| Soundwave said: I know it's not the be all end all, but wouldn't two Tegra X2's in unison constitute about 1.25 TFLOP of performance, and if we average out that Nvidia's floating point performance is generally 30% higher, that would put the two at 1.625 TFLOPS in AMD terms. Maybe if they added say 24-32MB of high speed eDRAM onto the SCD version of the TX2 ... would that change things? That would kinda cancel out the XB1's memory bandwidth advantage. Where would it be lacking vs a XB1 in that scenario? Surely two X2 units would boost things like the poly count and fillrate? |
Nope. Because there is inefficiency's being added into the system thanks to Multi-GPU technologies, they don't scale linearly.
Plus the bandwidth doesn't get combined either... It's all well and good to have insane levels of performance... But if you can't squeeze enough data through the small pipes, then you aren't going anywhere.
The numbers you are looking at are only theoretical anyway.
You can't look at a couple of gaming benchmarks (Which represents an entire chip) and quanitify it as nVidia being 30% better at floating point (Which is only a small part of a chip), when AMD can beat nVidia rather soundly in some floating point tasks. (I.E. Anything using Async Compute is AMD's domain, I.E. Anything modern and forward looking.)
Also, you need more than just flops and bandwidth to win the graphics game, if that was all that was ever needed then designing GPU's would be far easier. :P
The Xbox One also has more texturing power, it can handle larger and more textures at once than Tegra, basically all the surfaces in the game with all the little details? Like the floor, heck even the sky? Furniture? Will all look better on the Xbox One, more Nomal Maps. Everything.
There is a reason why Xbox 360 ports to Tegra X1 still looked inferior on the Tegra X1 compared to the Xbox 360, despite the Tegra X1 having almost twice the Gflops, it could keep up with shader effects and lighting and such, but fell backwards with texturing, geometry, streaming and in some cases, resolution, Anti-Aliasing and filtering.
Geometry wise, the Xbox One would beat Tegra too, Tegra is pretty relaxed on it's Polymorph engines for good reason... So the Xbox One can have more detailed models... And thanks to Tessellation, more smaller "bumpy" surfaces like small rocks and pebbles rather than flat surfaces without definition.
Tegra is also not as proficient at single and double precision floating point as Graphics Core Next either, generally. Which can impact accuracy.
It also falls behind on integers and Vertex.
The eDRAM would help eat away at the Xbox One's bandwidth advantage, but I think at that point, Nintendo would be better off throwing a couple of Gigabytes of GDDR5X memory at the problem... And even then, it's not a clear-cut winner on who is faster between the NX and Xbox One.
| Werix357 said: Wish more people read this as people are jumping to conclusions based on GFLOP's |
I am constantly nagging people about it.
| JEMC said: Direct3D or DX is a propietary API from MSoft, so neither Sony nor Nintendo will use it in their development tools. That leaves us with OpenGL/Vulkan. Also, those theories with 2xTegras have, in my eyes, one big flaw: they are assuming a 100% scaling. That's something that won't happen. |
That's true.
But I took into account that Pascal is generally more efficient all around than Graphics Core Next 1.0/1.1 in the Xbox One.
It has better compression and culling which helps.
I stand by that two Tegra X2's should be roughly equivalent to the Xbox One, just forcing games to be at a lower 720P resolution and that isn't based on the flops either. :P
| JEMC said: Direct3D or DX is a propietary API from MSoft, so neither Sony nor Nintendo will use it in their development tools. That leaves us with OpenGL/Vulkan. |
Sony and Nintendo will likely have both OpenGL and Vulkan. Possibly a 3rd low-level API.
There is advantages to having both OpenGL and Vulkan, despite Vulkan being the technical successor to OpenGL.
www.youtube.com/@Pemalite
