Perhaps one was produced later than the other and by that time they added another brand IC due to the shortage? I don't know. Clearly its a different BIOS so different times when it was loaded onto the card. But I find that more logical than mixing and matching ICs... that just sounds like a nightmare for stability.
Take some pics of the cards in question showing all the ICs and see if they are different. That is the only way to confirm 100%.
Of course not the same version number. They have different information in them. I actually temporarily bricked the 3mem card, because I hastily flashed the bios from the 2 mem card to it (I foolishly assumed cards were identical because they had the same sku and both listed in GPUz as hynix). Fixed it by flashing the stock Rom back, and manually editing the card from there using my procedure I posted before. 3mem card also performs slightly better.
I’m not going to tear a card down, void the warranty, and interrupt the mining.
I understand... but that is a shame as its the only way to check, really. As I said, I fully believe there are different entries there for different ICs, but only one IC is likely to be used at one time.
EDIT: What vendor make/model is the card? Why don't I simply reach out and ask...
my overall point was that it's too dangerous to copy/paste raw hex values into a bios editor that comes from an unknown source and you have no idea what those values are.
one typo and boom, bricked card.
using the 1-click method built into PBE is fool proof. and if you're advanced enough to be able to create your own hex values for custom timings, you dont need to be copy pasting values from a forum post.
Got it. Thanks for giving it to me...glad you took the time to write that and not just answer it.
I reached out to MSI to clarify if its simply a bios entry/placeholder or if they are using mixed ICs. I will report back with the repsonse when I have it.
EDIT: I also reached out to W1zzard and asked him how GPUz reads the memory ICs on the board and if it would catch both if they were on there.
OK, so if you find you have settings in your VRAM BIOS with "1:1750" vs "2:1750" it means you have mixed VRAM. If you have "0:1750" it means you DO NOT have mixed RAM (easiest to deal with - they are both easy, but easier lol). Now for poops and giggles I tried the timings on the "1:1750,1:2000" vs (Hynix) "2:1750,2:2000,2:2250" - apparently there is something here with which RAM comes first, denoted by the "1:" vs "2:".
I now have the Samsung memory working on the timings I posted earlier and am now able to get 30Mh/s I now have 5 total cards which are all using the Sapphire Nitro timings for ALL cards with core clock @1135 and memory clock @2150. This is resulting in ALL 5 cards with a 30 MH/s rate on Claymore!!!!! More to come lol. sorry
have you checked for GPU hardware errors? reported hashrate means diddly if there are so many errors that your effective hashrate at the pool is subpar or it causes increased instability from all the errors.
download HWinfo and check the GPU errors listed under each GPU in the list.
i toyed around with 2100+ memspeeds briefly but backed them down to 2000 because i was seeing thousands of mem errors despite great reported hashrates.
have you checked for GPU hardware errors? reported hashrate means diddly if there are so many errors that your effective hashrate at the pool is subpar or it causes increased instability from all the errors.
I have 0 errors, and have 0 errors from claymore or any errors with mining etc. I have had 3 cards working this way for at least 7 days. The newest MSI Gaming will be 24 hours around 4PM tomorrow. Will let anyone know if I see smoke or cards blow up or they melt.
Never noticed the GPU errors line before! I got 3 identical as far as I can tell RX 580 8GB Samsung ram, bought at same time. I had overclocked ram on all to 2200 mining ethereum.
First one I looked at had 240k errors after about 2 days uptime (120k/day). I've backed that off to 2150 for now and will see how it goes. This is the same system with highly OC'd system ram and I have prime stability problems in another thread.
Second one has 2.6k errors after 23 days uptime (0.1k/day). Totally different! Only 1 rejected share out of 13409.
Third one has 36 days uptime. 1 rejected share out of 20462. 57k errors (approx 1.6k/day).
I suppose the question then is, how many errors is too many? There's a difference of a thousand times error rate between my best and worse. Maybe I can try putting a bit more mem voltage on, but that's something to look at another day.
Sorry I mistyped. I meant GPU memory errors. It’s something apparently only reported by the hwinfo program (I haven’t seen any other applications list it) claymore won’t report that.
Only available on AMD cards also if anyone wanted to check that. Doesn’t get reported for nvidia.
But if you checked with HWinfo and your pool hashrate lines up with reported then great.
Never noticed the GPU errors line before! I got 3 identical as far as I can tell RX 580 8GB Samsung ram, bought at same time. I had overclocked ram on all to 2200 mining ethereum.
First one I looked at had 240k errors after about 2 days uptime (120k/day). I've backed that off to 2150 for now and will see how it goes. This is the same system with highly OC'd system ram and I have prime stability problems in another thread.
Second one has 2.6k errors after 23 days uptime (0.1k/day). Totally different! Only 1 rejected share out of 13409.
Third one has 36 days uptime. 1 rejected share out of 20462. 57k errors (approx 1.6k/day).
I suppose the question then is, how many errors is too many? There's a difference of a thousand times error rate between my best and worse. Maybe I can try putting a bit more mem voltage on, but that's something to look at another day.
Yeah I think a small number of errors is ok. What defines small? No clue.
Before I discovered the GPU mem errors thing, my system was on the edge of stable. Would randomly crash. I checked it finally after a solid week of uptime and some cards had like millions of mem errors built up. Adjusted bios settings and clocks and now I get almost no errors. Maybe like less than 10 a day on a couple cards and 0 on others
though I was a bit reluctant to do it as bios modding is as you know not the safest thing, this thread prompted me to do it as I see a growing need of it, a guide to modding polaris bios and a thread for discussions on that, can be found here
if you guys dont mind, i will move the last two pages of discussions from here to that thread as we are slightly away from the OP by now, but more importantly its buried so deep .. new users wont see it and its a shame as its good info
though I was a bit reluctant to do it as bios modding is as you know not the safest thing, this thread prompted me to do it as I see a growing need of it, a guide to modding polaris bios and a thread for discussions on that, can be found here
if you guys dont mind, i will move the last two pages of discussions from here to that thread as we are slightly away from the OP by now, but more importantly its buried so deep .. new users wont see it and its a shame as its good info
I think this was an excellent idea to break this out ZL1 into it's own thread. Just got my EVGA SC17 Gaming laptop and will be pushing this 1070 to the limit. I will post my guides/screenshots soon about the RX580 8Gr's.
OK, so if you find you have settings in your VRAM BIOS with "1:1750" vs "2:1750" it means you have mixed VRAM. If you have "0:1750" it means you DO NOT have mixed RAM (easiest to deal with - they are both easy, but easier lol). Now for poops and giggles I tried the timings on the "1:1750,1:2000" vs (Hynix) "2:1750,2:2000,2:2250" - apparently there is something here with which RAM comes first, denoted by the "1:" vs "2:".
Just saw this... where does it reference those are for mixed memory environments? Is it possible those are there for different memory types used, but not mixed?
PS - MSI marketing is asking HQ and will get back to me.
PPS - GPUz is not able to read different types of GDDR on the same card due to the registers not being able to give that information out.
Just saw this... where does it reference those are for mixed memory environments? Is it possible those are there for different memory types used, but not mixed?
PS - MSI marketing is asking HQ and will get back to me.
PPS - GPUz is not able to read different types of GDDR on the same card due to the registers not being able to give that information out.
It seems logical to me. Again, I am only going off experience with changing timings and trying to comprehend what the values mean.
Cards which have different memory in the polaris editor have 2 memory values (represented as "1:" vs "2:", in my experience either Samsung or Hynix. ethOS represents ONLY the first memory footprint for some unknown reason. The cards which are represented as primary memory through ethOS (Hynix require a "2:" modification), cards which represent Samsung as the primary memory in ethOS required the "1:" to be modified to get the 30 MH/s hash rates desired. I also have 2 cards which are labeled as "0:" and I take this as meaning there is only one type of memory on the card.
Again I have no idea what I'm talking about, it just seems logical to me. Additionally, if all these cards have the same types of memory and the timings are the only thing separating them in regard to high hash rates, why not apply those timings if they are the same memory to begin with? I have had no errors whatsoever and still do not understand why different vendors would have different timings if the memory is the same? AND the fact the ONLY differences in timings are the 1750,2000,2250 values is perplexing. Could it only be due to heat? (this can be changed too and I have made the power changes dropping temps in some cases over 15C!!!)
Seems more logical to me that they are profiles for each type of ram that is used instead of going through the potential nightmare of trying to run mismatched ram. Each gddr type ram has its own timings so it makes sense we see different profiles for each brand on the same bios for compatibilty and stability reasons.
I dont suppose you are willing to take a picture of the card you think is using mismatched ram, are you? Otherwise, ill report back when i hear from msi.
I dont suppose you are willing to take a picture of the card you think is using mismatched ram, are you? Otherwise, ill report back when i hear from msi.
It wasnt a "guess" so much as a hypothesis, lol. I will post my findings shortly, I have some time today and have been bitching. I will get this done and explain why I strongly believe this is absolutely fine to do, AS LONG AS you are not pushing the mem clock beyond the original specs too far. 2150 for me has worked for 4 different cards very well from a stability stand point.
But there isn't any way to prove it through explanation, is there? Not sure what more you have than what was said, but... I don't think there is anything we can do outside of snapping a pic of the actual memory ICs on the card and reaching out to MSI (which we are waiting for a response).
The fact that they seem to be working lends credence to the point that Memory brand A can run with Memory B timings, but that doesn't mean there are two there either.
I'm all ears and hope more facts are brought to the table, but until I see a picture or MSI says so, I'm squarely on the side of it not being mixed.
EDIT: And I am guessing you are not willing to snap a pic either (why do I have to ask that question multiple times to get a response out of people???
Alright fellas, below we have 3 screenshots. These are ALL stock BIOS settings. Curiously, the Sapphire Nitro has the best hash rate and was confused as to why, additionally, this particular card has temperatures WAYYYYY below all the other cards.
Let's review.
XFX SuperClocked - RX 580 8G
This may seem to be out of order, but the reason why I mark up the XFX was because this was my first testbed card which I just applied so many different types of settings you have no idea. I was super frustrated with the inability to get above 27 MH/s (following cybergator's guide or whatever his name was by staying within your memory and just changing the 1750 strap and applying it to 2000,2250 etc bleh).
Now, the blue and purple arrows below denote the differences between the Sapphire Nitro+ (out of the box 29.5 MH/s in Claymore - btw all these hash rates are in Claymore ONLY [not sure why]). The blue arrows are the memory timings and GPU changes made. My thought process here again is these are the same cards with the same types of memory, why not try and see what happens when duplicating these settings across all my cards. The purple arrows denote how the Sapphire handles the power settings and ultimately believe this is how the card manages the temperatures (postulation). Amazingly, by changing these values the temps on the card are reduced by 10C / avg!!!
Sapphire Nitro+ RX 580 8G
Out of the box this card gives you 29.5 MH/s and was perplexed why. It has Hynix memory as the primary (postulation based on information coming from ethOS) - in my experience the Hynix memory is the primary memory for the XFX, ASUS and Sapphire Nitro+. However, it is labeled in VRAM timings settings as "2:" (this is a postulation to be clear, I do not know for sure any of the comments are true lol). Changing the "2:" timings to the Sapphire Nitro+ settings for what I believe to be the Hynix memory allows for the increased hash rates from 27 MH/s to 30 mH/s!!!! YUUUGGEEE (in an annoying Trump voice).
PowerColor RedDevil RX 580 8G
Out of the box this also gives you 29.5 MH/s hash rate with stock timings. This one however is different, you will notice "0:" which allows me to postulate this card has singular memory and NOT mixed memory. Additionally, this card leverages "Micron" memory (I have yet to test this variant and will be taking down my miner on Saturday - after my payout of ETH this week lol )
I will also post my "guess" for this card and see where I can go with it based on my experience with Samsung/Hynix etc. Hope you guys enjoyed this reading. I have to walk away from this for a bit to play around with a new toy lol.
Sorry, is that the same hypothesis again? What is different and what should I grasp onto here that wasn't mentioned previously?
As I said, you think its there because its mixed and matched on the same PCB and we believe its just there as you can get the card with 2/3 different types of memory on it. We have seen this before quite a bit where the same card can use Samsung or Hynix or Micron. I am not saying it isn't possible, but what has been posted so far isn't conclusive. I can see the logic from both sides, but just think it is playing with fire to mix and match ICs on the same PCB which all have different timings...
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
