Getting modded: The Polaris (RX460-480/RX560-580) modding thread

[chawks2]

Sorry, is that the same hypothesis again? What is different and what should I grasp onto here that wasn't mentioned previously?

ED, sorry for not explaining it better. Look at all 3 screenshots. Notice the difference between screenshot #1/#2, they have timings "1:"(Samsung) vs "2:" (Hynix). The third screenshot shows "0:" (Micron). This is confirmed by the upper right drop down list. Curiously, you will notice screenshots 1/2 are showing Samsung, yet I had to apply the memory timings to "2:" to get the hash rates. If you pull down that menu option you will see 2 different types of memory listed, Samsung and Hynix. If I pull down on the 3rd screenshot I only see Micron.

There is a 4th screenshot I have not included which is the MSI Gaming RX580. This one would not get the 30 MH/s rate until I changed the "1:" (Samsung) timings. ethOS confirms the primary type of memory being used by their status page. (I can take a pic of this too if needed)

Hope I'm understanding what you are asking. As for the pictures of the ICs, if there is a location on the card you would like me to take a pic of, I will do my best. My miner setup is not particularly easy to work on (need Naviq lol).

 

[EarthDog]

Yes, again, I believe there are entries for different types of GDDR as there is different types of GDDR. I believe you will get different performance from selecting a different entry as there are different timings/subtimings on each. Just as if you change the timings on DRAM you can get different performance of the same exact sticks. But seeing a performance difference doesn't mean its mixed. It means whatever the hell is on there is running using that set of timings.

So again, just because there are entries in the BIOS does not mean they are mixed on the same PCB. I suppose they CAN be, but its not something I have ever seen in my decades doing this stuff. Now, that doesn't mean it isn't possible or didn't happen, but the evidence presented thus far isn't conclusive (from either side). Correlation is not causation and all that.

As far as a picture, quite obviously we would need to see the PCB with the memory ICs on it. This means powering off and removing the heatsink to snap the picture. That would shut me up in a heartbeat seeing different ICs on there. As it stands, nothing that has been presented so far is conclusive.

 

[chawks2]

As far as a picture, quite obviously we would need to see the PCB with the memory ICs on it. This means powering off and removing the heatsink to snap the picture. That would shut me up in a heartbeat seeing different ICs on there. As it stands, nothing that has been presented so far is conclusive.

I will do my best, and the easiest one for me to pull the heatsink off would be the MSI Gaming, I can do this for you, but Saturday is the earliest.

Google can help us too.

 

[EarthDog]

We would need to see a board that has the multple entries in its BIOS. A random google image won't do it. Consider gsx has two of the same cards, one that doesn't have 0/1/2 and one that does. So the only way is to show the PCB of the card that has multple BIOS entries where you believe its for mixing and matching. If your MSI does, then you are good. If not, we need to see something like the Nitro+.

 

[chawks2]

We would need to see a board that has the multple entries in its BIOS. A random google image won't do it. Consider gsx has two of the same cards, one that doesn't have 0/1/2 and one that does. So the only way is to show the PCB of the card that has multple BIOS entries where you believe its for mixing and matching. If your MSI does, then you are good. If not, we need to see something like the Nitro+.

I'm really lost. The MSI Gaming RX580 is not going to be different unless a new iteration card is released. Arent these practically new cards? Why would a "random" image not prove your point?

I am beyond confused what you are looking for. The software tells us there are 2 types of RAM on my MSI, XFX, and Sapphire Nitro+, it is very evident that is the case via the Polaris BIOS software. The PowerColor clearly is represented by a single type of memory as stated.

Are you just trying to prove a point to prove it? Do you believe the software is giving you incorrect information? Irregardless I will still remove the stock cooler on the MSI, but I am not clear on what you are trying to prove. Sorry.

 

[chawks2]

Based on this review of the Sapphire, which states correctly it uses Hynix, I still do see Samsung on my drop down. Maybe you are right ED, maybe the software is broken!

EDIT - This would also explain why the "2:" changes to the timings did nothing. There is no Hynix on the MSI Gaming, thus editing the Samsung "1:" was correctly interpreted, just turns our the software creator is not interpreting the information properly from the card.

K now I understand.

 

[EarthDog]

I'm really lost. The MSI Gaming RX580 is not going to be different unless a new iteration card is released. Arent these practically new cards? Why would a "random" image not prove your point?

I am beyond confused what you are looking for. The software tells us there are 2 types of RAM on my MSI, XFX, and Sapphire Nitro+, it is very evident that is the case via the Polaris BIOS software. The PowerColor clearly is represented by a single type of memory as stated.

Are you just trying to prove a point to prove it? Do you believe the software is giving you incorrect information? Irregardless I will still remove the stock cooler on the MSI, but I am not clear on what you are trying to prove. Sorry.

I already gave you a reason why.... I said that another member, gsxr said he had two cards of the exact same type. One had all 0's, the other had multiple entries. Prior to that, he mentioned that his video card had multiple different memory ICs on because of what he is seeing in the bios (those multiple entries we see).

I am not saying the software is wrong or broken. I am saying just because there are multiple entries in the bios for different brand ram, doesn't mean the card is mixing and matching ICs on the same PCB. In other words, what was brought up as proof, I am saying isn't actually proof (I mentioned this earlier). The software is likely reading it correctly. Again, as I said earlier, the same card can have different memory ICs on it so the BIOS includes all of these so they don't have to flash different BIOS' to cards with different memory on them.

The only way to check is to see the PCB of a card with multiple entries. If it doesn't then that is a useless image for this excersize. It really is that simple...it has to be a card which shows multiple entries to prove/disprove there are different memory ICs on the same PCB.

Hopefully that helps as I really can't break it down any more than I have.


EDIT: This is what I am trying to get to the bottom of............:

i will also reiterate the need to not simply use GPUz to check what kind of memory you have. (I know you didnt say that, but thats what most people do). GPUz will only tell you the first memory manufacturer. I have one card (RX580) that GPUz reports as Hynix, but inspection of the stock BIOS with PBE reveals that it actually has a mix of Hynix/Elpida/Samsung, each with their own sections for memory timings.

He is saying there are 2/3 different memory types on the card. I am saying I highly doubt that is true and the evidence 'proving' it isn't actually proof.

 

[chawks2]

EDIT: This is what I am trying to get to the bottom of............: He is saying there are 2/3 different memory types on the card. I am saying I highly doubt that is true and the evidence 'proving' it isn't actually proof.

Sorry ED, I will post my pics sometime on Saturday.

 

[mackerel]

EDIT: This is what I am trying to get to the bottom of............: He is saying there are 2/3 different memory types on the card. I am saying I highly doubt that is true and the evidence 'proving' it isn't actually proof.

To think through it logically, let's say AIB maker has the parts for a batch of 1000 cards, but of that, they have enough memory A for 750 cards, and memory B for 250 cards. Are they more likely to make 1000 identical cards, each card with 75% memory A and 25% memory B on it, or would they make 750 cards with memory A, and 250 cards with memory B. They could make one bios to detect and set the appropriate timings for memory A and B. I just can't see them mixing different types of ram chips on a single board. Unless the ram was identical in spec, it would be more pain to mix and match on a single board than it is to make separate batches using the same type on each board.

 

[EarthDog]

Sorry ED, I will post my pics sometime on Saturday.

Thank you. I appreciate you taking the time out to do so.

To think through it logically, let's say AIB maker has the parts for a batch of 1000 cards, but of that, they have enough memory A for 750 cards, and memory B for 250 cards. Are they more likely to make 1000 identical cards, each card with 75% memory A and 25% memory B on it, or would they make 750 cards with memory A, and 250 cards with memory B. They could make one bios to detect and set the appropriate timings for memory A and B. I just can't see them mixing different types of ram chips on a single board. Unless the ram was identical in spec, it would be more pain to mix and match on a single board than it is to make separate batches using the same type on each board.

Which is exactly my point (made previously).

 

[chawks2]

 

[EarthDog]

Samsung all around.

...and a random VRM pic.


Edit: it proves, at least on your card, what i was saying - the entries in the bios doesnt mean there are mixed ICs on the gpu. Now, his may be, though after seeing this, im willing to bet the farm that isnt happening.

Thanks for taking the time to check.