Does memory errors reported in HWINFO actually indicate a problem?

[Toxicable]

So I've been overclocking my cards and i've noticed that once I go up to about 2100 on the memory frequency I start getting memory errors reporting on HWINFO, about 1000 errors over 5 minutes but claymores doesn't complain so I'm wondering if this is something to be concerned about?

[Vann]

Yes, overclocking too much will cause hardware errors, which will cause the miner to report more valid shares than you are actually getting credit for. Older miners like sgminer and cgminer used to report HW errors. I don't know why miners like Claymore's stopped reporting them.

[Toxicable]

Yes, overclocking too much will cause hardware errors, which will cause the miner to report more valid shares than you are actually getting credit for. Older miners like sgminer and cgminer used to report HW errors. I don't know why miners like Claymore's stopped reporting them.

Ahhh okay thanks. I've clocked it back to 2075 which reports about 1-5 errors over a 30 min period, do you think that's fine?

[Vann]

Ideally you should overclock just to the point where it's stable and without HW errors. It's not just about getting the maximum possible hashrate out of your cards. Stability and power consumption are also a factor. Typically the more more you overclock, the more power is used, so there is a diminishing return point where overclocking is just pushing the cards over their capacity for little to no benefit in your bottom line.

[miropp]

I have many rigs with the same cards (the same part number, the same series of serial numbers, the same memory chip) and the same settings (core / memory). Usually in one rig, 3-4 cards have 0 memory errors, 1-2 cards have about several hundreds errors and 0-1 has thousands errors (HWINFO64). What with this? Underclock this cards with memory errors or sell them and buy others?

[Toxicable]

Ideally you should overclock just to the point where it's stable and without HW errors. It's not just about getting the maximum possible hashrate out of your cards. Stability and power consumption are also a factor. Typically the more more you overclock, the more power is used, so there is a diminishing return point where overclocking is just pushing the cards over their capacity for little to no benefit in your bottom line.

From the calculations i've done it's more profitable for me to get a higher hashrate rather than balance the power usage out, however stability is still important. As mentioned above with 2075 mem freq I get 29.6 MH on ETH and 860 MH on DRC at 150W which I think is reasonable enough aslong as those 1-5 errors per 30 min isn't an issue

[Arthas30]

I've got 2 Asus RX 560, one works without memory errors and second one has 9000 errors every second, both work on stock settings with modded bios at 13MH/s