Just How Bad is Mixing Memory?

Summary of Key Technological Findings (Mixing RAM Memory)

• Common claim tested: The video challenges the idea that mixing “different” RAM is always terrible. It investigates how bad it really is, and whether it can be safe and performance-neutral in practice.
• Sponsor aside: GlassWire is promoted (network activity visibility, malware detection, and blocking misbehaving apps), but the core content focuses on RAM mixing methodology and results.

---

Testing / Methodology (what they did)

They tested mixing RAM using five different DDR4 memory sets (varying speeds and latencies) on an AMD platform:

• CPU: Ryzen 5 3600X
• GPU: RTX 2060 Super

They performed many benchmark runs, swapping RAM modules one configuration at a time, then comparing results.

Baseline configuration

• Two 8GB DDR4 sticks at 3200 MHz
• Timings: 6-18-18-36

Mixing scenarios evaluated

• Mixing similar-speed kits
• Mixing half-and-half kits with different speeds/latencies
• Mixing large speed differences (e.g., 2666 vs 4000)
• Populating more than two DIMM slots (3 and 4 sticks)

---

Main Results

• Short version: It often “barely matters”—especially with 2 sticks.
  • With two modules, compatibility is usually fine on DDR4 systems because modern DDR4 behavior typically “just works.”
  • Faster RAM often clocks down to match the slower modules.
  • Across many configurations, performance stayed close to baseline in their tests.

---

Dual-Channel with Mixed Speeds/Timings

When mixing:

• a 3200 MHz baseline kit with a slower kit using similar timing style (example: 1600/16-18-18-35),

they expected performance to degrade due to the “rule of thumb” that higher speed often requires higher CAS latency.

What they observed

• Performance remained almost identical in their benchmarks when the system could run the mismatched pair at the target speed.
• In other words, if it “works,” it likely works well.

---

When Mixing Becomes Risky

Large speed gaps (e.g., 2666 + 4000)

• “4000 MHz kits” often cost more partly because they’re binned/sorted to higher internal performance targets.
• In their extreme mismatch test:
  • A 2666 kit might be electrically compatible sometimes,
  • but could fail to run at the higher rated speed and even prevent proper boot/powering.

More DIMMs increases memory-controller load (AMD IMC)

As they populated 3 DIMM slots:

• The system could sometimes start/post and run benchmarks initially,
• but later became flaky and produced memory errors.

The video emphasizes that mixing becomes more unpredictable when:

• installing more than two modules
• and/or relying on Flex Mode

---

Flex Mode Behavior (Ryzen-specific)

AMD Ryzen has Flex Mode:

• two modules run in dual-channel
• a third module runs in single-channel

However, the results show Flex Mode can require careful slot placement. Example outcome:

• A 2666 module in a specific slot (B1) caused:
  • memory error codes
  • a restart loop
• They resolved it by:
  • resetting CMOS
  • lowering speed back to 2666

---

4-Slot (4 DIMMs) Outcome

Running all four slots was difficult:

• They found only one configuration that posted reliably.
• It required tuning down to around 3200 MHz (below top-rated modules tested, like 3400+).
• Even repeating the “same” settings with the same sticks/slots later could still fail to post, showing non-deterministic behavior.

---

Asynchronous Dual-Channel with Different Capacities

They also tested mixing different capacities (example: 8GB + 16GB) with similar speeds/timings.

• It worked properly via asynchronous dual-channel in their testing.
• As long as the workload didn’t exceed the smallest module’s capacity, they reported no noticeable performance impact.

---

Practical Guidance / Recommendations

What improves odds of success

• Buy modules rated for the same speed and primary timings
• Prefer matching memory chip dies from the same manufacturer (e.g., Samsung B-dies), even if the RAM brand/logo differs

What they do NOT recommend

• Don’t buy random “on sale” RAM sets and mix them and hope for the best—especially for upgrades later where you may need to fill more slots.

Strong recommendation for reliability

• For upgrades, buy matching replacements (e.g., sell old sticks and replace with a matched kit) rather than mixing mismatched sets.

---

Performance/Compatibility Bottom Line

• Two-stick mixing: often fine and frequently near-performance-neutral.
• More sticks (3–4), big speed mismatches, and slot placement: introduce higher risk of:
  • instability
  • boot failures
  • memory errors
  • unpredictable behavior
  • crashes/blue screens
• Outcomes can vary between systems (“silicon lottery”), so results aren’t guaranteed.

---

Sources / Main Speaker(s)

• Linus (Linus Tech Tips)
• References to Andy (mentioned as part of an “Intel home upgrade series” and a man-cave segment)
• Sponsored mentions: GlassWire, Micro Center, and LTTStore.com