Memory overclocking 1060 (Samsung)

next-Jin

Supreme [H]ardness
Joined
Mar 29, 2006
Messages
6,633
I have 5 1060s and have been out of the GPU game for awhile. (Switched to Macs in 2012).

I see posts with varying degrees of people's clocks (ex. 2000, 9000, etc), what should I aim for with Samsung memory for its technical spec?

Folks say it's 9000 but that's not how it's displayed in Afterburner or GPUZ iirc.

I'm at +500 rock solid atm

I'm mining Ether and Burstcoins between 2 systems and have 21.8 or so MHs between them with 80% power limit and -200 on the core.
 

Nexillus

[H]ard|Gawd
Joined
Nov 6, 2012
Messages
1,116
If you are OCing within windows, I would try doing -100 to -200 and Memory around +700 to +950 (seen different levels based on micron and Samsung) With Samsung it usually OCs better)
 

PiERiT

2[H]4U
Joined
Oct 8, 2010
Messages
2,548
I think Afterburner halves it, meaning 9000 would be 4500. So you're already there with +500.
 

next-Jin

Supreme [H]ardness
Joined
Mar 29, 2006
Messages
6,633
See what I'm saying lol, everyone's comment had completely different values.

I'm sitting at +600 in Afterburner at +700 I think it locks up (machine is headless and I teamviewer in)
 

Nexillus

[H]ard|Gawd
Joined
Nov 6, 2012
Messages
1,116
I've got two 1060's running +1750 and a third running +1500.

Are you running on Windows or Linux? There is a huge difference in OC between the two OS's because of how the drivers are. On Linux I see standard OC between 1100-1700.
 

BinarySynapse

[H]F Junkie
Joined
Feb 6, 2006
Messages
14,978
Are you running on Windows or Linux? There is a huge difference in OC between the two OS's because of how the drivers are. On Linux I see standard OC between 1100-1700.


I'm running Windows 10. I was running on Ubuntu 17.04 with 375.xx drivers, but I didn't try to overclock then.
 

Vaulter98c

[H]ard|DCer of the Month - October 2009
Joined
May 21, 2008
Messages
5,730
Just keep taking the memory higher, let it run for an hour or so, take it higher, let it run for an hour or so. Every few bumps up let it run for a longer stretch, because some things will be stable for a short time but crash long run. Eventually when you hit a wall back it down a bit and run it there for a whole day. Once you have that number figured out, start lowering your power and keep an eye on the hash rate. You can drop the TDP pretty far and only lose like 1 MH or so, but you'll gain that back in efficiency by lowering your power bill and keeping things cooler

People have difff numbers because every card will be different, some will OC well with crap memory, some will OC crap with good memory, you get the idea. Just bump up each card slowly and find a good "wall" for each then start to reel it back in on clock and power side until you find a happy balance you can live with

I hope you don't mean you're getting 21 MH between all of your cards tho, because that's WAY off

OC'ing is kind of in the eye of the beholder, you can go as far or no so far, all up to what you're end goals are with heat and power usage and stress on the cards. Lots of variables play in and thats also part of why everyone has diff numbers
 

Vaulter98c

[H]ard|DCer of the Month - October 2009
Joined
May 21, 2008
Messages
5,730
Also, you can look into using the rc client to try and bump your numbers up, I believe it was designed with the 1060 in mind but using the RC I took my 1050 TI ITX card from 11.5/13.2 all the way to 13.8/15 just by switching the miner I used. It does less for 1080 but if you have 1060's I'd give that a look
 

Nexillus

[H]ard|Gawd
Joined
Nov 6, 2012
Messages
1,116
Just got my 1060s yesterday and compared to my 1070s, they OC a lot better on Memory, currently doing +1700 on 5 cards and 1650 on one card. Long term stability is not known yet but will be TBD.
 

next-Jin

Supreme [H]ardness
Joined
Mar 29, 2006
Messages
6,633
People have difff numbers because every card will be different, some will OC well with crap memory, some will OC crap with good memory, you get the idea. Just bump up each card slowly and find a good "wall" for each then start to reel it back in on clock and power side until you find a happy balance you can live with

I hope you don't mean you're getting 21 MH between all of your cards tho, because that's WAY off

OC'ing is kind of in the eye of the beholder, you can go as far or no so far, all up to what you're end goals are with heat and power usage and stress on the cards. Lots of variables play in and thats also part of why everyone has diff numbers

21MH average, nah I don't mean variances in OC on each card. I'm talking about how people are reporting their numbers.

Just looking at this thread there are very different ranges.

If I say I'm +500 on memory if the base clocks were 1500 you'd think 2000. That's not what I'm seeing. Someone might say they are +500 and report 2225 or some other number.

Iirc Pascal 4x's the memory so 2000 base would report 8000, it's just confusing when people report all sorts of different numbers when we all should be within a certain average if that makes sense.
 

BinarySynapse

[H]F Junkie
Joined
Feb 6, 2006
Messages
14,978
GDDR5 uses three clocks, the command clock which is the base or reference clock, and two write clocks (WCK) that are twice the command clock.

Then because it's a DDR device, data is written to and read from the chip two times for every tick of the WCKs. So it's effectively 4x the command clock.

People are only reporting what their tools are telling them. The Gigabyte tool that came with my 1060s reports the data rate (8,008 MHz); so when I said +1750 for my 1060s, the context was the data rate not the command or write clock rates.

On the other hand MSI Afterburner for my 980Ti seems to be reporting the write clock rate (3,505 MHz). I haven't used one yet, but I'm sure there's a tool that reports the command clock rate
 

next-Jin

Supreme [H]ardness
Joined
Mar 29, 2006
Messages
6,633
GDDR5 uses three clocks, the command clock which is the base or reference clock, and two write clocks (WCK) that are twice the command clock.

Then because it's a DDR device, data is written to and read from the chip two times for every tick of the WCKs. So it's effectively 4x the command clock.

People are only reporting what their tools are telling them. The Gigabyte tool that came with my 1060s reports the data rate (8,008 MHz); so when I said +1750 for my 1060s, the context was the data rate not the command or write clock rates.

On the other hand MSI Afterburner for my 980Ti seems to be reporting the write clock rate (3,505 MHz). I haven't used one yet, but I'm sure there's a tool that reports the command clock rate

Oh wow ok I didn't know they reported different numbers. That makes sense then lol
 

xrayos

n00b
Joined
May 31, 2017
Messages
54
I have a couple of 1060's with Samsung memory, baseline w/o o/c I get 18MH/s using ethminer. after +900 memory o/c with afterburner, I get about 24-25 MH/s. Overall very happy.
 
Top