Question GIGABYTE 5700 XT Bios mod fails

[AlleyCat]

Hi,
I am failing to update Gigabyte 5700 XT. The same procedure updates the BIOS on MSI cards with no problem. I follow the instructions on Igors Lab.

The sign of having trouble with the card bios flash shows in GPU-Z. After flash using amdvbflash the values of GPU and memory frequencies is empty. With stock bios there are Mhz frequencies.

Is there any known problems with flashing Gigabyte cards?
Any suggestions on what other forums I may ask for assistance?
Is it possible that the OEM bios is signed, and any modified bios will be rejected?
If the bios is signed, any tools to resign, or would I need to buy a card from a different vendor?

Thanks,

Alley Cat

 

[Mini_Me]

Hello everyone and thank you for this gold mine of a thread!
Really a lot of very useful and interesting information here.
Your dedication to benching/testing out all these different settings and especially doing so publicly and sharing all info is greatly appreciated!

We've been following this for a while and also successfully used the parameters outlined here: https://www.igorslab.de/community/threads/gigabyte-5700-xt-bios-mod-fails.3304/page-17#post-83096
in combination with the apple timings.

Now thank you for the recent update and the new MPT settings, we will create new modded Bios files for our cards later and try them.

I've got some questions though:
Are you suggesting of leaving the VDD and VDCCI (as well as the MVDD for Samsungs) fixed at what you wrote above and only vary only slightly in clock speed?
In the older post linked above there was a whole range of clock speeds and relating voltages. We tried out some of these and nearly all our cards run pretty stable at lower voltages than the fixed 800/850mV you wrote in today's update.
I've attached a recent screenshot to this post (yes, they're much too hot and we need to improve airflow in the rig).

Is this because if clocks are increased the chips need more power, but if you at the same time lover voltages naturally the current has to go up and higher current is heat and stress for the chip, so you are rather not undervolting too much?
I guess it also somehow relates to not exceeding the clock/voltage relation of 1.8 you mentioned so pages earlier when talking about the SoC?

Also I'd like to ask something else:
We're always modding the bios for every card using it's own stock bios as a base. For some cards this works well, but some (especially Asus Dual and XFX DD) just won't got below 130W reported in HiveOS. Now we once flashed a Sapphire Pulse with the Bios of another Pulse which reported lower and this helped very much.
So there has to be lot's of other stuff in that Bios's which we cannot alter with RBE or MPT and so we thought about using Bios's of other cards which run with nice low watts as base for modding a Bios for the Asus and XFX.
Does someone have any experiences in doing so and may be able to give any guide on how to choose a source Bios?
I read somewhere here that Fan settings/curve is important, which makes sense, as the fand are one obvious physical difference between custom cards.
Do you know of a way to transfer fan settings between Bios files, to maybe use another Bios as base for a card, but implement the fan stuff from the target cards stock?
Or are there other things which makes cross flashing Bios's between cards a rather bad idea? I've seen someone using an XFX Bios on Gigabyte cards once.

Thank you very much and best regards.

Hello and welcome,

Regarding your appreciation I am going to direct it all to AlleyCat as he is the one who started this and performed all the tests by himself to prove the information that were collected is not only theoretically set however practically as well and he is still doing that,

Regarding the previous parameters on page 17 when the Maximum Voltage GFX was set at 850 mV even thought it is working fine however on the long run the memory may degrade rapidly and die, it may not be a problem for some people to buy a replacement for the memory as it is only cost 8$ per 1 piece equal of 1GB however the replacement process requires a special oven to achieve that which I believe many do not have it or can afford it as well as this process is a time and effort consuming which make it not worth the hassle for the masses,

Regarding the safe minimum ratio of 1.8 between the clock and voltage is for the Core only as for the SoC and other parameters are unknown at the meantime,

Regarding the current parameters as advised before, I highly recommend it if anyone wants their cards not to fail and die on the long run,
When increasing the clock either for the memory or the core it will needs sufficient power to operate properly, the ratio of 1.8 is the safe minimum required voltage as lower than that the voltage may not be sufficient and if it is still sufficient then another factor is involved which is the amount of the current pumped to the core will increase and make it unstable, therefore a balance is required between the voltage and current, by setting a pool of an amount of voltage and limiting the current with the safe minimum amount required for that amount of voltage set before then at a specific clock, it will pull sufficient voltage from that pool and at the same time it will pull the safe minimum required current as well to achieve the required power at that clock,

Regarding porting vbios of a card to another card with different brand while maintaining the existing fan curve, it is possible and it is easy to achieve that as you mentioned in your post "to maybe use another Bios as base for a card, but implement the fan stuff from the target cards stock" however it is recommended to consider the number of fans and their power requirements, and this method is only for those who want to use the stock vbios of another card to try a different power consumption and skip vbios modding, other than that modding the vbios is better for controlled operation.

Regarding cross flashing vbios, it may has its own risks as the vbios was designed to meet the specification of the card cooling solution, one can try to test and verify the results by himself.

 

[AlleyCat]

You are very welcome,

I would like to request your help regarding one final thing to disclose the optimization of the vbios for the RX 5700 which as below,

As you have Gigabyte RX 5700 that comes only with Micron which is perfect for this final test,

What the test is about? Copying the vram timings of Samsung memory to the Micron memory once using the Apple straps and once using the non Apple with option 2 mentioned before,

Please no need to rush nd feel free to perform this test whenever it is possible and whenever you like.

Thank you for waiting for fore the results.
There is little difference between the three options memory straps.

Tests Recap:
Test 1) Apple Straps
Test 2) Apple vram timings of Samsung memory to the Micron memory
Test 3) timing as bellow.
- K4Z80325BC (Samsung)
(1500 or 1550) MHz (Copy) - - - > (Paste) 1800 MHz
1800 MHz - - - > 2000 MHz
2000 MHz - - - > 2250 MHz

- MT61K256M32 (Micron)
(1500 or 1550) MHz - - - > (1750 - 1800) MHz
1750 MHz - - - > (1875 - 2000) MHz

Memory Clock - tREF
1000 MHz - 3900 (Samsung & Micron)
1250 MHz - 4875 (Samsung & Micron)
(1500 or 1550) MHz - 6045 (Samsung & Micron)
1750 MHz - 6825 (Micron only)
1800 MHz - 7020 (Samsung & Micron)
1875 MHz - 7315 (Micron only)
2000 MHz - 7800 (Samsung & Micron)
2250 MHz - 8775 (Samsung only)

If I have to choose the "best mod", I would choose the Apple straps, because they are consistent over multiple cards and looks reliable.

HiveOS settings: Core: 1395, VDD 770mv Mem 935Mhz, VDDCI 790mv MVDD 0 (default)
Results:
Test 1) Speed: Around 57.34 MH/s, 111W
Test 2) Speed: Around 56.61 MH/s, 111W
Test 3) Speed: Around 57.14 MH/s, 112W

 

[Mini_Me]

Thank you for waiting for fore the results.
There is little difference between the three options memory straps.

Tests Recap:
Test 1) Apple Straps
Test 2) Apple vram timings of Samsung memory to the Micron memory
Test 3) timing as bellow.
- K4Z80325BC (Samsung)
(1500 or 1550) MHz (Copy) - - - > (Paste) 1800 MHz
1800 MHz - - - > 2000 MHz
2000 MHz - - - > 2250 MHz

- MT61K256M32 (Micron)
(1500 or 1550) MHz - - - > (1750 - 1800) MHz
1750 MHz - - - > (1875 - 2000) MHz

Memory Clock - tREF
1000 MHz - 3900 (Samsung & Micron)
1250 MHz - 4875 (Samsung & Micron)
(1500 or 1550) MHz - 6045 (Samsung & Micron)
1750 MHz - 6825 (Micron only)
1800 MHz - 7020 (Samsung & Micron)
1875 MHz - 7315 (Micron only)
2000 MHz - 7800 (Samsung & Micron)
2250 MHz - 8775 (Samsung only)

If I have to choose the "best mod", I would choose the Apple straps, because they are consistent over multiple cards and looks reliable.

HiveOS settings: Core: 1395, VDD 770mv Mem 935Mhz, VDDCI 790mv MVDD 0 (default)
Results:
Test 1) Speed: Around 57.34 MH/s, 111W
Test 2) Speed: Around 56.61 MH/s, 111W
Test 3) Speed: Around 57.14 MH/s, 112W

Thank you very much for your time and effort,

Have you noticed any other differences regarding the power consumption and the memory temperature for the Micron when using Samsung timings compared to its original timings?

 

[AlleyCat]

Thank you very much for your time and effort,

Have you noticed any other differences regarding the power consumption and the memory temperature for the Micron when using Samsung timings compared to its original timings?

Not really.

 

[Mini_Me]

Not really.

Very well,
Regarding the current power table, is there any difference from the previous one? And would like to suggest or recommend or advise anything to add to the post below,

Question - GIGABYTE 5700 XT Bios mod fails

Hi, I am failing to update Gigabyte 5700 XT. The same procedure updates the BIOS on MSI cards with no problem. I follow the instructions on Igors Lab. The sign of having trouble with the card bios flash shows in GPU-Z. After flash using amdvbflash the values of GPU and memory frequencies is...

www.igorslab.de

By the way, would you like to test the SoC max clock at 1093 MHz to check whether it will impact the power consumption or not however using a script or MorePowerTool from inside the system to set it at 950 MHz which is the safer way?

 

[heavyarms1912]

Just to clarify, you are referring to the memory speed, and not the core speed.

I can run the Samsung memory chips at 900 clocks easily with mvddci at 725mV and mvdd at 1350mV. Higher clocks cause instability. Also, reduced the SoC clocks to 950 max. Helps to reduce power. Note: SoC clocks will remain at 950 only if memory clocks are below or at 912. Higher memory clocks will change SoC clocks to higher values.

 

[heavyarms1912]

Understandable, the reason is not the parameters however unfortunately it is the cooling of the card is not sufficient,
The blower fan is not enough to cool down the memory at 1820 MHz,
and for the dual fan beyond 1840 MHz as well and if someone is lucky he may be able to run it at max 1860 MHz,
However on water cooling is different story, it is easy to run it stably at 990 MHz and even at 1100 MHz for the 50th Anniversary edition.

Please note that it is important for Samsung memory to run 24/7 safely without degradation, the temperature must not exceed 74 ° C, therefore it is highly recommended to fix the MVDD at 1350 mV (it is the default value at 1750 MHz with already added headroom) as low insufficient MVVD will increase the memory temperature and to increase the VDDCI (default for 1750 MHz is 850 mV) until reaching the desired memory clock without exceeding the 74 ° C, if the temperature was exceeding the 74 ° C to 85 ° C then every 48 hours of continuous running, the system should be given a break of 1 to 2 hours depending on the ambient temperature, if the memory is running at temperature more than 85 ° C then each 24 hour of continuous running, the system should be given a break of 1 to 2 hours depending on the ambient temperature.

74 degrees on memory temps? Almost everyone would be hitting that temperature and even gamers on daily runs.

 

[AlleyCat]

Very well,
Regarding the current power table, is there any difference from the previous one? And would like to suggest or recommend or advise anything to add to the post below,

Question - GIGABYTE 5700 XT Bios mod fails

Hi, I am failing to update Gigabyte 5700 XT. The same procedure updates the BIOS on MSI cards with no problem. I follow the instructions on Igors Lab. The sign of having trouble with the card bios flash shows in GPU-Z. After flash using amdvbflash the values of GPU and memory frequencies is...

www.igorslab.de

By the way, would you like to test the SoC max clock at 1093 MHz to check whether it will impact the power consumption or not?

I modified the SoC max clock to 1093 and at 935 Mhz the SoC clock is the same at 1085.
The power table is the same table in your link.

I love Rule 1.8. I can tune the power by selecting the core mv and multiply it by 1.8 to get the corresponding frequency.

In other words, I can select efficiency or performance at the same memory clock speed by simply increasing the clock and the clock/1.8 for VDD.

 

[AlleyCat]

I can run the Samsung memory chips at 900 clocks easily with mvddci at 725mV and mvdd at 1350mV. Higher clocks cause instability. Also, reduced the SoC clocks to 950 max. Helps to reduce power. Note: SoC clocks will remain at 950 only if memory clocks are below or at 912. Higher memory clocks will change SoC clocks to higher values.

I confirm that at Mem speed 912 I get SoC clock 950.
However, I don't know the long term effect of running the memory at 725mv.

 

[heavyarms1912]

I confirm that at Mem speed 912 I get SoC clock 950.
However, I don't know the long term effect of running the memory at 725mv.

I've been running them for couple of months now. I found instability in reducing mvdd though. Some do run fine at 1300mV.
Also, I see a lot of folks mentioning Phoenix Miner. It's just exaggerated hashrate on dashboard in comparison to TRM. Do the test yourself if you would like to. TRM actually is more consistent and better even though the below article has rated it higher over PM.

Claymore’s Ethereum Miner Stops Working. Best Alternatives –...

The popular Claymore’s miner will stop mining Ethereum Classic and Ethereum soon. We tested popular alternatives for the Ethash algorithm. 2Miners launched the test pool and experimented with the AMD and Nvidia graphics cards. ContentsClaymore’s Dual MinerClaymore’s Dual MinerPool sent wrong...

2miners.com

 

[Mini_Me]

I confirm that at Mem speed 912 I get SoC clock 950.
However, I don't know the long term effect of running the memory at 725mv.

Great information, one last thing regarding SoC clock,

Please, could you check the amd-info and the soc value for the settings below,

Core clock 1440 MHz, VDD at 800 mV,
Memory clock (960 and 950) MHz, VDDCI 865 mV, MVDD 1365 MHz.

 

[AlleyCat]

Great information, one last thing regarding SoC clock,

Please, could you check the amd-info and the soc value for the settings below,

Core clock 1440 MHz, VDD at 800 mV,
Memory clock (960 and 950) MHz, VDDCI 865 mV, MVDD 1365 MHz.

With the settings above, ": final Power Table, SoC Mx Freq. 1093.
Mem 950 Mhz results with 58.36MH/s @117W (RTM reports SoCMHz 1085)
Mem 960 Mhz results with 58.37MH/s for 1 min and the GPU crashed.

 

[Mini_Me]

I can run the Samsung memory chips at 900 clocks easily with mvddci at 725mV and mvdd at 1350mV. Higher clocks cause instability. Also, reduced the SoC clocks to 950 max. Helps to reduce power. Note: SoC clocks will remain at 950 only if memory clocks are below or at 912. Higher memory clocks will change SoC clocks to higher values.

Greetings and welcome,

Thank you for your kind and informative sharing,

Regarding the SoC, it is already being known reducing it will decrease the power consumption in a percentage however the problem is we do not have the solid knowledge to determine what soc clock reduction will impact on the long run other than the memory clock,

Because the lack of knowledge at the meantime, unfortunately there are many things were not made available to the users when creating the MorePowerTool,

I am trying my best to dig the information by even visiting the manufacturer factory for the pcb components of RX 5700 in other countries however it seems AMD made sure that many informations regarding their products to stay in the dark however I believe it is a matter of time before this informations see the light.

 

[AlleyCat]

Greetings and welcome,

Thank you for your kind and informative sharing,

Regarding the SoC, it is already being known reducing it will decrease the power consumption in a percentage however the problem is we do not have the solid knowledge to determine what soc clock reduction will impact on the long run other than the memory clock,

Because the lack of knowledge at the meantime, unfortunately there are many things were not made available to the users when creating the MorePowerTool,

I am trying my best to dig the information by even visiting the manufacturer factory for the pcb components of RX 5700 in other countries however it seems AMD made sure that many informations regarding their products to stay in the dark however I believe it is a matter of time before this informations see the light.

You rock Mini_me.
I am sure you will be able to figure, and with the community's assistance if requires.
Best.
AlleyCat

 

[Mini_Me]

With the settings above, ": final Power Table, SoC Mx Freq. 1093.
Mem 950 Mhz results with 58.36MH/s @117W (RTM reports SoCMHz 1085)
Mem 960 Mhz results with 58.37MH/s for 1 min and the GPU crashed.

thank you very much, I believe for Mem 960 MHz the soc clock should be 1100 MHz,

It seems the information I got from one of the RX 5700 vbios programmer that the value of SoC clock is 1090 MHz that the value which set for the max stable clock 1910 MHz and the max SoC value of 1267 MHz is for Mem clock at 2250 MHz.

 

[Sartori71]

Please use below for best result safely running 24/7,

Regarding parameters settings as below,

- Core parameters,
Core clock from (1360 to 1440) MHz, VDD at 800 mV,

- Memory parameters,

~Samsung K4Z80325BC-HC14,
Memory clock at (905 or 910) MHz, VDDCI at 850 mV (less than this is not recommended), MVDD at 1350 mV (less than this is not recommended),

~Micron MT61K256M32JE-14,
Memory clock at (905 or 910) MHz, VDDCI at 850 mV (Recommended minimum is 845 mV), MVDD (1310 or 1350) mV [the recommended minimum for MVDD Micron MT61K256M32 at POD135 DC operating conditions is 13095 mV according to Micron Inc.],

Ok, so I've used the Micron settings and core parameters as suggested. Rig is currently running stable, but it's only been an hour or so. With that being said, I'm now seeing a definite drop in MH/s on the rev 2.0 cards only (Gigabyte 5700xt Gaming OC rev 2.0). The rev 1.0 cards have a negligible drop, and about a 10W increase. With my previous settings, all cards had identical settings as well, the MH/s was similar between the rev 1.0 and rev 2.0, it was just the watts that were different (about 20W higher on the rev 2.0) I still have the same 20W difference. I would like to bring the wattage down on all cards still, and try to close the gap on the MH difference between rev 1 and rev 2 cards. Also to clarify, I'm using apple straps and the suggested MPT settings for the gigabyte cards from this thread. My current settings in the screenshot, the two circled in red are the rev 2.0 cards.

 

[AlleyCat]

thank you very much, I believe for Mem 960 MHz the soc clock should be 1100 MHz,

It seems the information I got from one of the RX 5700 vbios programmer that the value of SoC clock is 1090 MHz that AMD set for the max stable clock 1910 MHz and the max SoC value of 1267 MHz is for Mem clock at 2250 MHz.

I think we need to focus on the Power curve.
Allow me to explain.
We want to find a simple methodology for tuning cards. The Rule 1.8 is a good start for tuning the core speed. It gives a core clock ranges from 1300MHz to 1450 to achieve the speed of the efficiency. For example, I found that at core speed 1400, I am getting a balance between speed and efficiency., The card AMD RX5700 reference (Micron mem, non-XT) runs at about 111W. If I wish to run the card at lower power, I can start reducing the core and the corresponding VDD (clock/1.8) to the desired power.

The issue starts when attempting to tune the memory speed. The SoC looks like a big factor in power consumption. If we set a "safe" curve for the memory, we could perhaps ignore the values of VDDCI and MVDD, letting the curve decide the values.

With SoC max frequency set to 1093, I am running the test card at 1400/770 (1.82) 111w. I set the memory frequency to 945, the most stable speed I achieved, and put no value for VDDCI or MVDD. I assume that the values are taken from the curve. The speed is 57.24MH/s, and the configuration is simple.

If we could get a "performance" curve, we will need no values for VDDCI or MVDD.

Do you agree with my assertion?

AlleyCat

 

[Mini_Me]

You rock Mini_me.
I am sure you will be able to figure, and with the community's assistance if requires.
Best.
AlleyCat

Please no need to mention it and you rock more than me dear,

What I have been informed as well regarding the SoC clock, if we set it on a value for an example 950 MHz for Mem 910 MHz or we left at the default values, the soc will automatically be set according to the memory clock.

I would really appreciate whenever it is possible to verify this to check if the power consumption still be the same for memory clock at 910 MHz when setting the soc in vbios at 950 MHz or when leaving the soc at the default values,

You may have a good day,

Best Wishes.

 

[Mini_Me]

Ok, so I've used the Micron settings and core parameters as suggested. Rig is currently running stable, but it's only been an hour or so. With that being said, I'm now seeing a definite drop in MH/s on the rev 2.0 cards only (Gigabyte 5700xt Gaming OC rev 2.0). The rev 1.0 cards have a negligible drop, and about a 10W increase. With my previous settings, all cards had identical settings as well, the MH/s was similar between the rev 1.0 and rev 2.0, it was just the watts that were different (about 20W higher on the rev 2.0) I still have the same 20W difference. I would like to bring the wattage down on all cards still, and try to close the gap on the MH difference between rev 1 and rev 2 cards. Also to clarify, I'm using apple straps and the suggested MPT settings for the gigabyte cards from this thread. My current settings in the screenshot, the two circled in red are the rev 2.0 cards.

Please follow the guide in the kindly attached link below for your rev. 2.0 cards,

Question - GIGABYTE 5700 XT Bios mod fails

Hi, I am failing to update Gigabyte 5700 XT. The same procedure updates the BIOS on MSI cards with no problem. I follow the instructions on Igors Lab. The sign of having trouble with the card bios flash shows in GPU-Z. After flash using amdvbflash the values of GPU and memory frequencies is...

www.igorslab.de

 

[AlleyCat]

Please no need to mention it and you rock more than me dear,

What I have been informed as well regarding the SoC clock, if we set it on a value for an example 950 MHz for Mem 910 MHz or we left at the default values, the soc will automatically be set according to the memory clock.

I would really appreciate whenever it is possible to verify this to check if the power consumption still be the same for memory clock at 910 MHz when setting the soc in vbios at 950 MHz or when leaving the soc at the default values,

You may have a good day,

Best Wishes.

Here are the results:

SoC freq is the same at 950MHz.

At memory clock 910 with SoC Max Freq. 950, the RTM report SoCMHz = 950. MHs/s = 55.93 and Power = 111W.
At memory clock 910 with SoC Max Freq. 1267 (default), the RTM report SoCMHz = 950. MHs/s = 55.92 and Power = 110W.

The difference is insignificant.

Both tests used core 1400 MHz, VDD=770.