New desktop build with a DRAM debug LED and a PSU upgrade question
A shop assembles a desktop for a customer: an Intel i5-12400 (LGA1700, 65 W base, about 117 W at PL2) on a B660M board with four DIMM slots, 2 × 8 GB DDR4-3200 fitted in slots A1 and A2, a GTX 1660 Super (125 W, one 8-pin) in the top PCIe slot, a 1 TB NVMe SSD, and a 550 W 80+ Bronze PSU whose single 12 V rail is rated 42 A. On first power-on the fans spin, the DRAM debug LED stays lit and there is no video; the board has no buzzer. The monitor cable is plugged into the motherboard HDMI port. The assembler says the PSU is 'only 550 W, too weak', and wants to swap the motherboard. The shop ambient is 35 °C, and the customer later wants 16 GB more RAM, a second NVMe drive, two 140 mm fans, and is asking whether the same PSU can run a partner-card RTX 3070 (220 W, 2 × 8-pin) at full load.
1. Give the order of diagnosis for the DRAM LED with no video, covering slot population, minimum configuration and the monitor connection.
Read the worked answer
The DRAM LED means the board is stopping during memory training or has not detected usable memory – it is not a PSU symptom. A 550 W unit does not cause a DRAM LED.
Order of work
- Slot population. On a four-slot B660M board with two sticks, the correct pair is A2 and B2 – the second and fourth slots counting from the CPU socket. A1 + A2 is the wrong pairing and can prevent training on some boards. Re-seat with firm pressure until both latches click; a half-seated stick is the most common cause of a stuck DRAM LED.
- Clear CMOS with the jumper or by removing the coin cell for 30 s with the PSU unplugged, then retry.
- Minimum configuration. One stick in A2, no GPU, no SSD, using the i5-12400's UHD 730 output. If it POSTs, add components back one at a time. Test the second stick and the other slots individually to find a bad stick or slot.
- CPU socket and cooler. LGA1700 pins bend easily; inspect with a light and magnifier. Over-tightened cooler screws cause uneven contact and memory-channel faults.
- Video path. With a discrete GPU fitted, the monitor must be on the graphics card, not the motherboard HDMI – many boards disable or do not route the iGPU when a card is present. Explain this to the assembler before anything is replaced.
Replacing the motherboard first is guesswork; suspect it only after a known-good stick in A2, a cleared CMOS and a confirmed card output fail to POST.
2. Check the PSU sizing with numbers and answer the RTX 3070 question.
Read the worked answer
Current build
- CPU: 65 W base, 117 W at PL2 → about 10 A on 12 V at peak
- GPU: 125 W → about 10.5 A on 12 V
- Board, RAM, NVMe, fans, USB: 40–50 W → about 4 A
- Total DC draw ≈ 250–300 W; with 88% efficiency the AC draw is about 300–340 W.
That is roughly 55% of a 550 W unit, and the 12 V rail (42 A = 504 W) carries only about 25 A at peak. The PSU is correctly sized – the assembler is wrong. Note that the 35 °C ambient derates the unit slightly; keep the intake clear and the fan unobstructed.
Adding the planned parts – 16 GB more RAM, an NVMe drive and two fans add perhaps 15 W. Still comfortable.
RTX 3070
- 220 W GPU + 117 W CPU + 50 W rest ≈ 390 W DC, about 450 W at the wall.
- Transient spikes on a 3070 class card can be 1.5–2× TGP for milliseconds; a 550 W Bronze designed several years ago may trip OCP or produce rail sag.
- The partner card needs 2 × 8-pin; a single daisy-chained cable is not acceptable.
Verdict: a 650–750 W 80+ Gold unit with two separate PCIe 8-pin cables is the correct answer for a 3070. The existing 550 W is fine for the current build and can be kept as a spare, but it is marginal for the upgrade, and 750 W leaves room for the 35 °C ambient and future changes.
3. The customer adds the extra parts later. What checks and configuration steps would you document before delivery?
Read the worked answer
Firmware and configuration
- Update the B660M BIOS to the current release before fitting more memory; early revisions had DDR4-3200 compatibility issues.
- Enable XMP for DDR4-3200 and confirm the board trains at 3200 with four DIMMs, not just two. Four sticks often drop to a lower supported speed – record what it actually runs at.
- Enable Resizable BAR / Above 4G decoding if the GPU supports it, set the boot order and confirm the second NVMe is enumerated.
- Set fan curves for a 35 °C ambient rather than the default silent profile.
Verification
- 30-minute combined stress test (CPU and GPU) with temperature logging; nothing should throttle within the first ten minutes.
- Measure the 12 V rail under load with a DMM at a Molex or PCIe connector: ATX allows ±5%, i.e. 11.4–12.6 V. A rail sitting at 11.5 V under load indicates an ageing PSU.
- Test every front-panel USB port, audio jack and the reset/power buttons, and check the front-panel headers are on the correct pin groups.
Documentation
- A build sheet with component serial numbers, BIOS version, memory speed achieved, stress-test temperatures and rail voltages.
- A short note to the customer on the 2 × 8-pin requirement if a 3070-class card is fitted later, and a reminder that the PSU warranty is void if it is opened.
- Handling notes: ESD wrist strap during assembly, anti-static bag for the old parts, and the removed 550 W unit returned if it is not reused.