Nintendo Switch

Nintendo Switch Firmware 5.0 Has A Reference To A Model With A Newer SoC

The latest firmware launched last night for the Nintendo Switch and the team over at Switchbrew have discovered that it includes a new check for a SoC (system on a chip) Reset Era member Atheerios says that it’s most likely been changed due to a hardware vulnerability that makes hacking possible in all firmware versions. This in turn means that future Nintendo Switch units will most likely have a different SoC.

Source / Via



    1. It seems far too early for that to show up… I’d imagine a reference to an upgraded Tegra or whatever wouldn’t make it into the firmware until closer to launch (maybe in 6.0 or 7.0) – and we know they aren’t launching a “Switch Pro” this year (but hopefully they are looking into it!)


  1. Before anybody gets their hopes up, it’s likely that the new SoC will be a minor revision just to fix this vulnerability. Of course, it could very well be possible that they go with an X2 and clock it the same way just to take the power savings but it’s to both Nintendo and Nvidia’s advantage to fix an vulnerabilities in the TX1 especially when the TX1’s other market is self driving cars.

    Liked by 2 people

      1. Nope, it uses a TX1. The die shots are identical. A TX2 uses A57s as the performance cores, Denver cores as the performance cores, has an ethernet block, more camera interfaces, and was manufactured on a 16nm process so the dies would look noticeably different.

        The only possible difference, which was mentioned in developer documentation, is that fill rate is lower though I’m highly skeptical of that.


      2. What about the programming of Switch. The Switch uses half floating points, and people say the Switch is confirmed doing 3 Tflops.


      3. Nobody with any credible at all is saying that it’s hitting 3 TFLOPS. No GPU ever hits its max theoretical performance and 3 TFLOPS is over three times the peak theoretical about of 16b FLOPS it can execute and twice what the peak FLOPS the theoretical peak flops of the TX2 .

        When you hear numbers about how many FLOPS something can execute, it’s not based on any benchmarks, it’s determined by an algorithm: clock speed x ALU count x 2 FLOPS . That number will get you the amount of FLOPS it could execute if memory were infinitely fast and the entire work flow were MADDs. This gets you 32-bit FLOPS. In the case of GPU’s with ALUs that support doing two of the same operations at half precision, you would just multiply this number by two again. Again, this is performance in a purely theoretical and statistically impossible scenario. In reality, there’s always latencies and the types of operations and their precision vary from task to task.

        Saying the TX1 “uses half floating points” suggests that that’s its only precision it uses. If this were true then all Switch games would use vertices with the same precision as the N64 and artifacts would be noticeable. In reality, most vertex shaders need 32-bits of precision but half floats would be suitable for most fragment shaders.

        The Switch’s max theoretical full-precision shader performance is just 196.608 GFLOPS in handheld mode and 393.216 GFLOPS in docked mode. To put that in perspective, the difference in peak full-precision GFLOPS between the PS4 and Xbox One is 140 GFLOPS higher than the maximum the Switch can execute. As for memory bandwidth situation, the max amount of memory bandwidth than can be accessed by just the CPU in the PS4’s APU, is twice as much as is available to the entirety to the entirety of the TX1.

        In real world performance, PS4 is likely 10-12x more powerful than Switch in handheld mode and about 5x more than its docked mode. That’s why we see a game like Doom running on the PS4 at twice the frame rate and 2.25 times higher resolution (with higher resolution textures) than the Switch. That alone shows at least a 4.5x difference in performance between both platforms.

        Liked by 1 person

      4. Thanks. So its TX1 not TX2 chip. But what about the 2 years of shortening the source codes all the people of NVIDIA spent to get more & more performance while draining less & less power/electricity at the same time ?

        Liked by 1 person

      5. That never happened because source code becomes software not hardware.

        The only exception to that would be a Hardware Description Language which is compiled into an actual chip design. If any changes were made to the TX1’s HDL that would have any effect on the actual peak performance or power usage, it would changed the physical layout of the chip or the architecture as a whole and it wouldn’t be a TX1 anymore. Short of very minor revisions for security fixes or bugs, chips do not get changed after they’re finalized and released.

        So yea, the TX1 in the Switch is the same TX1 released two years ago but set to run at slower clockspeeds by the firmware.

        Liked by 1 person

    1. I’m pretty sure they can already do that but at the cost of CPU speed. They need to keep below certain thermal and power restraints so modes with higher GPU clocks will have lower CPU clocks and vice versa.


Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s