1.5V on 1.65V ram... how fast?

dosmastr

Distinguished
Jan 28, 2013
191
7
18,695
Hi,

Got haswell and some g.skill ram rated for 2133. did research which indicated that running XMP over long term could burn out the haswell FIVR as 1.65 is more than 5% over the spec... so currently its running 1600 and 1.5V
memtest says its good in current config... but just curious what you think I might be able to push it to and still be safe.

In other news I had some 1866 running 1.5V and memtest was like DUUUUDE-- 1 error per second minimum... oddly it only bsod'ed once a month. backed that one down to 1600 (stupid bios had no voltage adjust for ram!) no issues

think I'll hit 1866 without issue or am I dreaming?
 
Solution
1.5 is the recommendation for running with stock 1600 DRAM (I stick with it at 1866 also), higher freqs require more voltage and are perfectly safe for 2133 and up look for 1.6-1.65, Intel even certifies these type sticks, so you'll be fine

Tradesman1

Legenda in Aeternum
1.5 is the recommendation for running with stock 1600 DRAM (I stick with it at 1866 also), higher freqs require more voltage and are perfectly safe for 2133 and up look for 1.6-1.65, Intel even certifies these type sticks, so you'll be fine
 
Solution

dosmastr

Distinguished
Jan 28, 2013
191
7
18,695
ug... as always I can't find the article I read.. google is fail....

I read a couple places that intel says yes its good under XMP but then also always says how the jedec spec is 1.5V and that's whats recommended for their procs. Further pcpartpicker popped up a warning when I added it.

all that fine, But I read a reviewer (anand?) had a mysterious cpu failure and all he could figure was the FIVR was running out of its spec and caused it to burn out, and thus cited the above and then said "well, haswell doesn't really see much improvement from increase memory speed anyway so just stick to 1600"

i'll give 1866 a shot at 1.5 and see what happens.
 


1.5 volts is the JEDEC standard and should be observed when stability is of paramount concern, such as in a production server. However, if performance is desired it is perfectly safe to take it up to the maximum tolerated by the memory controller or the maximum tolerated by the DDR3-SDRAM IC, which are around 1.8 volts and 1.9 volts respectively before damage is incurred. If the FIVR on the Haswell microprocessor failed, then it's a dud chip.