Bitcoin Forum
June 28, 2024, 11:51:38 AM *
News: Latest Bitcoin Core release: 27.0 [Torrent]
 
   Home   Help Search Login Register More  
Pages: [1]
  Print  
Author Topic: GPU memory load (=hash rate) low with more than one GPU installed  (Read 78 times)
vollmilch-trinker (OP)
Newbie
*
Offline Offline

Activity: 1
Merit: 0


View Profile
February 03, 2018, 09:32:48 AM
 #1

Hi there,

My system runs smoothly with one GPU installed (hash rate as expected), but drops significantly with the second GPU installed.

The setup is fairly standard:
- MSI Z270-A, G4400, 4GB RAM, 120GB SSD
- Claymore on Win10 with AMD BC driver
- 2x Powercolor RX580 RD GS (Samsung/Micron) no bios flash
- 400W bequiet

With either of the GPUs installed, I get out-of-the-box 24.5 MH/s ETH only on the Samsung and 27.0 MH/s ETH only on the Micron. With Claymores Dual Mining Mode I can add 600 MH/s (SIA). VRAM Overclocking (2000>2250MHz) gets 27.5 and 29.0 MH/s resp. GPU clock and voltage is on std settings. That's the values I'd expect.
Now, running both cards simultaneously has a combined performance of 36.0 MH/s ETH only (16.5 on Samsung, 18.5 on Micron) with std. mclock. Obviously I'd rather see somewhat >51 MH/s.

So what is wrong here? exclusion principle:
- Bios setup: set to "above 4G for mining", changing PCI latency and PCI Gen doesn't change anything - OK!
- Riser cable: same rate w/o riser - OK!
- Power supply: I switched to a 600W bequiet, but no change - OK!
- CPU: load low - OK!
- RAM: The system gets slower after a couple of hours. The 4GB RAM fills up til nothing is free. This might cause the slower windows system. The miner also finds less shares although reported hash rate is stable. However, virtual memory is set to 16-32GB. I'll get another 4G RAM, but I think thats not the bottle neck here. - OK!
- Internet connection: latency is around 150ms, however same hash rate with cable connection (30ms) - OK!
- GPU: GPU-Z sensors a 69% (Samsung) and a 73% (Micron) GPU memory load (GPU load 100% both)! The mem load on the Samsung sometimes suddenly drops to 0% for a second, Mircon is stable. Having just one GPU installed, both VRAM und GPU load is 100% stable. Now calculating for Samsung: 16.5 MH/s * 100%/69% = 24MH/s, i.e. if the memory load would be 100% I'd see the full 24 MH/s as in single mode.

So here we are: Why is the GPU memory load that low?

Any ideas really appreciated!!

best


 
Shazam!!!
Full Member
***
Offline Offline

Activity: 350
Merit: 158


#takeminingback


View Profile WWW
February 03, 2018, 01:45:22 PM
 #2

Move this to the Alt mining section for quick answers. There is an option to move it in the lower left corner of YOUR post.
Goodluck!!!

Click these links to learn some truth about Big Corporate mining pools stealing your money and centralizing BTCitcoin!!!
Help support the BTCitcoin community!!! Mine your BTCitcoin at a non-Corporate pool!!! BTC: 1ShazamjsPnpWDNnk3n2tAiKGMdXaSjay
I mine BTCitcoin at a non-Corporate pool with no hidden fees, no empty BTClocks: Kano Pool   ■ get Kano Pool signature banners here ■
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!