JRPGfan said:
Bandwidth matters, as you start running higher resolutions. So while that 960 might beat a 7970 at like 720p, I bet you once you start running 1440p, it has the advantage right? Probably because of that 135% more memory bandwidth.
Would you look at that, a factory OCed 960, is slower than a stock 7970 at 1440 on avg. Stock to stock, its around 17%.
*edit: actually this review, shows the 960 is slower than the 7970 on all resolutions they tested.
|
That 135% more bandwidth and 60% more flops only increased performance by 9% at 1440P.
That is pretty insignificant.
You are still proving my point. That Architecture is more important than flops or bandwidth.
Overclock that 960 heavily and it will match/exceed the 7970 even at 1440P.
Not to mention the 960 was always more of a 1080P card anyway.
As for benchmarks themselves, Anandtech is a more legitimate source.
https://www.anandtech.com/bench/product/1722?vs=1744
Leynos said:
The Xavier is not meant for mobile hardware. It's a large mother board and meant for cars/trucks. It's not designed for something mobile with a battery.
|
It can scale downwards, it has configurable TDP's.
The 350mm2 die size is the difficult part, but 12nm is mature as it's based on 14/16nm which in turn is based on 20nm, so yields would be absolutely fantastic... Plus there is room to move downwards by porting it to 7nm.
haxxiy said:
Power consumption and manufacturing node are factors here, though.
The Tegra X1 consumed 10 - 15 watts to deliver 500 - 600 GFLOPS and even then it had to be underclocked to fit on the Switch. The figure you're quoting probably refers to the 30W TDP option. The Xavier chip under similar power constraints is a 600 - 800 GFLOPS chip. Even when you factor in the IPC gains of the Volta microarchitecture, it's unlikely the average improvement for games would reach a 2 times increase.
Now, we do more or less know how the next three or so upcoming nodes will perform in terms of feature size and power consumption. How close to a PS4 Pro a ~3 nm or so Switch 2 would perform, though, remains to be seen considering we know nothing about future architectural improvements or which clocks or die size Nintendo would use to fit their power and cost constraints.
But to assume it could match it certainly risks falling into the same optimistic trap a few members here fell into some four years ago, when they were sure the Tegra X1 could deliver a portable console at least matching the Xbox One. And we know how that one turned out.
|
The jump between Tegra X1 and Xavier is massive.
Tegra X2 can already have 50% more performance than the X1 at the same powerlevel as it's more power optimized for higher clockrates. (Typical advantage of Pascal over Maxwell.)
There were also improvements to Delta Colour Compression along the way which means more bandwidth even with the same memory setup.
And Xavier has twice the functional units on top of that.
So yes, we can take Xavier and underclock it significantly to meet a Switch-like handheld requirements and still garner a significant increase in performance.
Plus I would hope Nintendo would include a larger (I.E. Not a tiny 6.2" 720P display) so there should be more physical space for a larger battery.
haxxiy said:
But to assume it could match it certainly risks falling into the same optimistic trap a few members here fell into some four years ago, when they were sure the Tegra X1 could deliver a portable console at least matching the Xbox One. And we know how that one turned out.
|
Yeah, I think that was always a bit of a stretch.
As soon as I knew the hardware was Tegra based, I always asserted that the hardware would fall inline between an Xbox 360 and Xbox One, the reduced clockrates that came out later didn't change that.
Still, it's impressive what developers have been able to do with the hardware... Links Awakening for example looks absolutely adorable considering the hardware constraints.
Doctor_MG said:
Yes, the die size is roughly two and a half times that of the X1, but the power consumption is still in the 10-15W range. I'll concede, though.
The Adreno 650 is supposed to be capable 1.2TFLOPS of theoretical performance (yet to release), and that GPU is made for the mobile market. The 640 gets around 1.04TFLOPS of performance (released this year), and the A12X from Apple is supposedly at 1.3 TFLOPS of performance (released last year).
|
Larger die sizes doesn't always equate to more power consumption, sometimes it does the opposite as there is dark silicon in order to reduce leakage and drive up clockrates... Or spare functional blocks in order to keep yields high.
The Adreno 650 is an interesting beast, it should beat the Xbox One even with less flops as it's a more efficient architecture than Graphics Core Next... And tends to rely on tiled based rendering so it should get extremely efficient use out of it's limited memory bandwidth.
Flops just doesn't tell the whole story.