gtx 780 3gb

[mystro1213]

i used to mine on it at 21mhs now its 7 mhs
used all drivers old and new i need help
cpu i7 4770k
8gb ram
gtx 780 3gb

[ZiDanRO]

What had changed between 21 and 7mhs? You installed something, another ethermine version etc?

[jean-m-cyr]

I don't get much more than 10 with an old gtx 980 ti, so I can't see a 780 doing 21.

[arainchik]

Why is that? According to https://www.videocardbenchmark.net/high_end_gpus.html 980 and 970 should be on par with RX580, or even faster.

Why do we see such drop in performance on 970 and 970 cards as DAG size is increasing?

[jean-m-cyr]

passmark g3d is a directx video graphic test that focuses on the card's rendering ability, more specifically vertex and pixel shading. Not a very good indicator of mining performance. I'm sure there are sites out there with more relevant info. I have an old 980 ti which I don't don't mine with since the new 10 series are nearly twice as power efficient. That card, not overclocked, would get about 10 Mh/s.

[chfast]

Can you run a simulation mode ethminer -G -Z? My 280X was mining with 28 Mh/s, now is doing 5 Mh/s. The DAG size affects the performance a lot.

[jean-m-cyr]

Interesting! With 'ethminer -U -Z' on the 980 I get 19Mh/S, but the same cards only reports about 9.5 Mh/S when really mining? Double counting in simulation bug?

[chfast]

Try ethminer -U -Z 5000000.

[mystro1213]

chfast thanks man its works getting 21mhs/s

[chfast]

@mystro1213 what works? If you get 21 mh/s with ethminer -U -Z 5000000 that's actually something is wrong.

[jean-m-cyr]

Try ethminer -U -Z 5000000.

Ah ok. Adjusted DAG size.
Confirmed. With this the 980 says 9.5Mh/S... same as live.

[arainchik]

There is a dramatic difference between -U -Z 4000000 (17MHs/) and -U -Z 4500000 (11MHs/) on my 970.

I think it's somewhere betweeb 4200000 and 4300000 blocks.

When I'm looking at GPU utilization stats with "nvidia-smi pmon" you can see that with 4000000 block I get 100% sm and men utilization, but when I do 4500000 sm stays at 100% but mem drops to 65%

Try it, run "nvidia-smi pmon" in a separate window when running different block sizes.

[ddobreff]

On older maxwell/kepler GPUs page_size or as AMD put it fragment_size is default 64kb thats why your hashrate is dropping, if you mine ethash forks with <2GB DAG you'll be fine.
P.S. If you use amdgpu driver instead of fglrx you might be able to revive some of the lost hashrate on Tonga/Tahiti.

[arainchik]

Block number 4,000,000 has 2.04Gb DAG size, so it's > 2Gb, but still get's decent hash rate.

I don't understand how 64Kb page size can have any impact on that...

I'm still wondering if an updated CUDA kernel can bring back 19MH/s on 970/980s :)

[Qwertzi01]

I have the same issue with my GTX 970 4GB without OC -> benchmark AND local mining are both around 22 MH/s. But when I start mining in pools like ethermine then the performance drops to 6.5 MH/s around. I don't understand this problem!?

[arainchik]

Qwertzi01 - benchmark on current block/DAG: ethminer -U -M 4700000 or ethminer -U -Z 4700000 - you'll get the same speed you are getting in pool mining. You are benchmarking on DAG size from years ago :)

[DeadManWalkingTO]

I think this issue can be closed.