Anyone know why Phoenixminer lowers the core clockrate on some GPUs (and not others of the same model on the same rig) after they start up and successfully hash at ~50 MH/s for a few minutes? I am starting all my RX 5700s at 1300/900 (core/memory) and they all start out at ~50 then for some reason PM lowers the clock rates on some which end up hashing at around 40 MH/s or less. I am setting GT value at startup so it's not auto-tuning. Any thoughts?
Have you tried letting it autotune? ive noticed that sometimes the hardsetting of GT values results in lower hash on subsequent boots. Since most of my rigs are stable i dont bother setting and get consistently high hashes. 53-54mhs without bios mod on 1350/920/775. With bios mod about 56.7mhs per 5700xt...but i need to spend some time tuning them...57-60 should be achievable.
Thanks. Yes, same result when I let it auto-tune. One thing I just thought of is it might be SMOS that is lowering them although I haven 't seen anything in the SMOS interface that suggests that. Not sure if I can set the clock speeds directly in PM but will look into that next...
Interesting. I get slightly better total hashrate when I set the clocks and voltage directly in PM but the clocks still drop in the first minute or so on a few individual GPUs. I'm now thinking it could be the drivers; PM documentation mentions that the drivers sometimes don't hold the specified voltages but maybe that applies to clocks too? Not sure...
Its unlikely the miner is lowering clock rates after startup, I doubt that's in its capabilities - it should need to restart with different settings. Starting with the 5000 series cards there aren't the old p-states (p0-p8) of set timings but an curve of timings related to clockrate, voltage, and other factors. So the "set" clock rate will vary and due to differences in each card you can get different end timings with similar initial settings. I've got one 5700 that differs quite a bit from my others so to get the desired clockrate it (the driver/card) sets a higher voltage or to get the desired voltage the clock rate is lower.
Are you running PM with the -hstats 2 option? This shows the actual current clock rates on each card in the display along with 3 temperatures (gpu, memory-junction, and memory). I would check these values after start up and then after 5min. My 5700 I usually see a sight ramp-up in clockrate from about 1240 to 1265 after about 5mins.
I am now, but it looks like SMOS was reporting the same number only averaged for some number of seconds so I think I have a handle on the actual clock speed being used so I am now focusing on just one GPU specifically that exhibits this behavior. So just to be clear I am now setting the clock rates and voltages using the following inputs to PM:
-cclock 1300 -mclock 900 -cvddc 800
At startup SMOS shows those exact values and a hashrate of ~52 MH/s. After about five minutes the core clock drops to between 845 and 880 and changes every few seconds but is now hashing at only ~36 MH/s.
Seems like some loop is targeting the core clock to compensate for something it perceives to be wrong or suboptimal in some way sacrificing hashrate in the process. What I don't understand is 1) which program is doing this (by process of elimination I'd guess it's the driver) and 2) if it is capable of hashing at 52 MH/s why can't I (or how can I) just force it to retain those values and either crash or start throwing HW errors or something. My best guess at this point is that the driver could just be designed to protect the card at all costs and maybe the chip in this particular one is from a lower quality bin?
I can add that the card I'm focusing on is a Msi RX 5700 XT 8GB, but as I've said before others of the same make and model are running fine in parallel on the same rig.
Hmm, one other thing I just noticed, of all the cards on this rig (14 currently) the ones showing this behavior are all MSI XTs. In contrast all the XFX XTs and MSI XLs seem to hold clock and hashrates much better so I guess it could be a problem with these particular MSI cards. Maybe the next thing to try is to reflash them but I have not done that before so....