Memory Help please, with memory timings for VLP-RAM (ADATA PC DDR4 2666MHz)

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
Progress; and with stock voltage settings; But I guess the timings are what you would call loose;


 
  • Like
Reactions: anothernoob

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
Running Timings (16, 16, 17, 18, 38, 66) 1.38v-DRAM-VOLTAGE @3266MHz.
Not quite sure what is best to test the memory, I will do some more reading.

 

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
Made a USB Boot Drive from the PassMark MemTest86 v8.4 Free, and booted that to test memory. I am deep into the second of four passes with Zero Errors (and typing on her InWin Chopin) so I expect the other lengthy passes will also succeed with Zero Errors.

A couple of oddities about GigaByte BIOS settings;
  • The Dram voltage ends up slightly higher than what you set, both in a summary to the right of BIOS Dram Voltage setting, and on HWiNFO. So you might want to compensate for that extra small bump-up. One good way to observe this is to Boot-into-BIOS a second time and look at the new summary.
  • There are two places in BIOS for changing the memory timings. One of those places (Settings/AMD Overclocking) does nothing??
Thanks immensely for the help with AMD memory tuning. That was like a foreign language to me.

Oh btw> I said that I could not make this build get hot. That was wrong. When I let Prime95 run for a long time it did get to 80c. One thing which I am curious about. When you do get high temps, is it okay to allow temperature throttling to occur? Or should you hurry to shut it down?
 

JJSEA

Caliper Novice
Dec 30, 2019
23
7
I think something's not working quite right on your system because my system benchmarks faster using 2666MHz memory (with same CPU). I get read/write of 37/39 and a time of 171.

One possibility is that there are ECC errors that are being corrected: check in Event Viewer > Windows Logs > System. Then do Filter Current Log with an Event source of WHEA-Logger.

Another possibility is that FCLK is mismatched. HWINFO shows this as Infinity Fabric Clock. You want this to be the same as the memory clock (e.g. 1600MHz if you are running your RAM at 3200MHz).
 

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
  • No ECC Errors logged, see below.
  • Timings (16, 16, 17, 18, 38, 66) Bios DRAM-VOLTAGE of (1.35 = HWiNFO 1.38) @3266MHz.
  • Infinity Fabric clock/FCLK set to Auto to equal MCLK @2x1633=3266 MHz (see HWinfo below).
  • DRAM Calculator/MEMbench still running slower than yours @(read=28, write=24, time=217).
  • Tried higher DRAM voltage (1.38 = HWinfo 1.416) with same results, still slower than yours.

 
Last edited:

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
Hmmmm; I have done a lot of head-scratching over our different results from the Ryzen DRAM Calculator 1.7.3/MEMbench results:
  • RAM size (upper left) says 12000 and I have 2x8GB sticks = 16GB??
  • Version (bottom right) says this is a beta version (MEMbench 0.8 beta 4) 😮
  • If you hover over MEMbench mode upper-left and it says Mode-Easy (which was my test results) is for systems with 8GB or less and Mode-Default is for systems with more than 8GB. So my first test was using the wrong mode.
Bottom line is that I am not sure if I can trust the Ryzen DRAM Calculator 1.7.3/MEMbench beta results to compare RAM speed of different systems. Though, I could be wrong.

Here's the results from Ryzen DRAM Calculator 1.7.3/MEMbench using Mode-Default below. Also as an aside; I tested these timings (16, 16, 17, 18, 38, 66) Bios DRAM-VOLTAGE of (1.35 = HWiNFO 1.38) @3200MHz and @3266MHz. The results were the same.

 

JJSEA

Caliper Novice
Dec 30, 2019
23
7
Hmm. I'm using 2x16Gb (2666MHz CL19, the v-color ones mentioned earlier in this thread). These are dual-rank; I wonder if that has anything to do it. What do you get for read speed/write speed if you use the default settings (no overclocking)? We could also try the Cache and Memory benchmark in Aida64 (I get read speed of 39GB/s).
 

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
I got faster, (44GB/s compared to your 39GB/s).

But I am still skeptical about our memory-read tests being an apples-to-apples comparison because the test is on top of whatever else our systems are running. For example, I could also be running Firefox with 50 Tabs opened that are consuming extra memory and power??


 

JJSEA

Caliper Novice
Dec 30, 2019
23
7
I closed all other apps before I did the tests.

44Gb/s sounds reasonable, given that your memory is at 3233MHz and mine is at 2666MHz. I wonder why there is such a big difference in read speed on your system between membench and Aida64, whereas on my system they are quite close.
 

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
I have MemTest86 on a bootable flash drive. I am thinking of letting MemTest86 run through one cycle of the ten or twelve different memory tests and check to see how long it takes to complete one cycle of tests. If I recall correctly it provides the total time it took to run the tests. I will give it a try and post results.
 

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
MemTest86 runs tests 0 through 10 plus test-12 Hammer Test. It defaults to 4 passes of those tests but I was able to modify the number of passes to 1. The total time for one pass was 43:55/43 minutes and 55 seconds. Unless the tests are somehow random, I should be able to run another pass and come up with a similar time, which I will try to duplicate tomorrow, it's getting late.


 

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
I have MemTest86 on a bootable flash drive. I am thinking of letting MemTest86 run through one cycle of the ten or twelve different memory tests and check to see how long it takes to complete one cycle of tests. If I recall correctly it provides the total time it took to run the tests. I will give it a try and post results.
Scrap that idea!! It takes too long and my results on the second test did vary. Argh, I lost the 2nd test results, doesn't matter. Here's what I think might work better.

You can select which MemTest86 tests to run. I selected tests 0-6 for (7 tests total) and I changed the number of passes from 4 passes to 1 pass. It ran in 4:23. I ran it again and it ran in 4:24.

Tip: Keep your CLR CMOS tool handy when you switch to boot from USB flash drive.
Tip2: Store your Overclocked memory timings in a Saved Profile to get them back after CLR CMOS and set DRAM Voltage again because it is not stored in the Saved Profile.

 
Last edited:

PVC

SFF Lingo Aficionado
Original poster
Jul 12, 2020
132
39
Hmm. I'm using 2x16Gb (2666MHz CL19, the v-color ones mentioned earlier in this thread). These are dual-rank; I wonder if that has anything to do it. What do you get for read speed/write speed if you use the default settings (no overclocking)? We could also try the Cache and Memory benchmark in Aida64 (I get read speed of 39GB/s).
Back from documenting/logging my build while the details were still fresh. I am less busy now and finding time to get back to these memory timing issues. Thanks for your patience and help.

After BIOS (Load Optimized Defaults) to get back to default settings;
Speed/Read/Write/Total = (DEFAULT = 2666MHz/27/21/247) vs (OVERCLOCKED = 3266MHz/28/24/217)

ETA> Oh btw/FYI; When I click Calculate FAST on DRAM calculator it says Error/Coming soon

 
Last edited: