- Nov 23, 2001
- 6,712
- 142
- 106
I just wanted to start a thread for those with 32GB dimms to see what kind of overclocks you're all getting.
I have the 18ASF4G72AZ-3G2B1 with Micron Rev.B dies.
Here are results from my testing:
3400MHz 17-20-17-20-41-61 @ 1.30v
3400MHz 16-20-16-20-40-60 @ 1.36v
3400MHz 15-20-10-18-36-54 @ 1.42v (didn't try the lower rcd_wr with the others)
3466MHz any timings, are mostly* stable around the same voltage. ie: 2-3 ecc errors per day.
3533MHz same as 3466, just 12+ ecc errors per day
3600MHz booted, but would reboot itself or error out fast
All with tRFC @~313ns
What i'm seeing is CL scales well with voltage, but the rcd/rp/rfc don't scale at all.
The max MHz seems to be limited by signal integrity and/or clock skew on the command/address/chipselect/cke fly-by busses.
So moving the setup delay to the lowest for the CAD settings and increasing the drive strength to 20 is needed to boot/stabilize 3400MHz or higher.
I can't set them lower than 1 with a range of auto to 63. AMD tech docs seem to indicate that "auto" might be 32.
The pcb on these modules was not designed for high MHz, so it seems to make sense that they'd design about a ~10% margin of safety as far as signal integrity or skew is concerned. The chips themselves could likely clock 3600+.
The RCD/RP limitations are inherent to the chip design imo. wordline, sense amp, I/O, or support circuitry limitations
Also the wordline voltage can be lowered to 2.34v, raising it don't help anything.
I noticed with my hynix AFR 16GB modules, that raising vpp actually helped stabilize tighter timings. Micron is different.
Anyone else playing with 32GB modules ?
I have the 18ASF4G72AZ-3G2B1 with Micron Rev.B dies.
Here are results from my testing:
3400MHz 17-20-17-20-41-61 @ 1.30v
3400MHz 16-20-16-20-40-60 @ 1.36v
3400MHz 15-20-10-18-36-54 @ 1.42v (didn't try the lower rcd_wr with the others)
3466MHz any timings, are mostly* stable around the same voltage. ie: 2-3 ecc errors per day.
3533MHz same as 3466, just 12+ ecc errors per day
3600MHz booted, but would reboot itself or error out fast
All with tRFC @~313ns
What i'm seeing is CL scales well with voltage, but the rcd/rp/rfc don't scale at all.
The max MHz seems to be limited by signal integrity and/or clock skew on the command/address/chipselect/cke fly-by busses.
So moving the setup delay to the lowest for the CAD settings and increasing the drive strength to 20 is needed to boot/stabilize 3400MHz or higher.
I can't set them lower than 1 with a range of auto to 63. AMD tech docs seem to indicate that "auto" might be 32.
The pcb on these modules was not designed for high MHz, so it seems to make sense that they'd design about a ~10% margin of safety as far as signal integrity or skew is concerned. The chips themselves could likely clock 3600+.
The RCD/RP limitations are inherent to the chip design imo. wordline, sense amp, I/O, or support circuitry limitations
Also the wordline voltage can be lowered to 2.34v, raising it don't help anything.
I noticed with my hynix AFR 16GB modules, that raising vpp actually helped stabilize tighter timings. Micron is different.
Anyone else playing with 32GB modules ?
Last edited:
Facebook
Twitter