# The memory-slot reads behind the page's memory column

*Speeds from RAM alone, compared* (https://research.strata2signal.com/ram-speeds-compared/) prints each machine's
memory as sticks, rated speed and configured speed, and works out a paper peak from the configured speed. This file
says where each of those came from. Three machines' slots were read with `dmidecode -t 17`, the read-only utility
that prints what the motherboard's firmware reports for each memory slot; for the other two the page says the
memory was not read, and this file says the same. Dates are UTC. No serial number, asset tag or machine name is
printed here.

A paper peak is arithmetic, not a reading: the configured transfer rate (MT/s) × 8 bytes × the number of channels.

## EPYC 7532, 128 GB (the eight-channel desktop of *Speeds from RAM alone*)

Read 2026-11-03 by the RAM bench's own slot read on that machine, the run behind exhibit 105's eight-channel rows.

| Slot | Channel | Size | Type | Rank | Rated speed | Configured speed |
|---|---|---:|---|---:|---:|---:|
| DIMM 0 | A | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | B | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | C | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | D | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | E | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | F | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | G | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |
| DIMM 0 | H | 16 GB | DDR4 | 2 | 3,200 MT/s | 3,200 MT/s |

Eight 16 GB dual-rank DDR4 sticks, one per channel on eight channels, each configured at its rated 3,200 MT/s.
Paper peak: 3,200 × 8 × 8 = 204.8 GB/s. That read printed no part numbers.

## i5-12600H, 32 GB (the mini PC)

Read 2026-11-08 with `sudo dmidecode -t 17`.

| Slot | Bank | Size | Type | Rated speed | Configured speed | Manufacturer | Part number |
|---|---|---:|---|---:|---:|---|---|
| Controller0-ChannelA-DIMM0 | BANK 0 | 16 GB | DDR5 | 5,600 MT/s | 4,800 MT/s | Micron Technology | CT16G56C46S5.M8H1 |
| Controller1-ChannelA-DIMM0 | BANK 0 | 16 GB | DDR5 | 5,600 MT/s | 4,800 MT/s | Micron Technology | CT16G56C46S5.M8H1 |

Two 16 GB DDR5 SO-DIMMs, one on each memory controller, rated 5,600 MT/s and running at 4,800 MT/s. Paper peak at
the configured speed: 4,800 × 8 × 2 = 76.8 GB/s.

## 3700X, 32 GB DDR4 (the two-channel desktop of *Speeds from RAM alone*)

Read 2026-11-08 with `sudo dmidecode -t 17`.

| Slot | Bank | Size | Type | Rated speed | Configured speed | Manufacturer | Part number |
|---|---|---:|---|---:|---:|---|---|
| DIMM 0 | P0 CHANNEL A | 8 GB | DDR4 | 3,200 MT/s | 3,200 MT/s | Unknown | CMWB8G1L3200K16W4 |
| DIMM 1 | P0 CHANNEL A | 8 GB | DDR4 | 3,200 MT/s | 3,200 MT/s | Unknown | CL16-20-20 D4-3200 |
| DIMM 0 | P0 CHANNEL B | 8 GB | DDR4 | 3,200 MT/s | 3,200 MT/s | Unknown | CMWB8G1L3200K16W4 |
| DIMM 1 | P0 CHANNEL B | 8 GB | DDR4 | 3,200 MT/s | 3,200 MT/s | Unknown | CL16-20-20 D4-3200 |

Four 8 GB DDR4 sticks, two per channel on two channels, each configured at its rated 3,200 MT/s; two part strings,
one stick of each on each channel ("Unknown" is the manufacturer field as the firmware reported it). Paper peak:
3,200 × 8 × 2 = 51.2 GB/s. This read is the one behind exhibit 105's dated correction of 2026-11-09 (four 8 GB
sticks, not two 16 GB). The same machine's triad read 21.43 GB/s best on 2026-11-04 and 19.50 on 2026-11-08; why it
reads so far under the paper peak is not in the record, and neither this file nor the page guesses a cause.

## 290HX Plus, 64 GB (the laptop)

Not read by a bench run. The page's "two 32 GB SO-DIMMs" and "5,600 MT/s" are the house's hardware record for this
laptop, not a slot read, and no file in this kit carries them. Its paper peak on the page (5,600 × 8 × 2 = 89.6
GB/s) is arithmetic from that record and says so where it is printed.

## i9-10850K, 48 GB (the older desktop)

Not read. The page prints its memory as 48 GB of DDR4 (3,200 as built; never read) and 47,040 MiB as the system
reports it, the second from the older desktop's own result files (`free_m_before`, in `older-desktop/`). No triad
was run on it, so it has no paper peak or bandwidth figure on the page.