Whoa, I leave for a few days and when I come back, you guys have a ton of information and all new things posted! Where to start from…
This link doesn’t work for me. It says video is private. I would love to see the video!
This is such a radical finding!! Well done bro! You made me go back and fiddle with my Vega again. Will make a new post about it. But, really nice find! Good job!
I would love to see the straps you use that allows you to run your Hynix vega at 961 mem, and give 53.32 MH/s. My best ever hashrate was 53.11 MH/s but with more than 2% invalids, so that was apparently not legit.
So, I recently had to move my Vega 56 to my AMD rig, in order to make space for my 2 new 3060 Ti’s. When I put it in, I noticed that -with exactly the same settings- the wattage dropped a bit more (from 99W to 97W), but also the hashrate dropped a bit (from 52.71 MH/s to 52.66 MH/s). Weird.
After comparing some old straps I was using (that gave better hashrate but lots of invalids) with the ones I use now (props to @Smining570 for his really nice Hynix straps), and having in mind the really nice finding of @ss-silver regarding RRDL, I found out that in my old straps I was using a lower RC number (that gives more hashrate), but with high RRDL as well.
So, naturally, I thought “let’s drop the RRDL value to drop the hashrate and improve invalids, but try lowering RC as well to get better hashrate”. Because it seems, that RRDL has a bigger impact in invalids and lower impact in hashrate, whereas RC has bigger impact in hashrate.
From a starting point of these straps:
amdmemtweak --CL 20 --RC 36 --RP 11 --WR 14 --CWL 8 --FAW 12 --RAS 20 --REF 65535 --RFC 248 --RTP 5 --RRDL 6 --RRDS 3 --WTRL 9 --WTRS 4 --RCDRD 12 --RCDWR 12
and changing --RC 35 --RRDL 4 , I went up from 52.66 to 52.71 MH/s:
As there were no invalids for a couple of hours, I tried lowering the RC value even further, to 34. Seems solid so far and will have to let it run for a few hours. If no invalids again, I will either drop RC even more to 33, or try going one step up in RRDL to 5.
@oni@Smining570@ss-silver
I need your expertise, I am about to buy a second-hand MSI Vega 64 Air Boost. I don’t have an opinion on these, I only know they have Samsung memory. Are they decent if they get BIOS-modded to Vega 56?
This is a stress test the guy sent me:
I am a bit concerned about the memory temperature running at 82°C (reaching 86 max). Is this normal under a “gaming session”? Or is it an indicator that I will need to re-paste / replace thermals?
Thanks in advance! (a fast answer will be much appreciated)
and for my gigabyte 64 gaming oc (flashed to 56) use straps for samsung memory, i changed many --REF in the straps (17000, 21000, 65000, or 65535) it crashs, i have to low down its mem to 840 or 850, and it doesn’t crash, but i find the same hashrate arround 45 mh/s
hi. I dont have that one, but have been reading about that so after bios mod its decent card with samsung, those are like ref vegas, those i have 3pc. For lower wattage/temps its better to run those around ±53mh.
83c in gaming indicates that it needs new pads and paste, i have done all my second hand blower vegas pads right away, then it stays even with max clock´s under 70C if u have good air flow for them.
Personally i like 56 better than 64, 64 are more hassle to tune and normally u will end up with higher consumption…i all depends how much u want to sqeeze out from gpu. Hope this helps with your decicion. btw. pads and paste is quick job with these ones
Thanks for the quick reply. I will have to change pads and paste then. Any good guides and/or recommendations on what products to use?
I haven’t done this job before, although I am technically skilled enough (I been changing batteries on my iPhones myself for example). So, with the right directions it should be no problem for me.
I have used Noctua paste and arctic pads. U need 1mm and 1.5mm pads…1.5mm is only little piece, if u scroll up, there should be one picture of ref vega opened, u see from that where u need that little thicker pad, i dont think that msi bnoard is way different but naturally do your own research to confirm. And pads and paste chage is defo easy job if u have opened some electronics before…there is no parts with springs etc that shoots away when u open the card If and when u get the card, run it first with max clocks so u get reference temps to what is the impact of pad/paste chance…dont over do with paste and pads rather a bit too big than exact. let us know how it goes
I have a airboost Hynix its about 10 extra watts but it runs cooler than all my reference cards. In my experience that benchmarking runs hotter than mining at max hash with fans set.
Thank you, I thought It was a good it was a good Idea for two small rigs. The distance between GPU is even too much, but I had the space and was not going to mount more than 7 GPU.
Your RIG is OP! Nice airflow there!
Here you can see my straps and the result, including time of testing:
I can confirm this worked nice with no invalids, and the miner was giving those results. But do not allways believe the sofware calculation… For exemple, a Vega 64 @55MHs doesn’t consume 90W, it’s more like 180.
After this straps I did Hardware change, and rig became unstable. At the end went to old config (only with a swap position between Gainwards), and the rig was still unstable, so I had to change a little bit some straps, and put higher voltages in some cards.
I do not understand at all
Depends on how warm it was in his room
With a good repaste you’ll have a fine gpu. I’ve posted a few good videos somewhere above which I found on YouTube. It’s not that difficult, and certainly worth it.
okay, so it can be that way around allso, good to know. Anyway, for second hand blower its likely that pads s change is a good idea. So those benchmarks stress gpu more than mining, interesting
Thanks guys, you really helped me make a decision, I love vegas so I am gonna buy it and see. Almost certainly it will need a repaste job, so I am doing a small research on this one atm.
Thanks again!
I ran some tests and couldn’t get my Vega to work with your straps at 961 mem. It seems that as soon as my SoC changes state and goes to 1028 MHz, my card becomes unstable. So, back to my stable 960 mem / 960 SoC again.
I have one vega 56 Hynix. All I did is starting with samsung straps, and moving step by step to hynix straps. Because the card does not support the hynix straps, at least not at 960-961MHz.
But if you see the AVG in the miner, I think that’s the correct value to check, the AVG provided by the miner, not by Hive. The problem is you need time to get a consistent value.
If we check the accepted shares, GPU 3 seems to have the best performance, even when avg is not the highest.
I believe the number of accepted shares, so GPU 3 must be the one with better hashrate… Even when Hive doesn’t think the same.
If I go down to 960MHz, Hive value goes to 53.01MHs.
PD: I replaced all Vega stock thermal pads by Thermaltake pads. Some of the old ones were solid, 0% elasticity. And also added some thermals extra. But I was surprised temperature just improved few degrees… I think those thermal pads were good quality. But I still think replacing them is a must! Just take care.
Greetings all! I am struggling with a Vega 64 I have.
Its a sapphire blower vega 64, its been converted to water cooling (so no fan speed / reading)
This card overclocks great using windows / nicehash / amdmem tweak I get 50mhs on srbminer in nicehash.
I’m trying to get it setup in HiveOS (I have another computer w vega 56 already running HIve) and I cannot get this card to overclock any in Hive.
With stock config it will run, but it seems any overclock or strap I try it just failed gpu detected dead.
I’ve tried Teamredminer and SRBminer.
could the water cooling setup be causing Hiveos to act weird?
Any suggestions on a very basic starting point on overclocks and how to work it up?
I’ve tried just about every iteration I’ve found.
I really would like to mine Eth and get paid in eth which is why I want to use hive.
Kind Regards!
If and when u get the card, run it first with max clocks so u get reference temps to what is the impact of pad/paste chance…dont over do with paste and pads rather a bit too big than exact. let us know how it goes 
