How to stress-test RAM/fclk overclock?

seward

n00b
Joined
Mar 6, 2014
Messages
32
I've got an AMD Ryzen 5 5600G cpu in an Asus Prime B550-PLUS motherboard, with Corsair Vengeance DDR4 3600mhz RAM. I play old games, and the igpu on the chip runs them surprisingly well. Now I'm trying to tweak the igpu, and that also involves RAM and fclk settings (fclk = igpu memory clock). I'm trying to raise fclk, and I'm also trying to maintain 2:1 RAM/fclk ratio. Somewhat to my surprise, I'm able to boot at 4000mhz/2000 fclk with this RAM. All other settings were left on Auto; RAM volts at 4000mhz are the same as 3600mhz (1.35v), and only two timings loosened when I went from 3600 to 4000 (Trc went from 84 to 93, Trfc went from 630 to 700), so it looks like a legit overclock - same timings at faster speed, with tolerable volts.

I've heard that Ryzen 5000 chips don't like fclk above 2000, so instead of looking for better performance RAM (my first idea), I'd like to see if I can keep these speeds with this RAM. First, though, I'd like to check stability. How do folks stress-test fclk settings? Will Memtest do it? Parts of Prime95?
 
It sounds like RAM/fclk stress-testing should involve simultaneous use of high-demand programs (e.g. 3+ hours of Prime95 Large FFTs and OCCT VRAM test at the same time). Is that how folks are testing their RAM/fclk overclocks?
 
Why bother with stress test, just use the PC for whatever use you have, if it crashes or has thermal issues back off the OC.

Your PC is a tool for business or whatever, too many peeps obsessed with measurements or "measurbators" as I call them.
 
Why bother with stress test, just use the PC for whatever use you have, if it crashes or has thermal issues back off the OC.

It might have corrupted your OS install and data by that point, though.

RAM corruption is very dangerous for filesystem contents.
 
When I attempted to run my X99 systems RAM at the vendor speed of 3200 the XMP profile killed my Win 10. XMP and X99 do not play nice together.
 
Your PC is a tool for business or whatever, too many peeps obsessed with measurements or "measurbators" as I call them.
Paulie OG 300A legend.jpg

:D

Obligatory joke aside, I'd first free as much RAM as possible and run Prime's blend.
Followed up by something like a batch photoshop job, that hammers away on huge bitmaps for a day straight.
An idea I haven't yet tried: big-ass ramdisk, beat the snot out of it with a continous r/w test (as if it were a block device)
 
Back
Top