Video summary

Nvidia Tried to Stop This... 2X VRAM GPU Mod

Main summary

Key takeaways

Product Review

Product reviewed / modified

  • Nvidia GeForce RTX 3070 (ASUS “Tough” model) that the creator modded from 8GB VRAM to 16GB VRAM by:
    • Removing soldered GDDR6 memory chips
    • Reballing/replacing memory modules
    • Adjusting GPU “straps” (VBIOS hardware configuration resistors) and/or timings indirectly so the GPU recognizes the new memory capacity

Goal: reduce VRAM-limited stuttering/crashes and improve ray tracing / texture settings at higher resolutions.

Key features achieved (outcomes)

  • 16GB VRAM recognized by drivers
    • Nvidia drivers report 16GB.
  • Works in games
    • Described as “basically a normal RTX 3070 but with more VRAM.”
  • Higher usable texture settings and stability when games exceed 8GB VRAM.
  • Ray tracing capability improved due to reduced VRAM bottleneck.
  • DLSS (including mention of path tracing) becomes more “playable” than on the 8GB version (though still limited by performance and other factors).

Main pros (as stated)

  • Much smoother, more consistent gaming experience (less stutter) because the VRAM ceiling is raised.
  • Big performance uplift in VRAM-heavy scenarios, especially 1% lows:
    • ~19% faster average (1080p/1440p raster + RT mix)
    • ~36% faster 1% lows
  • Ray tracing / VRAM-heavy workloads gain heavily (examples at 1440p):
    • Alan Wake 2 (1440p medium, DLSS Quality)
      • 20% faster average, 43% faster 1% lows
    • Ratchet & Clank: Rift Apart (1440p high RT)
      • 49% faster average, 56% faster 1% lows
      • Still ~60 FPS range
    • Oblivion Remastered (1440p balanced hardware RT)
      • Only 9% faster average, but 50% faster 1% lows
  • VRAM utilization increases
    • Example mentioned: ~11GB allocated in some cases
    • Another mention: up to ~9GB in a scenario

Main cons / risks (very emphasized)

  • Extremely difficult and risky hardware mod
    • Removing and replacing memory chips is described as “the most technically challenging” the creator has done.
    • Requires special tools (hot-air rework, flux, soldering equipment, precision work).
    • Memory reballing is described as very hard; the team struggled for multiple days.
  • Not economically sensible
    • The creator states a modded 3070 is not worth it vs buying a 16GB card.
    • Costs mentioned:
      • RTX 3070 used: ~$220–$250
      • Memory itself: ~$150+
  • May break the card permanently
    • Failures mentioned: display issues, memory corruption, “no display,” needing donor parts/cap/resistor replacements, and additional testing using internal tools.
  • Some behavioral quirks
    • After the upgrade, one issue noted that game settings may not be saved correctly between launches (some games reset options because they assume GPU identity / VRAM characteristics).

Comparisons made

  • Compared against the original RTX 3070 Tough with 8GB VRAM (baseline).
  • Broader context:
    • 8GB has become limiting in newer games, leading to stutters/crashes and restricted settings.
    • Mentions stagnation in memory sizing across newer Nvidia consumer cards.
  • References an earlier example from other content:
    • “Gamers Nexus doubled capacity of a 4090” to justify that mods are possible in theory.

User experience / install experience (creator’s process)

A multi-day workflow:

  1. Removing memory modules
    • Includes preheating advice to avoid board damage.
  2. Wicking/tinning to clean pads
    • Uses leaded solder to lower the melt point for easier workability.
  3. Reballing chips
    • Identified as the biggest stumbling block.
  4. Strap changes
    • To get the GPU to detect the new memory size.
  5. Validation
    • Boot testing
    • Using leaked internal testing software (“MATS” via “mods”/Tiny Linux approach)

Validation outcomes

  • Initially: memory errors/failures indicated by MATS
  • Later: MATS passes after replacing failing modules
  • Additional fix later:
    • “Black screen flicker” resolved by changing power mode / prioritizing performance

Ratings / numerical scores

  • No formal star rating included.
  • Performance improvements presented as percent changes and 1% low gains across games.
  • Explicit mod-time stat:
    • ~35 hours, 47 minutes, 26 seconds of on-camera work
  • Total effort:
    • 35 hours+ overall, took 5 days
    • Huge footage volume: 700GB footage and 37 hours of footage

Concise verdict / recommendation

Verdict: The mod successfully turns an 8GB RTX 3070 into a working 16GB “effectively upgraded” GPU, delivering major gains in VRAM-bound situations, especially reduced stutter and ray tracing stability.

Recommendation: Do not attempt unless you’re extremely skilled with micro-soldering/reballing and willing to accept a high failure risk. It’s not economical compared with buying an actually 16GB-capable card.

Unique points mentioned

  • VRAM stagnation / 8GB limitation causing stutter, crashes, and settings restrictions.
  • Nvidia’s implied reasoning:
    • keeping higher VRAM out of consumer/workstation tiers.
  • Mod method:
    • physically remove memory modules and install higher-capacity ones.
  • RTX 3070 specifically singled out as powerful but limited by 8GB.
  • Memory matching requirements:
    • GDDR6, matching speed, capacity, voltage (1.35V), and temperature targets.
  • Cost/availability constraints:
    • module pricing increased; buying many modules over $300 for 20.
  • Tools used/sponsored (Fantic):
    • soldering iron, hot air blower, screwdriver.
  • Importance of preheating the PCB to avoid cracks.
  • Wicking difficulty:
    • led vs unleaded melting point issues; uses tinning to lower melt point.
  • Reballing difficulty:
    • stencil cleanliness, ball behavior, flux/no-flux timing, airflow control.
  • GPU straps concept:
    • hardware resistors control memory configuration/timings recognition.
    • straps are board-specific; schematics not public.
    • requires moving strap resistors to specific positions/values.
  • Validation steps:
    • boot/Windows detection
    • MATS internal test via mods on Tiny Linux
  • Observed faults during the process:
    • no display / bricking
    • memory corruption
    • capacitors/resistors knocked loose
    • solder contamination on pads/PCIE slot
    • MATS failures on specific modules
    • black screen flicker fixed by setting power mode / prioritization
  • Final results:
    • 16GB reported, games run, VRAM utilization rises, smoother performance
    • performance comparisons across named games (including Alan Wake 2, Ratchet & Clank: Rift Apart, Oblivion Remastered, and others mentioned)
  • Ongoing downside after success:
    • some games reset settings and don’t save configuration properly
  • Overall recommendation:
    • no—not economical; skill/tool intensive; high failure risk

Different speaker contributions

  • Primary speaker (creator/Cameron involved)

    • Drives the narrative: why VRAM matters, what’s being modified, outcomes and comparisons.
    • Performs/records major steps and reports failures and fixes (preheat, wicking, reballing, strap changes, testing).

  • Cameron (more electronics experience)

    • Provides expertise during memory removal/handling.
    • Takes over or performs cleaner removals of memory modules at times.
    • Stays engaged across long work sessions—described as crucial for success.

Original video