Originally Posted by Jakob27
My card has hynix memory. That means i need the F4 bioses. So, i should replace the D_D F10 bios with D_P F4 on the DVI-I side, and leave the other D_D F4 where it's at. Please confirm this. I really don't want to mess this up.
Yes, assuming you are aware of how to "switch" from DD to DP (vice versa) as there is no physical BIOS switch.
What I see from your logs is a power delivery
problem. This is obvious from the PWR perfcaps but the log tells a more detailed story. Your GPU hasn't ever exceeded 70% TDP, which is basically showing that it is starved for power. Every time the GPU attempts to boost, which requires an increased power draw, it is unable to "pull" the power it needs and therefore repeatedly power throttles. This will cause the voltage to be unable to reach its highest capability as well, so the cause is lack of power and the side affect is limited voltage.
A 980Ti with the MUMOD BIOS is capable of drawing up to 475W of power maximum.. You'll almost never reach this but it is possible. 200W maximum for each 8-pin PCI-e cable and 75W maximum for the PCI-e slot. I forget (been a while) but I think the stock BIOS is limited to 300W max or something like that.
The attached excel spreadsheet (pivot tab) will help you see that at each GPU boost speed (Mhz) the power consumption never really exceeds 70%... where it would be normal to see well over 100% TDP (estimation is that you are losing 30%-50% overall power). Looking closer (you can play around with the pivot) you can also see that the power issue is present on both 8-pin PCI-e cables as well as the PCI-e slot. I could be wrong but this information tells me that you've likely got a hardware issue here. If it was showing a power delivery problem on one of the 3 sources of power then it could be a bad cable, bad PCI-e slot, MB needs PCI-e supplemental power etc... but when all three of the power sources are not able to utilize (pull) all of the power allowed and the GPU is complaining of being starved for power, there is literally something preventing the power delivery from occurring and I think it is hardware based.
It goes withouth saying, IF you are using any PCI-e splitters/adapters/combiners... DON'T... and of course the PSU needs to be beefy enough (750W bare minimum). All PCI-e cables should be a dedicated 8-pin, straight run to the PSU directly.
I can only guess as to what the true cause is but you can figure it out through process of elimination:
(1) Try the GPU in another PC if you can.. if you see the same behavior then you have ruled out the PSU, cables, MB etc all in one test. You never know.
(2) Inspect the PCI-e solder joints for any cold solders or cracked solder. A weak connection would appear to work @ idle but unable to pass a lot of current @ load.
(3) Remove the cooler and inspect the GPU. You may find a damaged component or I guess possibly the TIM/PADS could be missing/dried/cracked and something is quickly overheating (but I don't think so at this point). It cannot hurt to look.
Was a power shunt mod ever done to this card? I'm not suggesting that you do it (we can modify the BIOS, so it is not necessary) but if one was done, that should be removed.
Good luck and I hope you can get it sorted.
Let us know
* Power. Indicating perf is limited by total power limit.
NV_GPU_PERF_POLICY_ID_SW_POWER = 1,
* Thermal. Indicating perf is limited by temperature limit.
NV_GPU_PERF_POLICY_ID_SW_THERMAL = 2,
* Reliability. Indicating perf is limited by reliability voltage.
NV_GPU_PERF_POLICY_ID_SW_RELIABILITY = 4,
* Operating. Indicating perf is limited by max operating voltage.
NV_GPU_PERF_POLICY_ID_SW_OPERATING = 8,
* Utilization. Indicating perf is limited by GPU utilization.
NV_GPU_PERF_POLICY_ID_SW_UTILIZATION = 16,