Soundwave said:
What type of bandwidth do you think they would need for a 720-800p display?
|
25GB/s+ seems to be the general hardware consensus of acceptable 720P gaming.
However, that's just a raw number... We need to take into account nVidia's tiled based approach and various compression techniques, the real-world bandwidth is likely to be much larger than that.
Soundwave said:
1440p obviously ain't happening. I have a 1280x800 resolution 8-inch Samsung tablet and honestly the display is pretty nice. Waaaaaaay better than the Wii U or 3DS screens, and I think I like it better than the Vita even, and t his was a dirt cheap $150 tablet that I got two years ago. LCDs are really dirt cheap these days so that's one benefit to Nintendo.
|
I know. But it's what I want. I really don't have room/time for displays that are only 1080P and under.
I have been gaming at 1440P on the PC for years and years, it's hard to go back to anything less.
Soundwave said:
Is there some reason why the iPad couldn't have a L3 cache? I think Nintendo could use a similar main memory setup for the LPDDR4 RAM, but have eDRAM on the die, that would be pretty good, no?
|
128bit LPDDR4+L3 is better than 128bit LPDDR4+eDRAM IMHO.
eDRAM can basically end up being a slower, larger capacity L3 anyway. (Hence why it's used as an L4 on some chips.)
The main reason Apple likely removed it was probably due to costs and to keep power consumption down low, transisters cost money and power.
Soundwave said:
The NX is going to have to be big, that's just how it is, but even a Vita is almost 3x the thickness of an iPad Pro. If NX is the same thickness of a Vita, but is the physical dimensions of say an iPad Mini (height x length), then that's basically like three iPad Minis compared to an iPad Pro ... or basically take your height and length and multiple by three.
|
I have a 7" Android Gamepad/Tablet thing, it's a good size.
If you go to big, then you start to impact on it's mobility... Which then compromises a massive aspect of the device.
shikamaru317 said:
Aren't Nvidia flops usually considered to be better than AMD flops though?
|
It's not as simple as that unfortunately.
Things change depending on the chip, how many TMU's, ROP's, Memory Bandwidth, Polymorph/Geometry engines, Cache sizes etc'.
And of course, the task at hand. There are some compute-only tasks (Aka. Floating point/Flop) that simply fly under AMD's hardware, yet falter on nVidia's and the reverse also holds true.
I think the idea people have that AMD's flops are worst is because people associate it with a certain level of gaming performance, when it's not to be used in that fashion anyway.
shikamaru317 said:
Look at the new 1060 for example, it has ~1000 less flops than AMD's RX 480 yet beats it in quite a few games.
|
Take the Radeon 265 at 1.8Tflop. (Aka. Slightly higher clocked Radeon 7850.)
Yet it's faster than the Radeon 6950 at 2.25Tflop.
And faster than the Radeon 5870 at 2.7 Tflop.
The main reason why the 1060 pulls ahead is because of it's tiled based apparoach.
shikamaru317 said:
Also, though the tablet portion of the NX from the Eurogamer leak will most certainly use LPDDR4, there should be nothing stopping them from putting better memory in the dock portion which could override or work with the LPDDR4 when the system is docked.
|
I doubt they would use different memory whilst docked, there are a few caveats in doing so.
shikamaru317 said:
So, let's say Tegra X2 is 800 gflops. The tablet runs in a battery saving mode when used by itself that lowers the system to around 400 gflops, but when docked it has accesss to it's full 800 gflops. Then a 2nd Tegra X2 in the dock works alongside it, and that Tegra X2 in the dock is paired with GDDR5 instead. I feel like in such a scenario NX would be able to handle PS4/XB1 games at roughly the same settings as XB1 when docked, as well as games designed to run on the tablet by itself.
|
I think you are missing the point entirely.
Even if Tegra was 2 Teraflops verses the Xbox One 1.3 Teraflops... It would still be slower.
Besides, the Xbox One's graphics quality isn't exactly setting people's worlds on fire... And it's about to be superceded by something even better.
Soundwave said:
Lets say octa-core CPU ... quad-core ARM A72 + quad-core ARM A53. Tegra X2 GPU, which is lets say 30-40% better performance over the Tegra X1. 50GB/sec main memory bandwidth (6GB LPDDR4 RAM at 50GB/sec on 128-bit bus).
|
The CPU will be two nVidia Denver cores backed up with Quad ARM A57 in a Big.Little configuration if they go with the Tegra "X2".
Otherwise expect 4x A57 + 4x A53 also in a Big.little configuration if they go with the Tegra X1 chip.
And I am strongly betting the memory will be 4 or 8Gb on a 64bit bus, it makes infinitely more sense.