| SonytendoAmiibo said: So what you are trying to tell us non-technophile neanderthals is that trying to use teraflops as a measurement of system power simply doesn't work because there are just too many variables involved. The days of comparing 8 bit vs. 16 bit have been over for decades and there is no easy way to make comparisons anymore. Maybe someone needs to come up with an easy to understand grading system for these chips so they are easier to compare. Thanks for the info. |
I don't want to call anyone names :) You got my point exactly -- there are too many variables that make up the various facets of the graphics chip/card. Think of the specifications of the videocard as paper specs but they don't always translate into real world performance. I'll prove it to you even if you don't know anything about technical specs. Just follow along on this one:
AMD Graphics Core 1.1 Architecture R9 290X : 2816 shaders (AMD calls them Stream Processors) @ 1000mhz => 5.63Tflops, 512-bit memory bus, GDDR5 320GB/sec memory bandwidth
NV GTX980Ti Maxwell architecture : 2816 shaders (NV calls them CUDA cores) @ 1000mhz with 1050mhz Boost => 5.63-5.91Tflops, 384-bit memory bus, GDDR5 336GB/sec memory bandwidth
http://hwbench.com/vgas/radeon-r9-290x-vs-geforce-gtx-980-ti
Let's say there were 2 consoles with these GPU specifications: Console #1: 5.63Tflops, 320GB/sec memory bandwidth vs. Console #2: 5.63-5.91Tflops and 336GB/sec memory bandwidth. Using these top-level specifications, a lot of people could erroneously conclude that Console #2 is at best 5.91Tflops / 5.63Tflops = 5% faster.
In reality, GTX980Ti is 38-39% faster than the R9 290X (100% / 72-73%):
https://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_980_Ti/31.html
You see why taking 1 specifiation, in this case the Arithmetic Logic Unit (GPU clock speed x Stream Processors/Shaders/CUDA cores x 2 ops/clock cycle) leads to incorrect conclusions?
It's not possible to assess how powerful the Switch is relative to the Wii U or XB1 using TeraFlops because the Switch uses Maxwell GPU Architecture, the Wii U uses VLIW-5 (Very Long Instruction Word architecture) and Xbox One uses Graphics Core Next Architecture.
All of these architectures have a different Performance Per Clock, or often described as Instructions Per Cycle (IPC).
It gets even more confusing since the paper specs also cannot be compared. For example, 2014 Maxwell's memory bandwidth is 33% more efficient than 2012 Kepler's. That means the 25.6GB/sec memory bandwidth of the Switch is "equivalent" to 34GB/sec of NV's 2012 GPU architecture:
http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/3
To summary, I am not defending the specs of the Switch as clearly it's a handheld console with a bonus feature to connect it to the HDTV. My point is don't try to directly compare XB1/PS4 vs. Wii U vs. the Switch using Tflops as it's too inaccurate of a comparison.
