How Will be Switch 2 Performance Wise?

[bonzobanana]

Pemalite said:

bonzobanana said:

Tegra Orin is a SOC with up to 60W TDP meant for premium automobiles with a high cost and a huge feature set for the time. Yes it was designed around the same time as the Switch 2 chipset back in 2018/19 and shares the same fabrication node but its certainly not relevant to the Switch 2 spec with regard built in support chips which is a much more cut down mobile device designed for a TDP of less than 10W likely around 6W. I don't understand why you have mentioned this it adds nothing to the conversation and isn't relevant at all. Yes ARM A78s are typically much more powerful than PS4 Jaguar cores but not if you run them at 1Ghz they are designed to be fabricated on a 5Nm node ideally with a speed of up to 4Ghz (actively cooled). Frankly they are poor performing at 1Ghz and we know the cache will be reduced to be fabricated on a mainly 10Nm node. Anyway there are benchmarks that clearly indicate their performance and this backed up by Virtuos in that the overall speed is a bit faster than PS4 which is exactly where a passmark of 2000 sits in CPU performance. Surely with all the data out there and the developers themselves stating this its not really debatable. I've got a Razer Edge android games console that I picked up for £100 and that has 3 ARM A78 cores at 2.4Ghz on the optimal 5Nm node plus 4x A55 at 2Ghz and 1 super core X1 at 3Ghz which is competitive with high performance x86 CPUs. All those together works out at about a passmark of 4000 for CPU performance, still less than half that of the Steam Decks 9000 figure. The Switch 2 has very weak CPU performance of less than a quarter of steam deck. It's strengths are all in its GPU not CPU performance.

Orin

Nvidia announced the next-gen SoC codename Orin on March 27, 2018, at GPU Technology Conference 2018.^[157]

It contains 17 billion transistors and 12 ARM Hercules cores and is capable of 200 INT8 TOPs @ 65W.^[158]The Drive AGX Orin board system family was announced on December 18, 2019, at GTC China 2019.

Nvidia has sent papers to the press documenting that the known (from Xavier series) clock and voltage scaling on the semiconductors

and by pairing multiple such chips a wider range of application can be realized with the thus resulting board concepts.^[159]In early 2021, Nvidia announced the Chinese vehicle company NIO will be using an Orin-based chip in their cars.^[160]

The so far published specifications for Orin are:

• CPU: 12× Arm Cortex-A78AE (Hercules) ARMv8.2-A (64-bit)^[161][162]
• GPU: Ampere-based, 2048^[163] CUDA cores and 64 tensor cores¹; "with up to 131 Sparse TOPs
  of INT8 Tensor compute, and up to 5.32 FP32 TFLOPs of CUDA compute."^[164]
  • 5.3 CUDA TFLOPs (FP32)^[165]
  • 10.6 CUDA TFLOPs (FP16)^[165]
• Samsung 8N process (derived from 8LPU)^[165][166]
• 275 TOPS (INT8) DL^[165]
  • 170 TOPS DL (INT8) via the GPU
  • 105 TOPS DL (INT8) via the 2x NVDLA 2.0 units (DLA, Deep Learning Accelerator)
• 85 TOPS DL (FP16)^[165]
• 5 TOPS in the PVA v2.0 unit (Programmable Vision Accelerator for Feature Tracking)
• 1.85 GPix/s in the ISP unit (Image Signal Processor, with native full-range HDR and tile processing support)
• Video processor for ? GPix/s encoding and ? GPix/s decode
• 4× 10 Gbit/s Ethernet, 1× 1 Gbit/s Ethernet

¹ Orin uses the double-rate tensor cores in the A100, not the standard tensor cores in consumer Ampere GPUs.

Nvidia announced the latest member of the family, "Orin Nano" in September 2022 at the GPU Technology Conference 2022.^[167]

The Orin product line now features SoC and SoM (System-On-Module) based on the core Orin design and scaled for different uses from 60W all the way down to 5W. While less is known about the exact SoC's that are being manufactured, Nvidia has publicly shared detailed technical specifications about the entire Jetson Orin SoM product line. These module specifications illustrate how Orin scales providing insight into future devices that contain an Orin derived SoC.

https://en.wikipedia.org/wiki/Tegra

Yes the Tegra in the Switch 2 is a deviation from the Tegra Orin, but it's not that much of a fucking deviation that was a chip built from the ground up which omitted fixed-function processing blocks, it's based on an existing design with a ratified stencil pattern, nVidia still needs to stick with it's establish design rules so they don't need to get the chip re-evaluated by outlets like FCC/IEEE/IEC/CISPR and more which takes time and money.
But this is all common sense and I am sure you were apprised of that, right?

Every ARM SoC for over a decade has literally included a network processing engine so as not to burden the CPU and save on power.
Same goes for Audio.
Same goes for video.
ARM SoC's were always designed around the concept of fixed function blocks to offload specialized tasks to save on power and increase performance.

If you disagree with that... Then I am sorry. But you are just wrong. - Your copied/pasted rubbish neither confirms nor refutes my previous post in any capacity.

As for the CPU performance... Again... If you even bothered to read my actual post, even if the Switch 2 CPU was equivalent to Jaguar CPU, the Switch 2 will be faster as it's offloading things like compression to a fixed function block.
As for the TDP argument, it's irrelevant. Chips operate on a voltage curve. You can have a single chip go from 10w to 1,000w if you want.

And again... You need to look beneath the black and white paper specifications of hardware and get a more nuanced, low-level understanding.

The Orin is 455mm2 surface area compared to 200mm2 for T239 and 200mm2 is the classic size for mobile devices and even many premium phones still rely on their CPU for some of the network processing because they are always fighting for space on the die. The 455mm2 size is a very premium and expensive size with low yields typically. You can't compare a super expensive feature rich chipset with a budget cutdown mobile chipset. The Switch 2 couldn't afford all the power requirements of such a chip and it has been argued that one of the benefits of the Switch 2 is its mobile optimised design rather than laptop chipset in Steam Deck which does have a huge amount of support processors but they have a cost in power consumption. You can't just make up random rubbish about the Switch 2 using an Orin type chipset when they are completely different designs made for different markets completely. Every chipset nowadays has fixed function blocks to optimise certain functions like video decoding, its nothing new that is across all processor types including x86 etc that doesn't mean network processing doesn't have a CPU cost if you don't have a dedicated network processing chip.

We do know the Switch 1 relies on the ARM processor for network data processing and the Switch 2 is compatible with Switch 1 software with network features. We also know Codemasters completely omitted network functionality for Grid Legends on Switch 2 for its initial release. I'm saying the Switch 2 likely processes network data like the Switch 1 and other mobile devices relying on its CPUs. This makes it more power efficient and a simpler design. Dedicated support chips cost power. I've looked and seen nothing that indicates the T239 incorporates a dedicated network processor. The issue with Switch 1 using CPU performance for network processing was a less robust network connection and more inconsistent frame rates as you had a CPU with more varying load. The network experience of Switch 1 was inferior to PC, Playstation and Xbox. My thinking is Grid Legends found networking poor or they simply couldn't afford the CPU resources for it. They may be optimising the game at this point to add it later or maybe they can't do it at all. The point is there is a reason why they have omitted and extremely important feature of this game completely. That is my guess at the reason based on low CPU resources of the console its very difficult to do. You may have another reason why such an important feature has been omitted.