Memories (from CATS?)

Chapter 6 — Memory (RAM)

RAM is the CPU's working space. Every application you open, every file you're editing, every Docker container running in the background, and every remote desktop stream being decoded occupies RAM. Run out and the OS starts swapping to disk — and the machine bogs down immediately. This chapter settles the DDR4 vs DDR5 question with real numbers, explains the single most commonly skipped configuration step (XMP activation), and gives you specific kit recommendations for each path.

Your existing kit: Corsair Vengeance RGB RS 64GB (2×32GB) DDR4-3600 C18. This is a genuinely good kit — DDR4-3600 is the sweet spot on the DDR4 performance curve and 64GB is exactly right for your workload. Whether you keep it or replace it depends entirely on which motherboard platform you choose. Both paths are fully covered below.

What RAM Does

RAM (Random Access Memory) holds the data and instructions the CPU is actively working with right now. Unlike storage (SSD/HDD), RAM is volatile — it loses its contents when power is removed — but it's orders of magnitude faster than any SSD. The CPU can access RAM in nanoseconds; it would take microseconds to wait for the same data from even the fastest NVMe drive.

When you open an application, its code and data load from storage into RAM. When you switch to it, the CPU reads from RAM. When you run out of RAM, the OS uses a swap file on the SSD as overflow — and the speed difference is immediately noticeable: things that were instant become slow.

For your dev workstation, RAM is constantly consumed by:

OS baseline (Windows 11 + system processes)~3–4 GB

IntelliJ IDEA / VS Code + project indexing~2–6 GB

Chrome browser (multiple dev tabs)~2–5 GB

Docker containers (3–5 running simultaneously)~4–10 GB

Local database (PostgreSQL / MySQL / Redis)~1–3 GB

X2GO + RDP remote sessions (2–3 active)~2–4 GB

WinSCP, Kitty/PuTTY, OpenVPN client~0.5–1 GB

Build process / compiler working set~2–8 GB

Realistic peak (heavy multitasking)16–41 GB

64GB gives 23–48GB headroom above peak — enough that you will never hit the wall in normal use, and plenty of room to grow your Docker footprint or add a local VM.

Why not 32GB? With your workload — multiple remote sessions, Docker containers, and IDE indexing running simultaneously — 32GB runs close to its limit. Any large compilation or added container will start paging to disk. 64GB costs only marginally more as a kit and removes the problem entirely. 128GB is unnecessary and expensive at this price point.

Understanding the Specs

DDR Generation

DDR4 vs DDR5 — not interchangeable

DDR4 and DDR5 sticks have different notch positions, different pin counts, and different voltage requirements. They cannot physically fit in the wrong motherboard slot. A DDR5 board takes only DDR5 sticks; a DDR4 board takes only DDR4. The generation is determined entirely by your motherboard choice.

Speed / Data Rate

e.g. DDR5-5600

The number after "DDR5-" is the data transfer rate in MT/s (megatransfers per second). The physical clock is half this (DDR5-5600 runs at 2800MHz). Higher is generally faster — but only after XMP is enabled in the BIOS. Without XMP, the kit runs at its JEDEC default (typically DDR5-4800 regardless of what the kit is rated for).

CAS Latency (CL)

e.g. C40 or C36

Number of clock cycles between a memory request and the first data response. Lower CL is better — but only meaningful relative to the clock speed. A DDR5-6000 C36 kit has the same real-world latency as a DDR5-4800 C28.8 kit because the absolute latency (in nanoseconds) is what matters.

True Latency (ns)

CL ÷ MHz × 2000

The real comparison metric. Formula: CL ÷ (data rate ÷ 2) × 1000 = ns. DDR4-3600 C18: 18 ÷ 1800 × 1000 = 10 ns. DDR5-5600 C40: 40 ÷ 2800 × 1000 = 14.3 ns. DDR5-6000 C36: 36 ÷ 3000 × 1000 = 12 ns. Your existing DDR4-3600 C18 actually has excellent latency — one reason to keep it if using a DDR4 board.

XMP / EXPO Profile

Must be manually enabled in BIOS

XMP (Intel eXtreme Memory Profile) and EXPO (AMD Extended Profiles for Overclocking) are the profiles stored on the RAM stick that tell the motherboard to run it at its rated speed. Without enabling XMP/EXPO in the BIOS, DDR5-5600 runs at DDR5-4800. DDR4-3600 runs at DDR4-2133. This is the single most commonly skipped step in PC builds.

Dual Channel

2 sticks, correct slots

Running two sticks of RAM in the right paired slots (A2 + B2 on most boards) enables dual-channel mode, effectively doubling memory bandwidth. A single 64GB stick running in single-channel mode would perform significantly worse than two 32GB sticks in dual-channel. Always use a matched pair — never one stick alone.

Capacity (GB)

64GB = 2 × 32GB

Total RAM available to the system. 2×32GB = 64GB is the right capacity for this workload. Two sticks (vs four) leaves the remaining two slots free for a future 128GB upgrade if needed — though that's unlikely to ever be necessary for this use case.

Voltage

DDR4: 1.35V / DDR5: 1.1V base

DDR5 runs at lower base voltage than DDR4 but requires a PMIC (Power Management IC) built onto the RAM stick itself — DDR5 manages its own power delivery rather than relying entirely on the motherboard. This is transparent to you as a user but affects compatibility between certain boards and RAM sticks.

XMP — The Step Everyone Forgets

WITHOUT XMP ENABLED WITH XMP ENABLED DDR5-5600 kit → runs at DDR5-4800 DDR5-5600 kit → runs at DDR5-5600 DDR4-3600 kit → runs at DDR4-2133 DDR4-3600 kit → runs at DDR4-3600 DDR4 example: JEDEC default XMP speed Bandwidth increase DDR4-2133 → DDR4-3600 +69% throughput DDR5 example: JEDEC default XMP speed Bandwidth increase DDR5-4800 → DDR5-5600 +17% throughput XMP is not overclocking — it simply tells the board to run the RAM at the speed it was designed and tested for. It is always safe to enable.

How to Enable XMP After First Boot

Restart and press Delete (or F2) to enter BIOS at the manufacturer splash screen

Find the memory/OC section — MSI: OC tab → AI OC Genie or DRAM Frequency; ASUS: AI Tweaker → AI Overclock Tuner; Gigabyte: Tweaker

Look for XMP or XMP 3.0 (Intel boards) / EXPO (AMD boards). Change the setting from Disabled to XMP Profile 1

The board will display the new RAM speed (e.g. DDR5-5600). Confirm this matches your kit's rated speed

Press F10 to save and exit. The board will reboot. If it fails to POST (three beeps or black screen), re-enter BIOS — it will have automatically reverted to safe settings. This is harmless; try XMP Profile 2 if available, or the next lower speed.

After boot, verify in CPU-Z → Memory tab: DRAM Frequency should show your rated speed ÷ 2 (CPU-Z reports the physical clock, not the effective data rate). DDR5-5600 shows as 2800 MHz; DDR4-3600 shows as 1800 MHz

Dual Channel — Always Use the Correct Slots

Installing RAM in the wrong slots produces single-channel operation — half the memory bandwidth. The performance impact is measurable in compilation and large file operations. Always check your motherboard manual for the dual-channel slot pairing. On most boards, the recommended pairing is slots A2 and B2 (the 2nd and 4th from the CPU).

Correct configuration — 2 sticks, slots A2 + B2

← CPU

A1
empty

Ch A

A2
32GB

Ch A

B1
empty

Ch B

B2
32GB

Ch B

✓ Dual-channel — correct

Total: 64GB across 2 channels

Full memory bandwidth

Slots A1 + B1 remain free for future upgrade

Wrong configuration — 2 sticks in A1 + A2 (same channel)

← CPU

A1
32GB

A2
32GB

B1
empty

B2
empty

✗ Single-channel — wrong

Both sticks on channel A only

Half the memory bandwidth

Windows still shows 64GB total — easily missed

Your board's manual overrules everything. Most boards use A2+B2 for 2-stick dual-channel, but some use A1+B1. The slots are usually colour-coded (two slots in one colour = one channel). Check the silkscreen printing on the motherboard itself, or the first few pages of the manual. CPU-Z → Memory tab confirms dual-channel operation: it shows "Channel #: Dual" if configured correctly.

DDR4 vs DDR5 — The Full Comparison

DDR4 — Keep your existing kit

Your kitCorsair Vengeance RGB RS 64GB DDR4-3600 C18

True latency10 ns (excellent)

Peak bandwidth (dual ch)~57.6 GB/s

Extra cost£0

64GB kit availabilityDDR4 boards becoming rarer

Ecosystem futureDeclining — DDR4 end of life approaching

Intel sweet spotDDR4-3600 is exactly right

Real-world dev performanceFully adequate

DDR5 — New kit required

Recommended kitG.Skill Trident Z5 or Kingston Fury 64GB DDR5-6000

True latency (DDR5-6000 C36)12 ns (vs 10 ns for DDR4-3600)

Peak bandwidth (dual ch)~96 GB/s (DDR5-6000)

Extra cost~£130–160 for 64GB kit

64GB kit availabilityWide — mainstream standard now

Ecosystem futureDDR5 will remain mainstream for 5+ years

Intel sweet spotDDR5-5600 to DDR5-6000

Real-world dev performanceComparable + higher bandwidth

The honest verdict: For your development workload, DDR4-3600 and DDR5-5600–6000 produce nearly identical day-to-day performance. Compilation is slightly faster on DDR5 in bandwidth-heavy scenarios. The real-world difference across a full working day is small. The DDR5 case is about future-proofing and board selection breadth — not a dramatic performance gain. If you choose a DDR4 board specifically to save ~£150 in RAM costs, you lose nothing meaningful in practice.

Your Specific Decision — Keep or Replace?

Board requiredB760 DDR4 variant (e.g. MSI MAG B760 TOMAHAWK DDR4)

Extra cost£0

True latency10 ns — excellent

Speed after XMPDDR4-3600 (rated speed)

Capacity64GB — right for this workload

RGBYes — can disable in Corsair iCUE or BIOS

Your existing kit is genuinely good — DDR4-3600 C18 is where DDR4 starts performing well on Intel platforms. No performance compromise keeping it. The only consideration: RGB DDR4 heatspreaders can be tall (yours are ~40mm). Check front fan clearance on the CPU cooler before installing — raise the front fan if needed.

Board requiredB760 DDR5 variant (e.g. MSI MAG B760 TOMAHAWK WiFi DDR5)

Extra cost~£130–160

Recommended speedDDR5-5600 (min) → DDR5-6000 (sweet spot)

True latency (DDR5-6000 C36)12 ns (slightly higher than DDR4-3600)

Bandwidth gain~65% more than DDR4-3600

Sell existing DDR4~£50–80 second-hand — offsets cost

Buying DDR5 future-proofs the memory subsystem for the life of this board. The net additional cost after selling the DDR4 kit is approximately £60–100. Specific DDR5 kit recommendations are listed below.

DDR5 Kit Recommendations (if choosing DDR5 path)

The sweet spot for DDR5 on Intel LGA1700 is DDR5-6000 C36 — it offers the best balance of bandwidth, latency, and price. Aim for a kit explicitly listed as compatible with your chosen motherboard (manufacturers maintain QVL — Qualified Vendor Lists — on their product pages).

Configuration2 × 32GB

Speed (XMP 3.0)DDR5-6000

Timings36-36-36-96

True latency12 ns

Voltage1.35V

Heatspreader height~36mm (no RGB) or ~44mm (RGB)

PlatformIntel LGA1700 + AMD AM5 (EXPO too)

£145 – £165

G.Skill's flagship consumer DDR5 line with a 6000 MT/s frequency and tight C36 timings. The Neo variant supports both Intel XMP 3.0 and AMD EXPO — useful if you ever switch platforms. Available in a low-profile non-RGB variant (Neo Silenz) that keeps heatspreader height under 36mm, eliminating any CPU cooler clearance concerns.

Configuration2 × 32GB

Speed (XMP 3.0)DDR5-6000

Timings36-38-38

True latency12 ns

HeatspreaderLow-profile, ~33mm

XMP / EXPOBoth supported

£130 – £155

Kingston Fury Beast is widely available, well-regarded, and broadly compatible with B760 boards. The low-profile heatspreader (33mm) means no CPU cooler front-fan clearance issues whatsoever — a practical advantage over taller kits. Often undercuts the G.Skill by £15–20 at the same speed and timings.

Configuration2 × 32GB

Speed (XMP 3.0)DDR5-5600

TimingsC40

True latency14.3 ns (higher than DDR4-3600)

Price advantage~£20–30 less than DDR5-6000

£115 – £140

Entry DDR5. The C40 timings at 5600 give higher true latency than your existing DDR4-3600 C18. In practice you won't notice the difference for dev work, but given the relatively small price gap to DDR5-6000 C36, the G.Skill or Kingston at 6000 C36 is a better value unless budget is tight. The Corsair brand and warranty are reliable.

What to Look For (and Avoid)

Matched pair (2×32GB) — never a single 64GB stick
XMP 3.0 (Intel) or EXPO (AMD) profile on the label
DDR5-5600 minimum; DDR5-6000 C36 is the sweet spot
Named brands: G.Skill, Corsair, Kingston Fury, Crucial Pro
Listed on your board's QVL (Qualified Vendor List)
Low-profile heatspreader if using a big dual-tower cooler
DDR4 path: your existing kit is already good — keep it
Heatspreader height under 40mm for safe cooler clearance

DDR4 in a DDR5 board, or DDR5 in a DDR4 board (physically impossible anyway)
Single 64GB stick — always loses bandwidth vs 2×32GB
DDR5 below C40 at DDR5-4800 — barely faster than JEDEC default
Unknown brands with no reviews or QVL presence
Buying 4×16GB instead of 2×32GB — fills all slots, no upgrade path
DDR5 kits above DDR5-7200 without checking your board's support
Forgetting to check heatspreader height vs cooler clearance
Skipping XMP activation — most common performance loss in new builds

Installing RAM

Anti-static: RAM is sensitive to static discharge. Wear your anti-static wrist strap or touch a grounded metal object before handling. Hold sticks by their edges — avoid touching the gold contact pads or the PCB.

Identify the correct slots

Locate slots A2 and B2 on the motherboard (the 2nd and 4th slot from the CPU, usually colour-coded). Confirm dual-channel pairing in the motherboard manual if unsure. The correct slots are typically labelled or printed on the PCB silkscreen.

Open the retention clips

Push the retention clips at each end of the target slots outward and away from each other until they click open. On some boards only one end has a movable clip (the other end is fixed); on others, both ends open. Do not force them — they swing outward freely when pushed correctly.

Some boards have very stiff retention clips on fresh-from-factory slots. A gentle but firm sideways push is correct. If a clip won't move, you may be pushing in the wrong direction (they open outward, not upward).

Align the notch and lower the stick

Each RAM stick has a single notch (gap) along the gold contacts. This notch must align with the raised key inside the RAM slot. DDR4 and DDR5 notches are in different positions — this is why they cannot be inserted in the wrong type of slot. Lower the stick into the slot aligned along its entire length, not at an angle.

DDR5 sticks are slightly longer than DDR4 and require more force to seat fully. This is normal.

Press firmly until both clips click

Apply even, firm downward pressure along the entire length of the stick using both thumbs — one near each end. The stick must go in all the way simultaneously, not end-by-end. You will hear and feel two clicks (one from each retention clip) when it's fully seated. Partially seated RAM is the most common cause of DRAM LED errors and boot failures.

RGB DDR4/DDR5 heatspreaders feel very solid — don't be surprised by how much pressure is needed to fully seat the stick. If you hear only one click, check the other end — one clip may not have engaged.

Verify both sticks are flush and both clips engaged

Visually confirm that the top of each stick is at the same height as the other (no tilt), and that both retention clips are closed and snug against the ends of the stick. Give each stick a gentle side-to-side wiggle — it should feel completely solid with no movement.

Check CPU cooler front-fan clearance

If you've already mounted the CPU cooler, confirm that the front fan clears the heatspreader on the nearest RAM stick. There should be at least 2–3mm of gap. If the heatspreader touches or is very close to the fan, raise the front fan one bracket position on the heatsink (most dual-tower coolers allow this adjustment).

Testing RAM

BIOS detection: On first boot, enter BIOS and check that the total installed RAM is reported correctly (e.g. 64.0 GB). If only 32GB is shown, one stick is not seated — power off, reseat, and retry. If 0GB or a DRAM LED lights up, check both sticks are in the correct slots and fully seated.

Enable XMP in BIOS and verify: Follow the BIOS steps shown earlier in this chapter. After enabling XMP and rebooting, confirm in CPU-Z (Memory tab): NB Frequency / DRAM Frequency should show half the rated speed (DDR5-6000 → 3000 MHz in CPU-Z; DDR4-3600 → 1800 MHz). The Channels field should read Dual.

CPU-Z SPD tab: Click the SPD tab in CPU-Z. Select Slot 2 and Slot 4 from the dropdown. Each slot should show your RAM stick's full details (manufacturer, part number, serial, XMP profiles). Empty slots (1 and 3) should show no data. Confirm both sticks are from the same kit (matching serial number prefix).

Windows memory: Open Task Manager → Performance → Memory. Confirm: Used should be 3–5 GB at idle (OS baseline), Speed should show the XMP-activated data rate (e.g. 6000 MT/s or 3600 MT/s), Slots used should show 2 of 4.

MemTest86 (strongly recommended for new RAM): Download MemTest86 from memtest86.com, write it to a USB drive using Rufus, and boot from it. Allow at least 2 full passes. A single pass on 64GB takes approximately 45–60 minutes. Look for PASS: 2/2 with 0 errors. Any errors indicate a faulty stick — identify which one by testing each alone.

Windows Memory Diagnostic (quick check): Type "Windows Memory Diagnostic" in Start → "Restart now and check for problems." Windows will run a basic memory test during reboot and report results. Less thorough than MemTest86 but catches obvious faults in under 10 minutes. Run this if MemTest86 is too slow to wait for right now.

Bandwidth benchmark — AIDA64 (optional): AIDA64's free trial includes a memory bandwidth test. DDR5-6000 in dual channel should achieve ~80–96 GB/s read bandwidth. DDR4-3600 in dual channel should achieve ~55–60 GB/s. Significantly below these numbers suggests single-channel operation or XMP not activated.

Next: Type PC7 to generate Chapter 7 — Storage, where we'll cover adding a new NVMe Gen 4 boot drive, keeping your existing Samsung 870 EVO and WD Blue SATA SSDs, and how to partition and organise a multi-drive setup for a dev workstation.