. I can't comprehend why people still release CPUs intended for any kind of performance yet omit the SIMD engine -- my suspicion is nVidia is trying to encourage developers to use their GPU APIs instead by crippling the ability of the CPU to do number crunching.

I highly doubt it's an overly significant power or die size issue -- with the amount they're already spending on L1 cache, L2 cache, and so forth, it seems dumb to omit the kind of powerful ALU necessary to make use of it all.

Is there a GPGPU API available on typical Tegra2 devices? OpenCL, CUDA? Without that, using the GPU for number crunching basically isn't going to happen.

Tegra 2 doesn't support OpenCL or CUDA, though Tegra 3 will. The PowerVR SGX543 supports OpenCL, though not, of course, CUDA.

EDIT: At the time, NVidia claimed that it _was_ a die size limitation.

Lack of NEON can't explain these benchmark results. This is GPU-bound stuff. The PowerVR SGX 543 core is simply a better GPU than NVIDIA's Tegra 2 GPU, and the A5 has two of them.

What's even more impressive is that Imagination Technologies claims this design can scale performance linearly up to 16 cores, for eight times the performance of the iPad 2 with no software changes. That headroom will come in handy for the inevitable retina display iPad.

I don't know if there is much to deduce from these benchmarks. They are good for bragging rights, that's for sure. But that's about it.

Note also that while the Xoom is a new product, its Tegra 2 processor is relatively old. NVIDIA first demoed it in January 2010. So here they're really comparing apples to oranges.

In any case, good for Apple. The iPad 2 does look like a good product.

Take a look at the screenshots from Infinity Blade on the iPad 1 vs. the iPad 2. There's a big difference in graphical quality. Most notably, the iPad 2 can handle antialiasing, more detailed geometry, and much more complex lighting. From these numbers, it looks like if the iPad 3 included a double-resolution display but no further GPU speed increases, it would still outperform the iPad 1.

I think I saw ambient occlusion done for the iPad2 vs. none on the iPad. I might be wrong though.

Real Racing HD 2 shows basically the same benefits in their iPad 2 optimized version.

I agree that comparing the GPUs is a little unfair, but the iPad and Xoom were released within a several weeks of each other, so I think it's entirely fair to compare system graphics performance.

Things like this show that we need a change of mindset when evaluating the technical capabilities of embedded devices instead of PCs. When a new graphics chip is released, you can't swap it for your old GPU, and it won't show up as a build-to-order option in 3 weeks, either. You have to wait for it to be integrated in to new SoCs.

Trying to excuse the Xoom's obsolete GPU (particularly when the Xoom's biggest selling point is the higher resolution screen) makes no more sense than trying to compare a current product to the announced but not shipping Cortex A15 processor.

Perhaps you're right. Personally, I would be more interested in comparing actual finished apps, as opposed to meaningless benchmarks.

Reply on news.ycombinator.com