Hardware Would ram upgrade increase switch performance?

[meha]

So I saw this old thread: https://gbatemp.net/threads/did-anyone-ever-tried-a-ram-upgrade-on-the-switch.557229/
and it seems that ram upgrade is possible.
what i'm curious though is if increasing ram capacity would benefit game performance?
my guess is that games are hardcoded to not use more than 4gb memory available ,so possible to write saltynx plugin that lifts this limit?
if this is not possible, would it still increase performance because more vram is available for the gpu or not?

 

[Reploid]

Hardcoded, defintelty hardcoded.

 

[mathew77]

@meha
More RAM (6-8 GB) may be usable with:

• custom CFW Graphical mods, like bigger textures;
• in debuggin' purposes (Dev's Switch officially has 6 GB RAM btw),
• and in Linux 4 Tegra or Switchroot Android projects, for better web browsing experience, for example.

 

[masagrator]

my guess is that games are hardcoded to not use more than 4gb memory available

All games are storing GPU pools hardcoded in executable or in romfs config file (for example IdTech games have them hardcoded in executable, but you can overwrite them with config file), but for CPU most games are like "take whatever is left as heap/alias". To improve performance you prefer to increase GPU pools, so yeah - you must mod those games.
Metro Collection for example is a weird case which hardcodes whole usable RAM to 2GB for whatever reason. Kirby567Fan tested that increasing available RAM for this collection improved slightly performance with 60 FPS mod. :v

 

[Cooler3D]

So I saw this old thread: https://gbatemp.net/threads/did-anyone-ever-tried-a-ram-upgrade-on-the-switch.557229/
and it seems that ram upgrade is possible.
what i'm curious though is if increasing ram capacity would benefit game performance?
my guess is that games are hardcoded to not use more than 4gb memory available ,so possible to write saltynx plugin that lifts this limit?
if this is not possible, would it still increase performance because more vram is available for the gpu or not?

A good and relevant question. I will try to provide a detailed answer that I can refer to in the future:

Nintendo Switch consoles come equipped with various memory chips with different hardware capabilities:

Spoiler: LIST

// LPDDR4 3200Mbps. LPDDR4_ICOSA_4GB_SAMSUNG_K4F6E304HB_MGCH = 0, LPDDR4_ICOSA_4GB_HYNIX_H9HCNNNBPUMLHR_NLE = 1, LPDDR4_ICOSA_4GB_MICRON_MT53B512M32D2NP_062_WT = 2, // WT:C. LPDDR4_ICOSA_6GB_SAMSUNG_K4FHE3D4HM_MGCH = 4, // LPDDR4X 4266Mbps. LPDDR4X_HOAG_4GB_HYNIX_H9HCNNNBKMMLXR_NEE = 3, // Replaced from Copper. Die-M. (1y-01). LPDDR4X_IOWA_4GB_HYNIX_H9HCNNNBKMMLXR_NEE = 5, // Replaced from Copper. Die-M. (1y-01). LPDDR4X_AULA_4GB_HYNIX_H9HCNNNBKMMLXR_NEE = 6, // Replaced from Copper. Die-M. (1y-01). // LPDDR4X 3733Mbps. LPDDR4X_IOWA_4GB_SAMSUNG_K4U6E3S4AM_MGCJ = 8, // Die-M. 1st gen. 8 banks. 3733Mbps. LPDDR4X_IOWA_8GB_SAMSUNG_K4UBE3D4AM_MGCJ = 9, // Die-M. LPDDR4X_IOWA_4GB_HYNIX_H9HCNNNBKMMLHR_NME = 10, // Die-M. LPDDR4X_IOWA_4GB_MICRON_MT53E512M32D2NP_046_WTE = 11, // 4266Mbps. Die-E. LPDDR4X_HOAG_4GB_SAMSUNG_K4U6E3S4AM_MGCJ = 12, // Die-M. 1st gen. 8 banks. 3733Mbps. LPDDR4X_HOAG_8GB_SAMSUNG_K4UBE3D4AM_MGCJ = 13, // Die-M. LPDDR4X_HOAG_4GB_HYNIX_H9HCNNNBKMMLHR_NME = 14, // Die-M. LPDDR4X_HOAG_4GB_MICRON_MT53E512M32D2NP_046_WTE = 15, // 4266Mbps. Die-E. // LPDDR4X 4266Mbps. LPDDR4X_IOWA_4GB_SAMSUNG_K4U6E3S4AA_MGCL = 17, // Die-A. (1y-X03). 2nd gen. 8 banks. 4266Mbps. LPDDR4X_IOWA_8GB_SAMSUNG_K4UBE3D4AA_MGCL = 18, // Die-A. (1y-X03). LPDDR4X_HOAG_4GB_SAMSUNG_K4U6E3S4AA_MGCL = 19, // Die-A. (1y-X03). 2nd gen. 8 banks. 4266Mbps. LPDDR4X_IOWA_4GB_SAMSUNG_1Z = 20, // 1z nm. 40% lower power usage. (1z-01). LPDDR4X_HOAG_4GB_SAMSUNG_1Z = 21, // 1z nm. 40% lower power usage. (1z-01). LPDDR4X_AULA_4GB_SAMSUNG_1Z = 22, // 1z nm. 40% lower power usage. (1z-01). LPDDR4X_HOAG_8GB_SAMSUNG_K4UBE3D4AA_MGCL = 23, // Die-A. (1y-X03). LPDDR4X_AULA_4GB_SAMSUNG_K4U6E3S4AA_MGCL = 24, // Die-A. (1y-X03). 2nd gen. 8 banks. 4266Mbps. LPDDR4X_IOWA_4GB_MICRON_MT53E512M32D2NP_046_WTF = 25, // 4266Mbps. Die-F. D9XRR. 10nm-class (1y-01). LPDDR4X_HOAG_4GB_MICRON_MT53E512M32D2NP_046_WTF = 26, // 4266Mbps. Die-F. D9XRR. 10nm-class (1y-01). LPDDR4X_AULA_4GB_MICRON_MT53E512M32D2NP_046_WTF = 27, // 4266Mbps. Die-F. D9XRR. 10nm-class (1y-01). LPDDR4X_AULA_8GB_SAMSUNG_K4UBE3D4AA_MGCL = 28, // Die-A. LPDDR4X_UNK0_4GB_HYNIX_1A = 29, // 1a nm. 61% lower power usage. (1a-01). LPDDR4X_UNK1_4GB_HYNIX_1A = 30, // 1a nm. 61% lower power usage. (1a-01). LPDDR4X_UNK2_4GB_HYNIX_1A = 31, // 1a nm. 61% lower power usage. (1a-01). LPDDR4X_UNK0_4GB_MICRON_1A = 32, // 1a nm. 61% lower power usage. (1a-01). LPDDR4X_UNK1_4GB_MICRON_1A = 33, // 1a nm. 61% lower power usage. (1a-01). LPDDR4X_UNK2_4GB_MICRON_1A = 34, // 1a nm. 61% lower power usage. (1a-01).

However, as many have noticed, the factory performance of all Nintendo Switches is identical, and replacing memory chips by itself will not affect game performance, even when transitioning from low-rated to high-rated chips, including those with increased capacity.

This is despite the fact that memory bandwidth in the Switch is the main bottleneck for gaming performance in the vast majority of graphically intensive projects (practically all, if you plan to play with graphic modifications to improve graphics, unlock frame rates up to 60 fps, fix resolution at 720p, etc.). And when I say "main bottleneck," that's putting it mildly, but more on that later.

The reason for this frustrating paradox is the heavily limited memory operating frequency and its monstrously high timings, standardized across all Switch models, regardless of the memory chips used. The trade-off is sacrificing "free" (due to the memory's negligible power consumption) performance in favor of unification.

The good (grandiose) news is that this problem is qualitatively solved in a matter of minutes:

Spoiler: SOLUTION

by a software method called 4IFIR. So effectively that a hypothetical Switch Lite with low-rated memory, by sacrificing only a negligible (~22%) decrease in battery life, was able to achieve stable 720p/60 fps in a game that runs at 360p (640×360) with regular frame rate drops to 20 fps in stock settings. That's 12 times more pixels per second! Moreover, most of the performance boost is due to the increased memory bandwidth (which also means that by disabling other optimizations, you can get a decent boost without reducing battery life in principle).

But that's not all! For those willing to hard-mode, 4IFIR also includes experimental support for read-on-write function for compatible memory modules (available for ~$24 from China), which literally doubles the already achieved results.

Bonus for developers: 4IFIR is freely distributable software, and I would be happy to provide the project code upon request.

As for increasing gaming performance by increasing memory capacity, the short answer is no. Games on the platform are designed with the available amount of RAM in mind, and there is no system swap file. To achieve any performance gains, it would require reverse engineering and partial refactoring of each game, which is a futile task.

P.S. My English is poor, forgive me.

 

[Redisk02]

What about upgrading to faster ram modules? Or is that likely locked too so you wouldn't alleviate any of the bottleneck with higher clocked memory?
Also, if I was to just buy larger chips that are faster that have the same layout, would I be able to just use them if I'm willing to accept that I won't get the full storage capability?