I got my 2 Sapphire 470 Nitro+ 4GB with Samsung rock solid @ 29.5MH/s on eth.
This is on Linux box... Maybe win would be a bit faster.
I got 30+ but it's not that stable...
Can you post the strap you are using? I'm not having much luck getting my Sapphire 470 over 27Mh. That's running memory at 1900 with the 1500 strap. Using the 1375 strap gives me errors mining ZEC.
Here's the 1375 & 1500 straps from the stock BIOS:
333000000000000022CC1C008CCD593AC0570D13AA09B4070048C7007A0014207A8900A002000000140D272E7B252B15
555000000000000022CC1C00AD515A3EC0570E142D4A64080048C700030114207A8900A003000000150E2A3186272E16
My best results with Samsung 4GB ( K4G41325FE ) have been using the 1625mhz strap at 1950-2050mhz but I'm still trying to determine long-term stability for each individual card
stock 1625mhz strap:
555000000000000022CC1C00CE596B44D0570F1531CB2409004AE7000B0314207A8900A003000000170F2E36922A3217
1150 / 2000mhz = 28.1mh (MSI 470 Gaming 4GB)
1150 / 1900mhz = 27.5mh (MSI 470 Armor 4GB)
Using Claymore in Windows, 16.11.5 drivers
I've noticed some odd behavior though... on my initial attempt I was able to gradually raise clocks on one card as high as ~2100mhz (I saw 29.5mh!), but it only seems to work if I raise the clocks while miner is already running and then pause/unpause, the initial miner / DAG loading does not seem stable with these timings at high clocks.
My other test card I seemingly had mining stable up to 2050mhz, only to find that it can't create the DAG at anything above 1965mhz without crashing the system
I've also noticed that the Samsung 4GB equipped cards doesn't seem to report any GPU memory errors in Hwinfo64, even when memory clocks are pushed high enough to cause errors / crashes in Claymore.