Topic: AMD Mem Tweak XL - Read/modify timings/pp/straps on the fly - page 52. (Read 58892 times)

Hi Guys,

Zorg33 said to me share this set, because it's impossible, but working good for me:
--CL 12 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 10 --RRDS 3 --RRDL 3 --RTP 5 --FAW 20 --CWL 7 --WTRS 4 --WTRL 9 --WR 16 --WRRD 1 --RDWR 19 --REF 17550 --MRD 8 --MOD 15 --PD 8 --CKSRE 10 --CKSRX 10 --RFC 250

Vega56 @ 64 sapphire samsung ram, Win10, 18.6.1 and 18.5.1 classic 1408/900core 1100/900mem

Turtle 22+
ETH 51+
Cr-Lite 10K khs (1600/990 1100/900, but crashed, 9600kh 1408/900 1100/900 super stable after 24h)

Ethash is even better than cn-trtl wrt being able to drop core clock on vega.  Where you may lose 5+% h/r on trtl by going to cclock p0 (852mhz), I saw zero h/r loss on ethash when doing the same, at least w/ stock timings.  Should allow you to get your voltage down significantly - closer to 800mv.

Interesting. I guess my point still stands then. Unusable currently on Polaris.

Most of my settings are just using values which seem to be gathering consensus here - from people such as yourself  I honestly didn't put much time into tinkering, other than adjusting a couple 'derived' values based on what seemed more appropriate mathematically.  I'd like to look into more, esp CL/FAW/WR...

I mainly wanted to point out the outsized effect of increasing REF, and put some thoughts to what it does, and why it's ok to go from 3.9K to say 15-17K.

I have noticed that if you throw out --cl value it doesn't crash when applying string. So it must be some bug in applying --CL. Also the timings are reset to default when i run miner. If i apply them while mining - Bluescreen.

I tested the base settings w/o modded timings on win10pro w/ 18.x drivers in the past, and saw the same results.  There may be differences w/ 19.x - I haven't tried yet.

That being said, a linux setup is pretty simple and equally performant for many algos these days.

Every combo I've tried on Polaris (even translating modded bios strap values) results in a crash to blue screen for me. At this stage, I'd say this tool is unusable for Polaris cards.

Anyone analyzed the built-in HBM tuning in the driver?

I just started looking at it and seems it modifies RRD values and probably some others too.
Might get some good ideas from there....

What bios are you using . I can get 20.1khs per card but have to run 1250 core 1100mem using 145w


Edit

Managed to get it to 21.8 khs @ 125w in hwinfo pulling 1060w from wall with 6 vegas approx 130khs trt

Looks good so far. About 1 kh/s boost to turtle. Didn't work with all my hynix tho. Need to figure out what to change to get to work with weaker card.

Edit: after longer time most gpu crash or have hashrate drop

I'm trying to take it slowly now.

So far I've been doing really well with these timings on CN-Turtle, i.e. improved hashrate on miner AND pool reflects it exactly (124.5 Kh/s miner hash, 124.5 Kh/s pool hash, according to TRM):

--CL 16 --RAS 30 --RCDRD 14 --RP 12 --RC 44

It's not the highest hashrate I've had using other timings posted here earlier, but the hash increase is 100% consistent on the pool, as opposed to what I experienced with too aggressive timings previously (I saw up to 21.3 Kh/s per Vega but the effective pool hash was A LOT lower).

I'll keep tweaking, but if anyone needs a basis to get started with, this works very well on my ref Samsung 56@64 rig.

Results:
Cards : Vega 64 (aircooled, reference, samsung)
Platform : Windows 10, 18.6.1
Miner : PheonixMiner 4.2c
Algo : Ethash
Clocks : 1216/1107/875 (ODT)
Actual : 1187/1107/850 (HWinfo)

Stock                                                                                                                                                           : 44.75 mh/s
Current Timing  --rp 12 --rc 44 --rfc 250 --rrds 3 --rrdl 3 --rcdrd 12 --rcdwr 5 --CL 19 --RAS 28 --REF 15600       : 50.12 (12% inc over stock)

Finally broke the Ethash 50mh/s barrier in windows 

pbfarmer is our resident magician, we strive to reach his god tier levels of efficiency  
Was waiting for his take on the timing to replicate  


Ill test ur CL, RAS and REF timing. Thx pbfarmer.

Results:
Cards : Vega 64 (aircooled, reference, samsung)
Miner : SRBMiner 1.8.3
Algo : Cryptonight litev7
Clocks : 1408/1107/875 (ODT)
Actual : 1354/1107/850 (HWinfo)

Stock                                                                                                                                                           : 4530
Todxx               --rp 10 --rc 44 --rfc 300 --rrd 3 --rcdrd 12 --rcdwr 5                                                                : 4800
My current        --rp 12 --rc 44 --rfc 250 --rrds 3 --rrdl 3 --rcdrd 12 --rcdwr 5                                                   : 4865 (7.4% inc over stock)
pbfarmer          --rp 12 --rc 44 --rfc 250 --rrds 3 --rrdl 3 --rcdrd 12 --rcdwr 5 --CL 19 --RAS 28 --REF 15600       : 4980 (9.9% inc over stock)

In all seriousness, was w8ing for ur numbers the most and u didnt disappoint  

Rx Vega 56 hynix memory  
1250/875mv   916/925mv
tartl cn 20.12 - 20.20  kh/s  L24+24
CN/r   2160   1450/900
60 min no errors.
WinAMDTweak.exe --rcdrd 19 --rcdwr 4 --rc 35 --rp 13 --rrds 4 --rrdl 5 --rfc 148 --REF 15600
Cool!!!

That is some crazy efficiency. I guess that is not even possible on windows, right? Maybe it is time for me to stop being scared and tried alternative operating systems.

First - quick results for cn-trtl, w/ efficiency-focused settings, using timings mostly borrowing from others here, w/ some minor tweaks:

Vega 64 air, ubuntu 18.04 + amdgpu-pro 18.50, TRM 0.4.3 (L18+18), 852 cclock (p0)/1107 mclock/818mv, power readings at the wall

stock timings:
--CL 20 --RAS 33 --RCDRD 16 --RCDWR 10 --RC 47 --RP 14 --RRDS 4 --RRDL 6 --RFC 260 (--REF 3900)
18.5 kh/s @ 135w (137 h/w)

modded timings 1
--CL 19 --RAS 28 --RCDRD 12 --RCDWR 5 --RC 44 --RP 12 --RRDS 3 --RRDL 3 --RFC 248
19.75 kh/s @ 137w (144 h/w)

modded timings 2
same as above, plus --REF 15600
20.71 kh/s @ 137w (151 h/w)


Second - notes on power...  I don't see any appreciable power differences - nor would I expect to.  Clocks and voltages are untouched, we simply have a bit more data being transferred.  Even the 2w difference I'm reporting here is conservative - taking natural fluctuations in my readings into account, my actual increase could be closer to <= 1w.  People seeing large power increases (at least on vega 64) seem to have something else going on.


Last - some conjecture / educated guessing re: THAT --REF THO!!!  I'm assuming --REF is the refresh frequency, in nanoseconds, and unlike most timings, a higher value (meaning less refreshing) is better.  Refreshes steal bandwidth, and AMD seems to have gone majorly conservative (aggressive?) on this, probably due to the super high temps of the HBM during normal/gaming use.  As leakage increases w/ temps, more refreshes would be required when running your GPU/HBM at high clocks/voltages.  Since (efficient) miners tend to run cooler, the crazy high default refresh rate is really unnecessary.  I found 4x the default to be around where returns quickly diminish, at least at my clocks - i can get maybe another 50 h/s (turtle) going 4.5x.  HOWEVER - if you run super aggressive for max h/r, or just aren't effectively cooled in general, you may want to dial this back, or you may start seeing mem errors / bad shares from corrupted data due to insufficient refreshing / leakage.

ETH and CNR ?

Overdriventool : reset and apply

noice!

 

Still running. There is no problem even on the lower quality silicon GPU's (yet).  Ref Vega 56@64 bios, samsung HBM.