2 Claymore
gtx1080tis are here, driver 425.31
-strap 1 - low 40+ hashrate, but system is stable
-strap 2 - need to apply only with std mem clock. Can raise mem clock after miner started and reach fantastic 56.5mhz but if miner closes then system crashes. Need to lower mem clock before exit.
I have my P2 cuda states disabled by the way to prevent memory clock spikes between different P-states switches.
-strap 3 - system crashes even with -502 mem clock.
So, it looks promising but unusable. Get back to EthPill and 55mhs stable and no crashes when miner starts or exits
2 Claymore. I think you should rework the way you do miner exit on gtx1080ti's with straps applied. I found a trick how to fix your timings and than run your miner without strap management. It keeps high speed of modded timing and doesn't crash system on exit. |
|
|
|
gtx1070 micron from 32.8 to 35 mhs - nice!
wow thats impressive, what memory setting though and what power limit Those are old palit gamerock and jetstream gtx 1070 cards with micron memory. They don't like aggressive downvolt with Curve (MSI Afterburner) below 775mv. So I need to use classic power target and gpu overclock to gain more efficiency ~1600+gpu clock, +550 mem, ~800mV, ~120watt according nvidia smi. settings are the same before was 32.8mhs now with claymore 35. Resulting gpu clock and voltage became lower automatically with high load ... |
|
|
|
|
gtx1070 micron from 32.8 to 35 mhs - nice!
|
|
|
|
|
2 Claymore
gtx1080tis are here, driver 425.31
-strap 1 - low 40+ hashrate, but system is stable
-strap 2 - need to apply only with std mem clock. Can raise mem clock after miner started and reach fantastic 56.5mhz but if miner closes then system crashes. Need to lower mem clock before exit.
I have my P2 cuda states disabled by the way to prevent memory clock spikes between different P-states switches.
-strap 3 - system crashes even with -502 mem clock.
So, it looks promising but unusable. Get back to EthPill and 55mhs stable and no crashes when miner starts or exits
|
|
|
|
Anyone have timings for frontier Vegas? They use different ram than the 64s and the bottom number is 350 not 260 like the Samsung vegas.
Most I can get by using the same settings is 46 MHz but stock with power play settings 1100 mem they get 42 so it’s not much of an improvement at all
Mine cryptonight-heavy algos with them. No need for dagger |
|
|
|
Contact me on PM. I solved this problem. Im not smart like DOCTOR but i solved this. use TRM and ignore DEVs that block programs that help us all like AMD MEM TWEAK
vega56 CNR= 2360HR (GPU1450mhz/870mv)(MEM 1100mhz/870mv)
Vega64 CNR= 2460HR (GPU1480mhz/870mv)(MEM 1100mhz/870mv) Nice results. Are you using kerney66 recommended timings or some custom ones? |
|
|
|
If you're lucky, you might find a Vega 64 vbios for that same card (if they made em both 56 & 64), which is unique.. not much out there, Pulse, only 56 afaik.
That however doesn't mean they're crappy, i've got a Sapphire Pulse with Hynix memory and it's running smooth at 47~Mh (memory at 975).
Got a Samsung Vega56 too and hits 48+Mh..
I have my Sapphire Pulse Vega56 Hynix one bios flashed to Nitro 64 for eth, it is working ok at 1040 mem clock than. But for cryptonight it is better to stay with original bios for low power with almost the same hash rate. |
|
|
|
@Eliovp "magic trick" is to disable p6 and p7 core state and stay on p5 core , p6 and p7 jump one soc clock step higher on some men frequency up to (1107soc) (example p7 core clock and p3 mem and your frequency is 830 hbm mem so your step is higher then 800soc so 847soc +"1 soc step" so you will get 900soc if you set p5 core you will get 847soc )
please explain this again , so yo usay on a 56 disable the p6 and p7 states on core then i can set higer memory than the 900 i seem to be capped at ? or you are saying by disabling those two the mem will automatically jump higer ? **edit*** @duck Hellen You are definitely on to something, I disabled the p6 an p7 states for my 4 card Hynix 56 rig and all of a sudden no more crashes or instability, power draw is also good 150 watts from the wall, I’m still at 1100/825mv and 900/825 Hash rate is 42 MHz on eth a bit higer than before. Timing are all the same only thing I did was disable p6 an p7 Can I attempt to bump up the mem now ? 940 mem clock is max I was able to get stable with hynix and std v56 1.25 mem voltage |
|
|
|
Gonna see same nice fried Vega )) Add some fan speed )) |
|
|
|
|
2 Claymore
Please, how to uninstall your driver and switch back windows from test to normal mode?
And about "paid straps from Eliovp" - I think everyone can examine them with Eliovp's tweak utility while your miner is running. Or are there some hidden gems?
|
|
|
|
Hy, my Vega'64 (Samsung) config for CNR and TeamRed miner. Draws more power but adds 1.8Khs.  your temps are high and hw errors in teamred miner is a bad sign |
|
|
|
Did you run the pill? What is your overclock, pls?
No pill, as I didn't notice that it gives any boost to Beam. OC in different for every card, for TI i use 0.8V and clocks can be 1670mhz or 1721mhz depending on card. In don't OC their memory, so its 5499mhz, so its even -5mhz - I noticed it give a little bit of boost over default 5503mhz in games\mining, probably there is timings step at 5500 and going over it by 3mhz is pointless. So for 1080 I set mem. clock to 5499 as well since it's basically the same gddr5x. For 1070 I tried to set 4498mhz mem for all, but some cards we unstable, so I lowered it until they stopped crashing. With core voltage you'll just have to find the value that will give you desired power consumption and then find max. stable clock for it(though raising clock would increase power consumption a bit as well:) ) It's just impossible to make a generalization on OC'ing polaris cards of the same model: some cards work 1746mhz@813mV and another 1860@881mV and later would have less power consumption... As I mentioned the pill, I meant gtx 1080 ti. It looks like TDP is about 70%. Palit Jetstreams 1080ti, 88% tdp, the pill and +200 mem clock, mem clock compute mode disabled in nvprofileinspector resulting 5702mhz effective mem clock. Gpu clock by curve and fixed to 1683mhz@775mv (afterburner) 42sols @ 195 watt |
|
|
|
|
2 GKumaran
Sorry, I remember I tried first samsung timings from TeamRed developers and they weren't good.
I didn't try second "safe" timings until twilightprotege's post.
And this second strap is really good for eth.
|
|
|
|
|
2 doktor83
Sorry, I missed this )) Yes, small hash rate increase is there ...
|
|
|
|
Hi, guys!
Who can explain what means compute_errors in SRB json reports?
One of my gpus (rx580_hynix8, Win10, 18.6.1) have a one error every 1 min. I reduced an overclocking and intensity, but the errors in srbminer still comes.
HWINFO no have memory errors.
Maybe somebody knows?
I think it is some kind of bug from version 1.8.4 All my polaris cards have 10-20 hw errors per day on heavy algo now according to srbminer. But they are perfectly stable and pool hash is good. All these cards had no hw errors before with previous versions of srbminer and other miner software with the same clock/voltage settings. |
|
|
|
just an update on my settings and results: have 3 vega's mining ETH on my rig (have a mix of rx470s, 480s and 570s, using phoenix 4.1c, amd driver 18.5.1 but will try 19.2.2 soon) all vega reference 56's flashed to 64. using wattman to adjust settings windoze 10 1803 gpu 1012mhz @ 0.825v mem 1085mhz @ 0.825v power limit -35% previously getting max of ~43.4mhs before winamdtweak changes now getting between ~49.95mhz and ~50.9mhz (depends on gpu) with : winamdtweak --i 6,8,9 --CL 19 --RAS 30 --RCDRD 12 --RCDWR 6 --RC 44 --RP 13 --RRDS 5 --RRDL 5 --RTP 4 --FAW 18 --CWL 6 --WTRS 4 --WTRL 9 --WR 15 --WRRD 1 --RDWR 18 --REF 17000 --RFC 248 0 errors over 3 gpus over 13+ hours. happy to leave everything alone now  Yes, good straps for eth! Thanx. 50.7mhs with vega64lc and 1090 mem clock For cn-r mine posted here are slightly better (2350 vs 2280 2300 yours) And I recommend latest (I use 19.4.1 but 19.4.3 should be good too) drivers. They are awesome! |
|
|
|
My Vega 56 cards (reference cards with Samsung mem) is still running V64 BIOS.
With all these changes with the mem tool etc is there any advantage of going back to stock V56 BIOS?
for low power - yes, stock v56bios + tight timings are good. (800mv and up to 2100h/s cn-r) But for max hashrate if you have low temps in your mining room and cheap electricity v64 bios is better (850-900mv and up to 2300h/s) |
|
|
|
|
2 kerney666
Can you please make txt files with correct new line symbol for windows version (in teamredminer-v0.4.5-win.zip file)?
It is real pain to read in std notepad )))
|
|
|
|
my timings are stable up to 1090mhz hbm clock. Yours feel fine with 1107 but I didn't test for long.
Mine are 2350h/s cn-r (teamred) vs yours 2200h/s (teamred) with 1090mhz clock. Claymore's is 49.5 vs 48 mhs eth.
With higher mem clock (1107mhz) with your timings results are getting better but didn't catch mine at 1090mhz.
Thanks for the stock values. That is a huge difference in CN/R and it shouldn't be, im wondering what settings did you get for 'my' timings. Can you share them ? <values cl="20" ras="31" rcdrd="12" rcdwr="5" rcab="45" rcpb="45" rpab="14" rppb="14" rrds="4" rrdl="6" rtp="3" faw="16" cwl="7" wtrs="4" wtrl="9" wr="16" rrefd="8" rdrddd="3" rdrdsd="2" rdrdsc="1" rdrdscl="3" wrwrdd="3" wrwrsd="2" wrwrsc="1" wrwrscl="3" wrrd="1" rdwr="17" ref="17300" mrd="8" mod="15" xs="270" xsmrs="8" pd="8" cksre="10" cksrx="10" rfcpb="160" stag="7" xp="8" cpded="2" cke="8" rddata="19" wrlat="5" rdlat="14" wrdata="1" ckestag="2" rfc="250" timing="0" zerorpm="False" tdplimit="0" maxtemp="70"> here what I see on vega64lc with srbminer 1.8.6 and profile 10 I used timing = 0 (auto) option from drivers with srbminer. When I applied custom timings shown above I used timing = 1 driver mode. Don't know if it makes any difference |
|
|
|
Hi, Dok!
Just examined your timings for profile 10.
They are good for hynix HBM2 but for samsung (both low voltage 800-940 mhz vega56 and high voltage 945-1100 mhz vega64 I have much better (more stable and fast) mem timings)
I can share to you if you want ))
Ok, so now i have examined your V64 samsung timings and i can't see why they would be much better, as you say. Can you run profile 10 on V64 with samsung mem? If you can, what's the hashrate with profile 10 vs your timing on the same algo? my timings are stable up to 1090mhz hbm clock. Yours feel fine with 1107 but I didn't test for long. Mine are 2350h/s cn-r (teamred) vs yours 2200h/s (teamred) with 1090mhz clock. Claymore's is 49.5 vs 48 mhs eth. With higher mem clock (1107mhz) with your timings results are getting better but didn't catch mine at 1090mhz. |
|
|
|
|
Show Posts
