Uhh… yeah, don’t try and flash a vBIOS intended for different vRAM to what you have, it won’t go well. Just don’t use a vBIOS other than the one that came on your card unless you are in a pinch, eg; you bought a card with a bricked vBIOS.
Hello
I have a MSI ARMOR RX 570 4GB HYNIX gpu
and I did a vbios with a sapphire and I got very good consumption (in addition to improving the timings)
Even so, I could not exceed 2100mhz in memory with stability
What I can do?
HYNIX.rar (107 KB)
Read the guide, its all in there 
@ket I seem to get clock throttling issues due to reaching a certain temperature point. Is there a way to adjust this value? I have adjusted temperature target but still experiencing clock throttling. This is the BIOS I am using: https://download.gigabyte.com/FileL…GD…b2472a425d9a010
EDIT: After monitoring using WattTool, it appears to be when VRM temp reaches past 78 degrees Celsius, clock throttling kicks in, is there a way to adjust VRM clock throttle temps in the BIOS?
I am using the RX 580 Aorus BIOS.
@rootuser123 , I’d need to know what VRMs the card is using to know exactly what the problem is but my guess would be the card is either trying to draw too much power or it has crap VRMs that are throttling to stop them from blowing up. That link is also dead.
@ket I have a RX 470 G1 Gaming which I have modded and replaced the thermal pads and changed the cooler to the RX 580 Aorus one. I also modded the RX 580 Aorus BIOS with the VOI table from the RX 470. Unfortunately, I don’t know which VRM chips are used on the PCB. BIOS Link: https://download.gigabyte.com/FileList/B…b2472a425d9a010
@ket Looks like I managed to locate what low side and high side chips are used. Same mosfets as the RX 480 G1 Gaming/RX 580 Aorus series
VCORE (6AON6414A) high side current: 180A@25-100°C, 132A@125°C.
VCORE (12AON6508) low side current: 385A@25°C, 300A@100°C.
Memory (1AON6414A) high side current: 30A@25-100°C, 22A@125°C.
Memory (2AON6508) low side current: 64A@25°C, 50A@100°C.
Auxiliary (1AON6414A) high side current: 30A@25-100°C, 22A@125°C.
Auxiliary (1AON6508) low side current: 32A@25°C, 25A@100°C.
@rootuser123 , from looking at a glance the issue is probably a heat related one combined with wattage the GB models had completely crap coolers not any better than the reference AMD blowers. In the RX4x0 days GB cards were second on the “stay away from” list, only bested by Powercolor and their RX480 where the VRMs were massively under specced, and top of the “stay away from” list in the RX5x0 days as Powercolor had fixed their VRM blunder by that point. Gigabyte = all show and no balls, basically.
Looking at some images for the card wiith the cooler off (first time I’ve bothered as I already knew the card is crap from a design sense) you might want to take it apart again, the plate used for the BGA memory doesn’t look like it uses any thermal epoxy meaning it’s bare metal screwed to bare metal, air is a very poor thermal transfer interface literally any thermal paste placed on the back of that plate will significantly improve thermal transfer to the primary heatsink and should significantly reduce the temperature on the memory (perhaps 7-10c+) so it doesn’t cook itself. From what I see the VRM design on the card is completely horrible as well, doubling the FETs is bad enough in most implementations but GB are tripling them meaning there is a stupid amount of load balancing the phase doublers are trying to do… on a card that doesn’t look like it has more than 6 inductors (they are covered with a thermal pad in the images I’m looking at) and worse still it’s like GB knew the power delivery design used is bad because the inductors have a thermal pad front and back to try and keep them cool. I’m not surprised with the FETs being tripled and phase doubled, its literally the cheapest possible VRM design you can do something you’d expect to see on a (very) budget model motherboard and even then it would be considered iffy never mind doing it on a GPU. Looking at the vBIOS the card isn’t using the IR3567B or even the NCP81022, so god knows what trashy voltage regulator GB have chucked on there.
The good news is that a lot of the bad design can be circumvented with a properly tuned vBIOS. If you want me to do it I can do so for a modest donation which will go toward more guides like this and hardware reviews so people know whats worth their cash, and what isn’t. The only thing I’ll need to know from you is the ASIC quality as I need to know that to optimise clock frequencies to voltage, power draw, and heat output. AMD always did chuck way too much voltage at Polaris cards, even the algorithm just chucks the maximum 1.2v at the card constantly.
@ket It uses a IR3567B PWM controller and yeah I replaced all thermal pads, repasted with Noctua NT-H1 and managed to obtain a RX 580 Aorus cooler to replace the stock G1 Gaming which was a crap cooler. I actually modded my own BIOS already but just looking to find which part of the ATOM BIOS structure dictates VRM temp throttle point. ASIC quality is 87.4% but card is poor clocker, best I could do was 1350MHz at 1.175v.
@rootuser123 if GB have put an IR3567B on the card then they have got completely the wrong code in place for that voltage regulator in the vBIOS there is no way I see to use voltage offsets, something that is easy to do with the 3567B, and the code is totally different to the NCP81022.
EDIT: with an 87% ASIC 1.175v is way too much, leakage is going to be minimal with that chip so you can probably get something like 1.32GHz on 1.1v or less. Increasing TDP and Power limit should get around the throttling but some extended tuning might be necessary I really wouldn’t recommend increasing either of those without first seeing what something like 1.075v - 1.1v nets you on the core I’m extremely dubious about that VRM.
@ket Here’s the modded BIOS I made, so I have added additional registers in the Voltage Object Info (VMAX register implemented, ported RX 470 G1 Gaming VOI over to the 580 Aorus BIOS), modded TDC to 410A and TDP to 450W, memory timing mods done and VDDCI adjusted to 1V.
RX580A4.zip (108 KB)
@rootuser123 good god man reduce that TDC! You are telling the firmware that the FETs are specced way, way, WAY above what they actually are that card will, in all probability, blow up in a glorious hail of smoke, sparks, fire and burning smell if any software ever tries to really stress that card… Amazon’s New World would likely equal instant death for that RX470. I can’t find a spec sheet for Hynix H5GC4H24AJ but those timings look weird even for custom ones, some values are a bit tight which is probably making them a bit unstable under prolonged use, fan and powertune stuff isn’t configured right either and VDDC looks like its set at 1.1v which is rather high even for a 400 series card.
@ket Yep, that’s why I haven’t tested Furmark lol. I was experimenting ways to stop the clock throttling in gaming, but looks like that didn’t stop the clock throttling from DPM 7 other DPM states when VRM temp reached 78 degrees which I monitored in WattTool. In terms of the mod that I did, I have adjusted ATOM table length with additional registers in VOI, I did change the IMC voltage to 1.1v as per guide from hellm @overclock.net, I also attempted to expose loop 2 on IR controller as Vento041 from OCN forums said it helps to reduce EDC errors but looks like Loop 2 gives me no output when monitoring with MSI Afterburner.
@rootuser123 Much like the RX5x0 series I found increasing VDDCI on the RX4x0 card I tested didn’t do anything, EDCs can come and go with a simple reboot I’ve not found any solid evidence that increasing VDDCI helps with memory OCs and EDCs, Buildzoid also came to the same conclusion. The best way to get rid of EDCs is to make sure the memory is cooled as effectively as possible, identify the maximum stable frequency, then start tuning memory timings. A good set of memory timings for Hynix/Micron memory won’t just be more stable with better performance you’ll likely improve OC potential as well, best card I had that used Micron improved it’s memory OC by about 100MHz. VDDCI in your modded vBIOS is pretty messed up there’s no 0.95v entry but two 1v entries, your throttling issues still look like they are temp related to me so I’d start by increasing the acoustic limit and locking GPU voltage to 1.075v and seeing how that impacts throttling.
EDIT: Why are you using the RX580 vBIOS on your RX470? You can’t wholesale just plonk a 500 series vBIOS on an RX400 series card you will get odd issues, I’ve told people this countless times. 
All you have to do is spend 10 seconds with a hex editor to make the card think it’s an RX580 to get around the driver level lock. Your card physically can’t have more than 225w delivered to it either, 1x 8pin PCIE + PCIE slot = 225w.
@ket Hellm’s guide had the technique https://www.overclock.net/threads/bios-m…-rx580.1634872/ to port RX 580 BIOS on RX 4x0 cards. I ported the whole VOI from the RX 470 BIOS over to the RX 580 so it wouldn’t mess up with the I2C settings. I’m having no crashes of any sort when doing 3D Mark stress tests and Unigine stress test.
@rootuser123 there is more to it than just porting over the VOI, RX5x0 series wasn’t just a die shrink even though it was the most notable change other tweaks were made to silicon and vBIOS alike so while an RX5x0 series vBIOS will “work” it will not work as fully intended on a 4x0 series card even doing what you have done there are still physical differences in the card designs that would cause unusual throttling behaviour such as different FETs and/or overall VRM design. I give an example warning about this twice in the first quarter or so of the guide but to reiterate;
Now is also the time to mention that the Powercolor RX480 Red Dragon used substantially inferior VRMs compared to any other 480 and thus had a very low TDC value, do NOT raise this unless you want to risk blowing out the VRMs.
The bold part is what you need to pay particular attention to, much like the RD480 the GB470 has an awful VRM, because of the terrible VRM it causes the card, and any other card designed with a similarly terrible VRM, to easily throttle. You cannot magically work around this physical limitation in software, merely work with it to reduce the severity of the throttling which is why I’ve told you to reduce vcore to 1.075v and see what frequency the core is stable up to, with an 87% ASIC you do not need huge amounts of volts because again as I said, leakage from a chip with an ASIC as high as that is going to be minimal the silicon simply doesn’t need the volts you will never achieve the same kind of OC people get with lower ASIC value cards because that is the flipside of having a card with high energy efficiency potential; they just don’t clock as high because the chips have low leakage.
On face value I have an old 8GB RX470 in my backup system that’s worse than yours because it has a high ASIC of 92% and Samsung memory, meaning it will never be able to do a huge memory OC or core OC but it still performs just as well as cards with massive OCs because Polaris itself is not terribly clock frequency starved, its bandwidth starved, as such my “bad” RX470 is actually better than most and cards like it massively underrated it is power efficient due to the high ASIC quality, and Samsung memory can roll with tighter timings than Hynix/Micron can so it doesn’t matter the memory won’t go above 2050MHz. Knowing these things you can make any Polaris card perform as well as any other. An example I can give here is a Powercolor RX580 Red Dragon and my best Sapphire RX580 Nitro, the former could do 1.45GHz and the memory being Samsung not over 2050MHz. The Nitro could do 1.425GHz and 2250MHz with excellent timings however neither card was really faster than the other because at that point both cards I had fully optimised for their hardware the limitation was not clock frequency or bandwidth but the hardware the GPU itself packs, there just wasn’t enough for further benefits. One of my optimised cards would perform amazingly well on a linux based system with Mesa drivers thats for sure, most likely beat out or equal stock GTX1070s and Vega 56/64s as the cards were very close to that mark even on Windblows.
@ket i have 3-4 memory errors in hwinfo when i playing games after one hour gaming at stock settings.
I have samsung rams.
To fix this i need to change the TCCDL from 3 to 4 only in the 1000 and 2000 straps or all of them?
My logic says only 1000 and 2000 as the card only uses the states 300 1000 and 2000.
(my card also take the same ammount of errors when i play at 1380 mhz and 1.075 volt)
@boombastik I don’t have the Samsung memory timing strap to hand I use but basically the only strap you need to change is the 2000MHz one although I change 1750MHz as well just to make sure performance is more consistent should the card throttle the memory speed when gaming, eg; at scene changes, that sort of thing.
I changed TCCDL from 3 to 4 to all straps (for samsung) and now i have zero errors.