Question GIGABYTE 5700 XT Bios mod fails

AlleyCat

Veteran
Mitglied seit
Okt 9, 2020
Beiträge
290
Bewertungspunkte
17
Punkte
17
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
 
Zuletzt bearbeitet :
OK. Thanks for the updates.
You are very welcome,

I hope that you had a comfortable sleep and sweet dreams,

When you have time, we are going to test 3 settings however before that I believe the memory clock from (920 - 950) MHz is not stable on Samsung memory therefore, first we are going to push the memory clock beyond that to 970 MHz with the modified straps like before once and with the original straps once again, adjusting the MVDD and VDDCI only if it is needed and to check each time, if it is not a success then we do the below:

One, copy the 1250 MHz strap of Samsung to the rest after it with tREF 6000 instead of 9000 for now,
If this did not work then,
Second, copy the (1500 or 1550) MHz strap of the micron to the 1550 MHz strap of the Samsung, change tREF to 6000 and copy to the rest after it,
Third, copy the 1875 MHz strap of the micron to the 1800 MHz strap of the Samsung, change tREF to 6000 and copy to the rest after it,

Start to test with core clock of 900 MHz,

Please after you check both do update me whenever it is possible.

Best Regards
 
Zuletzt bearbeitet :
OK. Thanks for the updates.
Finally I managed to get the most optimized vram timings for RX 5700/5700 XT crads attached kindly below,

It was done by Apple Inc. engineers for best performance, stability, safety to power consumption especially for Samsung memory and I believe that it is possible to run the memory stably from (920 to 950) MHz with these timings,
Copy and paste accordingly to the order of the core clock.
 

Anhänge

  • K4Z80325BC&MT61K256M32_GDDR6_Optimized_Timings.zip
    974 Bytes · Aufrufe : 2.626
Zuletzt bearbeitet :
Finally I managed to get the most optimized vram timings for RX 5700/5700 XT crads attached kindly below,

It was done by Apple Inc. engineers for best performance, stability, safety to power consumption especially for Samsung memory and I believe that it is possible to run the memory stability from (920 to 950) MHz with these timings,
Copy and paste accordingly to the order of the core clock.
Thank you for the great engineering work. I am testing now. My initial response that that the Sapphire with Samsung memory will not run steady above 915MHz memory clock. I am running now different power/speed scenarios and will get back shortly with results.
 
Thank you for the great engineering work. I am testing now. My initial response that that the Sapphire with Samsung memory will not run steady above 915MHz memory clock. I am running now different power/speed scenarios and will get back shortly with results.
It was not me however it was by expert professionals of Apple Inc.

I am looking for the results as I believe there is still a headroom to achieve better results.
 
It was not me however it was by expert professionals of Apple Inc.

I am looking for the results as I believe there is still a headroom to achieve better results.

Please find is a testing table. I progressively reduced the power, and the best stable performance I achieved at 1400 MHz core at 775mv. The best stable prformance on memory was achieved at 910MHz, 800MHz, 1.35V
 

Anhänge

  • Apple Inc straps (Samsung Mem).png
    Apple Inc straps (Samsung Mem).png
    328,6 KB · Aufrufe : 539
Zuletzt bearbeitet :
Please find is a testing table. I progressively reduced the power, and the best stable performance I achieved at 1400 MHz core at 775mv. The best stable prformance on memory was achieved at 910MHz, 800MHz, 1.35V
Alright, it seems this is a universal setting, let us improve it for mining purpose,
How about we proceed with what is mentioned before below,
You are very welcome,

I hope that you had a comfortable sleep and sweet dreams,

When you have time, we are going to test 3 settings however before that I believe the memory clock from (920 - 950) MHz is not stable on Samsung memory therefore, first we are going to push the memory clock beyond that to 970 MHz with the modified straps like before once and with the original straps once again, adjusting the MVDD and VDDCI only if it is needed and to check each time, if it is not a success then we do the below:

One, copy the 1250 MHz strap of Samsung to the rest after it with tREF 6000 instead of 9000 for now,
If this did not work then,
Second, copy the (1500 or 1550) MHz strap of the micron to the 1550 MHz strap of the Samsung, change tREF to 6000 and copy to the rest after it,
Third, copy the 1875 MHz strap of the micron to the 1800 MHz strap of the Samsung, change tREF to 6000 and copy to the rest after it,

Start to test with core clock of 900 MHz,

Please after you check both do update me whenever it is possible.

Best Regards
 
Alright, it seems this is a universal setting, let us improve it for mining purpose,
How about we proceed with what is mentioned before below,
The memory speed will not go to 970. The miner reports 900. Do we need to change the Power Table to allow the card to accept a 970?
Or, could it be a problem with HiveOS?
I will try to set a Phoenix miner switch -mclock 1940
 
The memory speed will not go to 970. The miner reports 900. Do we need to change the Power Table to allow the card to accept a 970?
Or, could it be a problem with HiveOS?
I will try to set a Phoenix miner switch -mclock 1940
Let us make another power table as below:

- Under Features tab enable all the overdive features,
- Under Overdrive Limits tab, set:
GFX Maximum Clock 2050 MHz
Memory Maximum Clock 1000 MHz
Under Power and Voltages, set:
Maximum Voltage GFX 1200 mv
Maximum Voltage SoC 1200 mv
Minimum Voltage GFX 750 mv
Minimum Voltage SoC 750 mv
Power Limit GPU 180 W
TDC Limit GFX 168 A
TDC Limit SoC 12 A
- Under Frequency tap, set:
GFX Maximum 1400 MHz
Memory DPM 3 - 950 MHz
- Under Fan tab, set:
Fan Acoustic Limit 1550 RPM
Fan Throttling 3200 RPM

After that we are going to test in windows with MSI afterburner.
 
Zuletzt bearbeitet :
The memory speed will not go to 970. The miner reports 900. Do we need to change the Power Table to allow the card to accept a 970?
Or, could it be a problem with HiveOS?
I will try to set a Phoenix miner switch -mclock 1940
I have results for test 1.
Straps made by Apple Inc
Mem speed set by Phoenix option -mclock 19xx
All the speeds crashed the driver

Test 2 will be with the above new Power Table, using Apple Inc straps.
 

Anhänge

  • Apple Inc straps (Samsung Mem) 970-955.png
    Apple Inc straps (Samsung Mem) 970-955.png
    109,9 KB · Aufrufe : 450
Let us make another power table as below:

- Under Features tab enable all the overdive features,
- Under Overdrive Limits tab, set:
GFX Maximum Clock 2250 MHz
Memory Maximum Clock 1125 MHz
Under Power and Voltages, set:
Maximum Voltage GFX 1250 mv
Maximum Voltage SoC 1250 mv
Minimum Voltage GFX 750 mv
Minimum Voltage SoC 750 mv
Power Limit GPU 190 W
TDC Limit GFX 176 A
TDC Limit SoC 14 A
- Under Frequency tap, set:
GFX Maximum 1550 MHz
Memory DPM 3 - 1125 MHz
- Under Fan tab, set:
Fan Acoustic Limit 1380 RPM
Fan Throttling 2900 RPM

After that we are going to test in windows with MSI afterburner.
WIth the above power table the Phoenix miner in Windows 10 crash .
Note that I used the straps provided by Apple Inc
 
WIth the above power table the Phoenix miner in Windows 10 crash .
Note that I used the straps provided by Apple Inc
This power table is meant to increase the voltages of the memory MVDD to 1390 mv, VDDCI to 900 mv, the SoC to 1200 mv, the memory clock to (970 - 990) MHz and the core voltage at 1200 mv. as a start and it is a success stably then we tune the voltages, if not a success then let us return to the power table of before and as a final attempt to do below for both standard straps and the Apple one:

One, copy the 1250 MHz strap of Samsung to the rest after it with tREF 6000 instead of 9000 for now,
If this did not work then,

Second, copy the (1500 or 1550) MHz strap of the micron to the 1550 MHz strap of the Samsung, change tREF to 6000 and copy to the rest after it,

Third, copy the 1875 MHz strap of the micron to the 1800 MHz strap of the Samsung, change tREF to 6000 and copy to the rest after it,

Start to test with core clock of 900 MHz,

Please do update me whenever it is possible.

Best Regards
 
I am lost. Too many tests variations.
What would be the next test?
I believe we can achieve memory clock of 925 MHz stably at most for your chip and around 58 MH/s and 110 to 115 watt,
This is the final resort after all attempts before to achieve better than 58 MH/s is not workibg, by setting the MVDD to 1390 mv and VDDCI to 900 mv and memory clock to 925 MHz after that decrease the MVDD and VDDCI until you reach minimum voltages to run the memory clock at 925 MHz or 920 MHz and that is the best we achieved before unfortunately for the Samsung memory.
 
I believe we can achieve memory clock of 925 MHz stably at most for your chip and around 58 MH/s and 110 to 115 watt,
This is the final resort after all attempts before to achieve better than 58 MH/s is not workibg, by setting the MVDD to 1390 mv and VDDCI to 900 mv and memory clock to 925 MHz after that decrease the MVDD and VDDCI until you reach minimum voltages to run the memory clock at 925 MHz or 920 MHz and that is the best we achieved before unfortunately for the Samsung memory.
It looks like Apple's straps work optimal on my Micron cards.
I think these are stunning numbers.
Total rig with 6 cards 296MH/s, power at 580W (reported by HiveOS) at a ratio of 0.51 MH/s/W
All the cards took the Apple Inc straps, and all the cards are with Micron memory.
I may modify the power consumption bellow 580 as long as I don't lose too much hash or the ratio below 0.5.

Please note that the rig is "fresh" and will need to "cook" for a day to establish that it is stable.

The remaining issue is what best configuration we could find for cards with Samsung memory. I will run the tests above with the Apple Inc straps.

Just to note, the miner if Phoenix 5.1c, and the configuraitons are:
-coin %COIN%
-amd
-hstats 2
-gt 6
 

Anhänge

  • Screen Shot 2020-10-19 at 2.13.27 PM.png
    Screen Shot 2020-10-19 at 2.13.27 PM.png
    246,9 KB · Aufrufe : 460
It looks like Apple's straps work optimal on my Micron cards.
I think these are stunning numbers.
Total rig with 6 cards 296MH/s, power at 580W (reported by HiveOS) at a ratio of 0.51 MH/s/W
All the cards took the Apple Inc straps, and all the cards are with Micron memory.
I may modify the power consumption bellow 580 as long as I don't lose too much hash or the ratio below 0.5.

Please note that the rig is "fresh" and will need to "cook" for a day to establish that it is stable.

The remaining issue is what best configuration we could find for cards with Samsung memory. I will run the tests above with the Apple Inc straps.

Just to note, the miner if Phoenix 5.1c, and the configuraitons are:
-coin %COIN%
-amd
-hstats 2
-gt 6
Great news, I am really glad that this was useful to you,

I would be grateful if you could help whenever it is possible in checking if applying the test 1 settings to Apple timings, will it work stably and will it yield to a better result?

Many thanks in advance.
 
I believe we can achieve memory clock of 925 MHz stably at most for your chip and around 58 MH/s and 110 to 115 watt,
This is the final resort after all attempts before to achieve better than 58 MH/s is not workibg, by setting the MVDD to 1390 mv and VDDCI to 900 mv and memory clock to 925 MHz after that decrease the MVDD and VDDCI until you reach minimum voltages to run the memory clock at 925 MHz or 920 MHz and that is the best we achieved before unfortunately for the Samsung memory.
It looks like Apple's straps work optimal on my Micron cards.
I think these are stunning numbers.
Total rig with 6 cards 296MH/s, power at 580W (reported by HiveOS) at a ratio of 0.51 MH/s/W
All the cards took the Apple Inc strips, and all are Micron memory.
I may modify the power consumption bellow 580 as long as I don't lose too much hash or the ratio below 0.5.

Please note that the rig is "fresh" and will need to "cook" for a day to establish that it is stable.

The remaining issue is what best configuration we could find for cards with Samsung memory. I will run the tests above with the Apple Inc straps.
Great news, I am really glad that this was useful to you,

I would be grateful if you could help whenever it is possible in checking if applying the test 1 settings to Apple timings, will it work stably and will it yield to a better result?

Many thanks in advance.
I accept the request, and I will do Test 1 with Apple's straps. I will do the test on the card with the Samsung memory.
 
Oben Unten