sc94597 said:
1. You say it is well-documented but you didn't provide said documentation. The Wikipedia page has no source for their Xenon and Cell statistics. Plus one must distinguish DMIPS (which are results from an actual benchmark called Dhrystone) from MIPS (which is a pretty useless measurement across architectures as Permalite noted earlier.) 2. Only if you can take advantage of six threads. Very few games do. There is only so much you can parallelize, and the more threads we're talking about the lower the returns. This is gaming, not video-editing. 3. Okay, but when we're talking about such a large difference in GPU power, why even care about the SPE's? These computations can just be done on the GPU. The SPE's were a hassle for PS3 development anyway, and many games suffered because of it. 4. I agree, there is no way. The Switch's CPU is definitely better for gaming. 5. This is mostly speculative/overly assertive on your part. 6. The switch's dock is there to prevent throttling. If throttling were a thing we'd notice when our Switch's got hot. Now there is likely throttling to reduce power-draw for less intensive games, but that is entirely dependent on the requirements of the game. It might be valuable to read the quote that quickrick provided, cited from his random developer. "Cell and Xenon are good in highly optimized SIMD code. Xenon = 3 cores at 3.2 GHz, four multiply-adds per cycle (76.8 GFLOP/s). That's significantly higher theoretical peak than the 4x ARM cores on Switch can achieve. But obviously it can never reach this peak. You can't assume that multiply-add is the most common instruction (see Broadwell vs Ryzen SIMD benchmarks for further proof). Also Xenon vector pipelines were very long, so you had to unroll huge loops to reach good perf with it. Branching and indexing based on vector math results was horrible (~40 cycle stall to move data between register files). ARM NEON is a much better instruction set and OoO and data prefetch helps even in SIMD code. |
You seem to have a strong bias in favour of the Switch rather than a sit on the fence of looking at the evidence and how the Switch performs. Your bias seems to be negating all evidence that differs with what you want to believe. Those mips/dmips figures are admittedly only a rough guide to comparing chips but you seem to be using that opportunity to believe it is the Switch under represented when its more likely the other way for reasons I've given previously. You would obviously understand the powerpc figures are pretty much inline with many other powerpc cpu's that are variants on the same processors in ps3 and 360. Just correlate them with other powerpc chips in that list of performance figures. There are no wild claims.
Dhrystone is just an integer performance check and those figures are mainly DMIPs anyway.
It's not like those PPC figures seem strange they are stating 2 mips per mhz per core on 360 where as wii was 2.3 mips per mhz which was out of order execution like ps3. So even with the dual thread on each core they are stating a low figure. It's only because there are 3 cores running at 3.2ghz that the final figure is so good. Basically 6 x 3200 approx which is 19,200. The cpu design of 360 seems to be designed around generating high floating point performance.
You look at LA Noire on Switch and you can clearly see a system with a CPU bottleneck in the design and this is hardly surprising with Nintendo.
Both the specification and real world performance support my view I believe but I'm happy to change my opinion if any evidence comes along to change it. I was an early adopter of the wii u but it was clear when I started playing it there was a cpu issue and that was made clear when the actual specs were leaked. I just don't think Nintendo are to bothered about CPU performance. Surely they could have run the Tegra at full CPU speed if they wanted to but didn't.
