Differences

This shows you the differences between two versions of the page.

--- 980ti [2020/12/05 06:19]
admin
+++ 980ti [2020/12/05 06:26] (current)
admin
@@ Line 25: / Line 25: @@
 FIXME IMGIMGIMG FIXME
+  * Disable voltage control – this feature prevent Precision X to control any voltage settings.
+  * Turn off KBOOST limitation – this feature allow you to still enable KBoost even if using VGA output (normally it’s disabled, due rare compatibility issues)
+===Enabling real voltage monitor===
+There is one more trick, regarding voltage readout in Precision X and K|NGP|N Edition cards (both 980 and 980Ti flavours). As you all already know, software is unable to show real voltage supplied to GPU, it shows what GPU VID setting currently is, not actual voltage. Most of users are not aware of this detail, and expect to see real voltage. That’s why we often get confused people in forums, who see LN2 records with 2000MHz clocks and asking how is that possible with only 1.212V?
+Well, KPE cards have a solution for this, as Precision X can actually read real voltage. KPE PCB have special circuitry to measure real voltage delivered to GPU chip and Precision X can get that reading. This is not default enabled feature, so to enable real GPU voltage monitoring, follow few simple steps before you start your session:
+  * Download latest Precision X, at moment of writing it was 5.3.6
+  * Set both OVERVOLTAGE and OVERBOOST mode in Voltage section top right. No need change voltages.
+  * Now VOLTAGE in center section will read actual voltage delivered by hardware.
+  * You can disable voltage control via secret menu after, reading still will work.
+Screenshot below shows example usage with 1.400V set. GPU Voltage is reported correctly in hardware monitor graph, log and OSD as well, which can be handy to monitor voltage droop/change during benchmark sessions. And after you found sweet spot for voltage, disable OSD and run for records (OSD have little performance hit when running, so you want it OFF when benching for records).
+FIXME IMGIMGIMG FIXME
+There is no limit on reading, so even 1.8V will be reported correctly. Again, this works only for EVGA GeForce GTX 980 KPE and EVGA GeForce GTX 980Ti KPE cards.
+====Overclocking methodology for Big Maxwell (GM200)====
+FIXME IMGIMGIMG FIXME
+EVGA 980 Ti KINGPIN as well as all other 980 Ti, 980 and Titan-X cards use a GPU featuring the NVIDIA Maxwell architecture. Given what we already know about it highlighted here before, we know that Maxwell GPUs don’t really benefit from much overvoltage on air and water cooling temps. If you give it some thought, it’s due to fact that power-efficient Maxwell is already close to pushing frequency boundaries when running at ambient temperatures. So then it seems Maxwell doesn’t seem to like much applied voltage at ambient temperatures. Well, how does applied GPU voltage normally work on other GPUs normally? There are two ways:
+===Voltage relation to chip operation===
+As your card clocks go up, it’s harder for the electrical gates in the GPU to switch due to more noise generated and the increased demand for current from the power supply. There is also an overall increase of temperatures inside the chip structures internally on the GPU. At some point, the applied voltage is just not enough to keep the internal gates switching reliably and this creates errors/artifacts. The result is the GPU will crash. Raising voltage will provide that extra power the GPU needs, but it will also skyrocket GPUs temperatures internally. If you are still below “maximum” temperatures for a specific frequency, the GPU will remain stable and continue working well. There are safe typical temperature ranges for different voltages. These are like “sweet spots” with regards to voltage/frequency/temperature.
+===Voltage in relation to power consumption and temperature===
+Increasing voltages and improving the cooling to reach max clocks, this is basic overclocking. Keep in mind increasing voltages always brings an increase in power consumption and internal GPU temperatures. This is not a linear relationship, but more like an exponential one. When you increase voltage 10%, your power usage and temperatures will not increase just 10%, but more like 25-30%. This is very different to frequency gains, which are more linear (10% faster clock gives close to same 10% power increase). Even at extreme cooling temperatures, this rule still applies because -100 °C at die contact surface is still not enough to cool down heavily overvolted chip logic structures buried deep in the silicon. It will still overheat at one point no matter how cold temperature is, if voltage and power consumption is too high. This behavior is important to know, as it is the basis for all overclocking, extreme or not.
+FIXME IMGIMGIMG FIXME
+Maxwell is designed and optimized very well to run at high frequencies. This means that while keeping voltages and temperatures at normal default levels, you can get amazing clocks. Most of the GM200 GPUs on Titan X and 980 Ti cards can run 1450+ MHz, which is already considered very high clocks given the chip complexity and size. There is some headroom left for scaling at ambient temps, but not much.
+Overclocking at frequencies beyond 1450 MHz really depends on quality of silicon and how high it can scale. Another important thing is whether the GPU is taking more or less power at default spec.
+====BIOS & Tools====

User Tools

Site Tools

Differences

Page Tools