bonzobanana said: Unless early Switch's are using the standard chip but later stock will have a customised chip. Maybe all launch Switch's are actually dev kit Switchs with a retail firmware and later Switch's will be different. So many problems with the early Switch's, it's practically beta hardware. I also think there is the possibility (remote possibility) that final retail Switch's will only be 2GB not 4GB and currently only 2GB is being accessed. For early adopters they may benefit later with being able to run dev kit software with a hack. Ok none of this is likely but it is possible. |
wtf?
Lafiel said: seems like this was the cheapest option the Tegra X1 has wasted silicone space, because the 4 A53 cores can't be used while the 4 A57 cores are active, but atleast they saved on a redesign for these chips I imagine when they shrink it to 16nm or lower process they'll get rid of these low powered cores entirely |
They should have ditched the A57 cores and just kept only the A53 cores, save on die-space and power... Which would hopefully be better spent on the GPU anyway.
They should have wen't with a Pascal based design which could have provided 50% better performance for the same amount of power. (If not more.)
JRPGfan said:
This is good news though, it means its likely we ll see a price cut early.
|
I think the only way we will see an early price cut is if the Switch sales start to falter.
It's not entirely impossible for that to happen, the Wii U and 3DS faltered shortly after their initial crazy releases.
Nintendo is a company that likes money first, so if they can get away with a higher price, you bet they will have that higher price.
JRPGfan said:
Thats another bonus.... the next shrink they do of these chips,.... you ll see drastically better battery life.
|
Provided Nintendo doesn't cut back on the battery of course to further decrease costs. It happens all time in the mobile world. ;)
Tryklon said: So, tech pros, whats the Teraflop output of a stock X1? PS4 has 1.84 and XB1 has 1.33, what about the X1? |
Comparing flops against different architectures is inaccurate anyway.
Maxwell is far more efficient than Graphics Core Next 1.0, so can pretty much do more work "per flop".
But for future reference... Shaders * Instructions per clock * clock rate is the formula to calculate flops.
For pretty much every AMD and nVidia GPU it's 2* instructions per clock. Tegra Maxwell and Pascal have 256 shaders.
Thus 256 * 2 * 768 or 307.2mhz. (Docked and portable respectively, Digital Foundry's numbers.)
Ergo 393Gflop and 157Gflop respectively.
Things get tricky if you start dabbling with half precision, quarter precision, double precision etc'.
In older GPU architectures, things like half precision performance had the same performance as single precision.. But starting with Maxwell and Vega, AMD and nVidia will now combine two FP16 operations and execute it as an FP32 operation, aka. Packed math, which doubles the theoretical FP16 performance over FP32. (But due to real-world inefficiencies, it's not a double of performance in real world scenario's. )
Tryklon said:
Not really that bad honestly, the Xbox 360 had 240 GFlops, and it remaisnto be seen if the X1 on the Switch is really stock.
Plus, what I see on Zelda is well beyond 157 in portable mode... that number seems way off
|
No. It's not way off. There is more to rendering a game than just flops you know.
caffeinade said: Also, remember the X1 does have support for half floats; meaning for math that does not require single precision (such as hair rendering) the chip can do 1 TFLOP. This is a feature that the Xbox One and the PS4 (non - Pro) does not have. |
No. Maxwell has double-rate FP16. So it's not ever going to have 1 Teraflop of performance.
It will have 786 and 314Gflop FP16 respectively.
Just remember that FP16 cannot be used for everything either, because of it's reduced precision there is a quality penalty in rendering some elements.