|
|
@@ -0,0 +1,47 @@
|
|
|
+
|
|
|
+# DDR5 Memory
|
|
|
+
|
|
|
+Besides the insane RAM prices due to AI gobbling up the market, I made a few big mistakes that I wanted document.
|
|
|
+
|
|
|
+
|
|
|
+## No Quad Channel
|
|
|
+
|
|
|
+First, I made a mistake with a RAM purchase back in 2024.
|
|
|
+
|
|
|
+I thought quad-channel was still the superior choice, but the move to AM5 apparently carried some big changes to specifications.
|
|
|
+
|
|
|
+Accordingly the maximum speed tops out at 3600 when running quad-channel, which is a significant performance hit.
|
|
|
+
|
|
|
+I did find it strange that nobody was selling quad channel kits, but at the time AM5 was new and nobody had clearly explained this limitation and I did not RTFM...
|
|
|
+
|
|
|
+I ordered two dual channel kits assuming that would work, and it did technically, but had a massive Code 15 at boot caused by setting EXPO 6000 when it could not operate above 3600.
|
|
|
+
|
|
|
+This forced it to run extensive memory checks at each boot to find a stable speed prior to actually starting. Sometimes it would take over 10 minutes.
|
|
|
+
|
|
|
+
|
|
|
+## Mixed Mode Refresh
|
|
|
+
|
|
|
+Apparently I encountered a second problem even after switching to dual channel memory.
|
|
|
+
|
|
|
+When running memtest86+ using EXPO 6000 I encountered errors.
|
|
|
+
|
|
|
+These errors disappeared if I ran in single channel, for either stick.
|
|
|
+
|
|
|
+I tried:
|
|
|
+
|
|
|
+- A different motherboard
|
|
|
+- Three different memory kits
|
|
|
+
|
|
|
+Each had basically the same errors.
|
|
|
+
|
|
|
+Turns out there is a new default called mixed mode refresh that messes with the tRFC1, tRFC2, and tRFCSb values.
|
|
|
+
|
|
|
+I managed to eliminate the instability by setting `tRFC2 = tRFC1`, then setting `tRFCSb = tRFC2 / 1.625`.
|
|
|
+
|
|
|
+
|
|
|
+## conclusions
|
|
|
+
|
|
|
+It never hurts to read the manual.
|
|
|
+
|
|
|
+DDR5 is expensive and volatile.
|
|
|
+
|
