Can iPad Pro out-game an XB360? *spawn
- Thread starter wco81
- Start date
It's in their blood to spin things for their own benefit.
The performance they claim probably for specific stuff and only at the most ideal condition.
The iPad Pro graphics power stomps all over the XB360.
In fact, in GfxBench Manhattan and T-Rex, it performs just under an aftermarket HD7770, which is roughly equivalent to the XBOne. (Fewer CUs, compensated by higher clocks and bandwidth.)
Depending on where you look in the synthetic test, the iPad Pro GPU go from higher (alpha blend) to just under half (fill) so the aggregate game test scores make sense.
Link
You guys need to keep up.
Preemptive Edit: Yes, that HD7770 is under OSX. Feel free to choose whatever other card of your liking. Also, on-sreen results are useless in general and for the iPad Pro in particular since it pushes just under 6MP. (!)
In fact, in GfxBench Manhattan and T-Rex, it performs just under an aftermarket HD7770, which is roughly equivalent to the XBOne. (Fewer CUs, compensated by higher clocks and bandwidth.)
Depending on where you look in the synthetic test, the iPad Pro GPU go from higher (alpha blend) to just under half (fill) so the aggregate game test scores make sense.
Link
You guys need to keep up.
Preemptive Edit: Yes, that HD7770 is under OSX. Feel free to choose whatever other card of your liking. Also, on-sreen results are useless in general and for the iPad Pro in particular since it pushes just under 6MP. (!)
Last edited:
Not if you don't hamstring the 7770's performance by looking at OSX OGL numbers. The DX numbers on Windows remain dramatically faster than the iPad Pro.
My preemptive edit was too slow.
Yes, running under OSX reduces the game test scores a bit. I couldn't find them, last time I looked, I found scores from a factory overclocked ASUS HD7770, and linked the comparison to B3D, but when I checked now it was missing scores. Hate how the GfxBench database works.
Doesn't change the conclusion one whit though. The iPad Pro massacres the XB360. Sometimes I just tire of the knee jerk mud slinging.
My preemptive edit was too slow.
Yes, running under OSX reduces the game test scores a bit. I couldn't find them, last time I looked, I found scores from a factory overclocked ASUS HD7770, and linked the comparison to B3D, but when I checked now it was missing scores. Hate how the GfxBench database works.
Doesn't change the conclusion one whit though. The iPad Pro massacres the XB360. Sometimes I just tire of the knee jerk mud slinging.
But shouldn't it ? The xbox 360 was released in 2005 and cost $300 bucks. The ipad pro released in 2015 and cost $1000. So it took 10 years and more than triple the price to massacre the xbox 360 performance. Even then how long would the ipad pro massacre it before it was forced to throttle performance due to limited cooling.
If we use the same scale it may take until 2023 for a mobile device to massacre the xbox one or ps4.
Aren't there huge bandwidth constraints in mobile chipsets throttling their performance as well? Even once they may reach FLOP parity.
yes , every part of the device is meant to throttle.
The ipad performance is impressive of course , but its not like the xbox 360 is new. A decade is a lot of time for advancement
No doubt the iPad pro does have a much better GPU than the 360, but the 360 can run flat out on both CPU and GPU indefinitely. Except for perhaps the launch 360s, which am died.
360S was actually tested to handle a power virus on both CPU and GPU without hitting any thermal issues. I do wonder how much of a performance advantage the iPad Pro (or any tablet) would maintain if a game was hammering down on all aspects of the system for a good amount of time. Even actively cooled i5 tablets are prone to throttling heavily if you really task them.
Dunno if the right place for this discussion but the technical details are worth it to read about.
It can get warm I'm willing to bet. Unreal engine based games tend to warm any mobile device, but they appear to play them indefinitely as long as they are plugged in. Then again I doubt the CPU is being tasked much comparatively to the GPU.
The iPad pro must be missing bandwidth though right? Hat should be a major bottleneck.
IPad Pro has a 128-bit interface to LPDDR4 running at 3200MHz interface or 51.2GB/s, as compared to the XB360 22.4GB/s. XBox1 is at 68.2 GB/s.
Additionally, in terms of bandwidth saving technologies, the PowerVR GPU is in a good place.
Reviews (example) fail to demonstrate throttling on the iPad Pro. Which is amazing.
incredible amount of bandwidth for what I assume is very little memory? <4GB?
D
Deleted member 13524
Guest
Unfortunately, they're not really comparable. GFXBench for Android makes use of FP16 for which the Rogue 7 series have a large portion of dedicated hardware, whereas the Windows version only uses FP32 shader effects. There's a substantial difference in shader output (Tegra X1's Maxwell 2.5 does twice the FP16 throughput than it does with FP32, using the same ALU resources) and in required memory bandwidth.
As impressive as the A9X performs for a tablet chip, its actual performance should be way behind the 7770 - and the XBone, for that matter.
steveOrino
Regular
It has 4GB of RAM.
You are ignoring the eDRAM/eSRAM in Xbox 360 and Xbox One. Xbox 360 render target writes cost zero memory bandwidth because RT is always in eDRAM. PowerVR tiling gives similar RT bandwidth gains. Xbox One eSRAM on the other hand is fully flexible read&write scratchpad. You can keep data in there and read it back with zero memory bandwidth used, and you can use it from compute shaders as well. It is a big advantage for Xbox One and saves significantly more bandwidth than tiling when used correctly (assuming modern rendering pipeline with complex post effects).
Tiling hardware should be more energy efficient and use considerably less die space compared to a large scratchpad (or a big L4 cache in Intel designs). The tiling buffers are small, allowing them to be very close to the execution units. Moving data for long distances is one of the biggest power consumers on modern chips. So I don't doubt that iPad's perf/watt is better compared to Xbox One or Intel's EDRAM designs (in rendering algorithms that are compatible with tiling).
Last edited:
iMacmatician
Regular
The existing iPad Pro 12.9" has 4 GB RAM, but it looks like the new iPad Pro 9.7" has 2 GB RAM.
I'm surprised, since I assumed that 4 GB would be a kind of "baseline" for the iPad Pro series, but then again Apple generally does not put a lot of RAM in the iDevices.
I'm surprised, since I assumed that 4 GB would be a kind of "baseline" for the iPad Pro series, but then again Apple generally does not put a lot of RAM in the iDevices.
On the other hand, there's little question that the iPad Pro's CPUs would be ahead of XBox 360's in performance with both fully loaded. For most code anyway, or the kind that isn't extremely cache/prefetch friendly, branch-free, highly data regular vector code. XBox 360 has 50% more cores and 42% more clock speed but I'm pretty confident that Twister typically offers way more than 2.1x the perf/MHz of Xenon.
So I guess the question I have is not if iPad Pro can run full tilt w/o throttling but if it can deliver consistently better than XBox 360 results. When the comparison is against real world games that were both CPU and GPU heavy on the XBox 360. And I think the answer is probably yes.
Only where there's really a strong case for using FP32 over FP16 though. I don't know how representative GFXBench is in this regard, but it's one datapoint where a developer made decisions for FP16 design that they find reasonable.
I ignored it, because I really had no idea how to weigh the different bandwidth saving approaches against each other as well as their overall impact on total performance. Still don't really even if your response left me somewhat wiser. I guess it also somewhat depends on how well the game targets the wrinkles of the particular architecture.
Overall though, it still seems to me that the A9x falls somewhere in the middle between the XB360 and the XB1.
Similar threads
