Apparently slow memory clock speed under Windows

Framework Desktop
1

Hello;

when I boot into Windows 11 (25H2), both CPU-Z and HWInfo claim that my memory clock is at 1000MHz, which seems a tad low for a system sporting LPDDR5-8000 RAM.

Are the sensor readings bogus, or is something limiting my memory speed?
If so, how can I unleash the full potential of my RAM?

BIOS is 03.04, AMD Chipset drivers are 8.01.20.513 (both should be current)

Cheers,
Thomas

8.01.20.513

2

Your mem speed is low because its idling. Use a memory stress tester such as MemTest64 on Windows, lower the “Parallel CPU Threads” by a half to avoid CPU bottleneck. Begin test then open HWINFO, your memory speed should be at or very close to 4000MHz

3

Hmm… so far I used CPU stress test like Prime95 and the like - still stuck at 1000.
I will fire up MemTest64 once I come home and come back with the results.

Thank you.

4

Thomas,

This is how the AMD intended the memory to work. There are a few things that you need to know how the CPU, GPU and Memory are wired up.

  1. The Memory is wired to the CPU through the Infinity Fabric to give 128Gb/s of bandwidth to each CCD (1 CCD = 8 cores, 16 threads) on the CPU.
  2. The GPU is wired up to the memory through the Infinity Fabric to give 256Gb/s of bandwidth.
  3. The Infinity Fabric bandwidth is shared between the CPU and GPU.

So when something like CPU-z show 1000Mhz it’s showing you the rate of data not the total bandwidth.

On the AMD Ryzen AI Max+ 395, the effective speed is much higher due to the unique architecture. LPDDR5X transfers data on both clock edges (×2 for DDR) and uses a 4× internal prefetch, giving an overall ×8 multiplier from the controller clock. So 1000 MHz × 8 = 8000 MT/s

This can be checked by using the latest version of GPU-Z and CPU-Z, as these have been updated to take into account Ryzen’s unique architecture.

5

Is it technically quad data rate (QDR)?

6

No it’s not QDR in that sense.

The Infinity Fabric is what talks to the DDR chips, the Fabric then passes the message to the CPU and GPU. The GPU is able to access full bandwidth and each CCD (CPU = 2x CCD’s) can access 1/2.

The Fabric is what also allows for the dynamic memory allocation between GPU and CPU as it’s all come from the same Fabric.