Ok, was watching Geekerwan’s SoC breakdown with the mainboard he got and noticed 2 GPU clusters. Why is that? Would that not cause latency issues? But it got me thinking after seeing 3gb or ram being used for system resource and 2 CPU cores as well. I always thought it was odd for a max of 102gbps bandwidth with only 9gb of ram at the rumored 3tflops of GPU power on 1536 CUDA cores all shared with the CPU with system memory just seemed off I thought or would be an issue with limited ram. But then after watching this video and seeing two TPC’s separated form each other…that’s 512 CUDA cores by itself with 16 tenor cores……so I think we have all been gaslighted on actual performance….lol
So really devs actually have 6 CPU cores,9 gb of shared ram with 1024 GPU cores as everything else is reserved for system resources and features. Like Gamechat for instance with AI green screen background feature. That does need resources to use. I think the performance estimates with the “leaked” clock speed are true but only if you’re actual using all 1536 cores that I now believe is false on what devs are actually using. So what do you guys think on why it has two separate GPU clusters? And can anyone do the maths on the clock speeds we have being seeing over and over again but doing it with 1024 core instead? To get the real performance number on what I think devs are actually using with 1024 cores for actual game development? I feel like it’s around Xbox One/S in handheld and a bit shy over PS4 in docks. IMO, or maybe I’m just crazy and looking into it to much. lol I know it was Nvidia That said it was 10 times more powerful over Switch 1?…….but as a main PC user……. Nvidia is the biggest gaslighter of them all….maybe with DLSS results not disclosed……but you do get the benefit of a more modern architecture and a better CPU performance over Jaguar cores from last gen systems with a light weight version of DLSS.
Anyways, I still will get a Switch 2 later with a node refresh and improved battery life but I’m really trying to get an actual true performance number on what devs will really have access to. Again, or maybe I’m just crazy but two clusters away from one another just seems strange and I would think it would cause latency issues over just having one large cluster unless they are actually used for two separate things. But I’m no engineer. lol
by wispytomato