I took delivery today of a 3070 FE which I bought on ebay, and after a few issues with performance, my troubleshooting efforts have indicated that the card appears to have a BIOS which doesnt belong on a FE.
Despite owning several NVIDIA GPUs in the past, including another 3070, I have never had to delve into this technical area before.
GPU-Z 2.52.0 and NVIDIA Control Panel show that installed BIOS version is 94.04.3A.40.A6 with a build date of 2020-12-03.
The info from the "General" section in the BIOS report from GPU-Z is as follows.
General
Version 94.04.3A.40.A6
Build Date 2020-12-03
Message PG142 SKU 10 VGA BIOS
Board Id GPU Board
The device ID reported in GPU-Z is 10DE 2484 - 10DE 146B. The one in NVIDIA Control Panel is 10DE 2484 146B10DE.
According to the BIOS database on this site, version 94.04.3A.40.A6 belongs to a Galax Boy SG, as shown at this link:
The same database suggests that the firmware on a 3070 FE should be 94.04.25.00.ED or 94.04.25.00.2B.
It looks like all of the limits in the installed BIOS are as they should be on the FE BIOS, and the card itself is physically immaculate. I'm aware that people flash alternative BIOS for mining etc, but it doesnt look like any mining card that I've seen on ebay - and there are many of them.
Considering that there are performance issues across two seperate systems that I've tried which are more than capable of handling the 3070, I think it's just going to have to go back anyway. But for my own peace of mind, am I misunderstanding what I'm seeing? Or could the database be wrong? The person who sold it didnt flash another BIOS themselves. They didn't buy it new either. But, its been working fine for them. They've offered me a refund if required, so I would like to confirm that my findings above are correct so I can give them some pointers. Unfortunately this goes beyond my level of understanding.
Thanks in advance for any advice.
Just a quick update...
I tried to grab the BIOS from the GPU using the latest version of GPU-Z, but got an error "BIOS reading not supported on this device".
Using v2.50.0 I was able to extract it, but the hashes dont match the ones for the two BIOSes which are held in the TPU database for 94.04.3A.40.A6. The ones I have are:
Algorithm Hash
--------- ----
SHA1 EA789522F74A54EDB4835DDAF31DFEDEA2124589
Algorithm Hash
--------- ----
MD5 8E6EA9172FF8434646172CA66BB4338B
The file sizes match - 976 KB (999,424 bytes) - So i guess the extraction worked. I cant find the BIOS that I have in the database. So, all I can assume at this point is that I somehow have a 3070 FE with an alternative BIOS which isn't in the database.
I was going to upload it into the database, but the upload box asks for only BIOSes which were shipped with the GPU, so I didn't bother.
From what I understand, the signing keys werent compromised in the hack last year and only signed BIOS can run, so it must be "official". The card is returning 3070 level results in all benchmarks - at least for a while, so it isnt a "fake".
Has anyone seen anything like this before? It all seems very weird.
Despite owning several NVIDIA GPUs in the past, including another 3070, I have never had to delve into this technical area before.
GPU-Z 2.52.0 and NVIDIA Control Panel show that installed BIOS version is 94.04.3A.40.A6 with a build date of 2020-12-03.
The info from the "General" section in the BIOS report from GPU-Z is as follows.
General
Version 94.04.3A.40.A6
Build Date 2020-12-03
Message PG142 SKU 10 VGA BIOS
Board Id GPU Board
The device ID reported in GPU-Z is 10DE 2484 - 10DE 146B. The one in NVIDIA Control Panel is 10DE 2484 146B10DE.
According to the BIOS database on this site, version 94.04.3A.40.A6 belongs to a Galax Boy SG, as shown at this link:
The same database suggests that the firmware on a 3070 FE should be 94.04.25.00.ED or 94.04.25.00.2B.
It looks like all of the limits in the installed BIOS are as they should be on the FE BIOS, and the card itself is physically immaculate. I'm aware that people flash alternative BIOS for mining etc, but it doesnt look like any mining card that I've seen on ebay - and there are many of them.
Considering that there are performance issues across two seperate systems that I've tried which are more than capable of handling the 3070, I think it's just going to have to go back anyway. But for my own peace of mind, am I misunderstanding what I'm seeing? Or could the database be wrong? The person who sold it didnt flash another BIOS themselves. They didn't buy it new either. But, its been working fine for them. They've offered me a refund if required, so I would like to confirm that my findings above are correct so I can give them some pointers. Unfortunately this goes beyond my level of understanding.
Thanks in advance for any advice.
Just a quick update...
I tried to grab the BIOS from the GPU using the latest version of GPU-Z, but got an error "BIOS reading not supported on this device".
Using v2.50.0 I was able to extract it, but the hashes dont match the ones for the two BIOSes which are held in the TPU database for 94.04.3A.40.A6. The ones I have are:
Algorithm Hash
--------- ----
SHA1 EA789522F74A54EDB4835DDAF31DFEDEA2124589
Algorithm Hash
--------- ----
MD5 8E6EA9172FF8434646172CA66BB4338B
The file sizes match - 976 KB (999,424 bytes) - So i guess the extraction worked. I cant find the BIOS that I have in the database. So, all I can assume at this point is that I somehow have a 3070 FE with an alternative BIOS which isn't in the database.
I was going to upload it into the database, but the upload box asks for only BIOSes which were shipped with the GPU, so I didn't bother.
From what I understand, the signing keys werent compromised in the hack last year and only signed BIOS can run, so it must be "official". The card is returning 3070 level results in all benchmarks - at least for a while, so it isnt a "fake".
Has anyone seen anything like this before? It all seems very weird.