Ti KPE OC Guide
Ti KPE OC Guide
Ti KPE OC Guide
Overclocking/overvoltaging your EVGA 980 Ti KINGPIN graphics card on air, water, and
LN2 temperatures can be fun and most of all pretty safe for the hardware, as long as it’s
done correctly. This includes not only card PCB preparation for extreme conditions, but
having the right software tools + capable supporting hardware(PSU) and a stable platform
to test on. Taking the time and effort to prepare the VGA properly for more extreme forms
of cooling and knowing some of the basics of overclocking, can go a long way to give a
better understanding of exactly what is going on when you overclock your VGA at any
temperature. Consider everything given below as “self-educational” and provided to assist
in getting the most out of your shiny new KP card. Not only EVGA, but I think most
vendors won’t warranty any hardware if it shows signs of being used and abused.
Proceed with caution and at your own risk. Having patience and doing things in small
steps can work wonders in providing the best user experience while keeping your
hardware safe and benching over and over again.
980 Ti KP has three different BIOS modes, which all have different maximum power
targets keeping nominal power spec the same. If you want maximum OC headroom on
air/water, you can just max this setting out and it will be enough.
Next up are the usual controls we use in EVGA Precision for adjusting everything on the
card in windows:
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)
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:
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.
Normal BIOS – This BIOS is stock baseline, +130% Power target limit
OC BIOS – This BIOS similar to Normal but with 7096MHz memory, +150% Power target
limit
LN2 BIOS – This BIOS have similar to OC, but with 170% Power target limit
Custom BIOS
OC BIOS
If you plan to run LN2 temps (below -60°C) should use LN2 switch position (Red LED
BIOS), to disable hardware thermal protection mechanism.
Watercooling compatibility
Bitspower 980 KINGPIN waterblocks are compatible with GTX980Ti KPE as well, with
minor limitation:
Due to different memory chip layout – few memory chips will not be in direct contact with
block. That’s not big issue from performance, as memory is not power hungry device and
does not need direct active cooling. You can add thermal pad between extra chips and
block, so nothing to worry about there.
Clockspeeds, voltages, and temperatures. All these variables work together and must be
dialed in correctly for the best results. The following are the normal overclocking ranges
you can expect for the majority of cards, however every card is not exactly the same.
More fine tuning will always get the best results on your card. Some cards will use
more/less voltage and for sure some cards will go colder than others on top end. As
suggested earlier, overclocking in small steps helps to learn exactly what your particular
GPU needs for every MHz upward.
This guide doesn’t cover insulation and condensation preparation of the card itself, as
there are different ways to do this and everyone has a preferred method. I figured maybe
it’s best to let some guys post their own methods here that work well. This lets you decide
how you want to insulate your card your own individual way. Check around KPC or the
usual internet forums for insulation guides for your graphics cards. Once the 980 Ti KP is
prepped, container is mounted, and card installed into the system, you’re ready. Do a
quick post/boot before putting ANY LN2 into the container just to be sure everything is still
100% functional on the card after you mounted it. Check that the drivers are still
working/installed, check for artifacts on the screen, even run a 3Dmark at default settings.
If the container warms to 30-40 °C under load, just splash it with some LN2, don’t let it
overheat is all you need to worry about. If it can’t pass this, it will never overclock. That
means it’s time to break down the rig, see what the problem is, and start over.
If everything is OK, then let’s move on. Word to the wise, you can NEVER just pull the
card down to min temp (-125 °C), set max voltage(1.65v+) and think it will run alright. This
is NEVER recommended and something I’ve always preached against since the
beginning. Impatience will lead to bad extreme overclocking sessions. The proper way to
overclock a VGA to the max clocks it can run is to go in steps, working your way up in
voltage and down in temperature, like the following. Adjustable on the fly Vdroop changes
can help when tuning in max clocks. Voltages shown are measurements taken at the
probe it reading point for core voltage:
1st boot/check Ambient Default voltage
-50 to -60 °C 1650 MHz 1.55 V,droop disable
-80 to -90 °C 1750 MHz 1.65 V,droop disable
-90 to -115 °C 1850 MHz – 1900 MHz 1.7 V, mild droop
-115 °C and beyond 2000 MHz+ 1.75+ V, med-most droop
Again these are typical LN2 scaling numbers of what most cards can achieve. While all
KP980Ti PCB ‘s are capable to push Maxwell 980 Ti GPU to well beyond the limits, all
GPUs are different. Factors such as ASIC, leakage, and container mounting can have a
big impact on the end result.
Memory overclocking
In modern benchmarks like 3Dmark Firestrike or Heaven, graphics memory overclocking
is just as important as the GPU.
Much thought was given on 980 Ti KINGPIN with regards to memory during the card
design to optimize memory performance to its fullest.
To start, all 980 Ti KPE cards are equipped with SAMSUNG 7Gbps memory chips.
Currently, this is the only 980 Ti card available which features Samsung memory. Why the
buzz about it?
1. Samsung is faster clock-per-clock against Hynix due to some tighter latencies and the
ability to run higher voltage. This improves the performance a bit over that of the
reference 980Ti.
2. Samsung memory can scale nicely with memory voltage. This means that if you
provide more voltage to memory than stock 1.60V, you will very likely get higher
overclocking on memory. Hynix memory fails at this as most IC’s are usually unable to
cope with elevated voltages. 3D apps will crash on cards using Hynix memory even at
stock 7Gbps clock if you raise the memory voltage. Usually over 1.7v is enough to make it
unstable. Because of this, Samsung memory is THE BEST choice for a high end
overclocking card and why 980 Ti KINGPIN is built only with Samsung IC’s. When you are
overclocking on LN2, you want to match the highest core speed + highest memory speed.
This is what gives the best score and Maxwell GPU cores actually need higher memory
voltage to keep the memory controller portion of the GPU from bugging on cold and
limiting max memory OC. Hynix memory can’t scale with voltage, so a lot of Ti’s and
Titan-X GPUs end up NOT being able to hit their max core clock due to this. We saw this
behavior over and over again in the lab on many Titan-X and reference 980ti as well as
some ES experimental 980 Ti KINGPIN cards with Hynix memory.
When pushing your card hard on LN2, always raise the memory voltage about 1.80 V TO
START, and the GPU will require more like around 1.85 V or higher depending on the
GPU to keep the memory clocks high on LN2 and scoring correctly. This is critical for 980
Ti overclocking and getting the best results.
You may find that your card can do ++700 on air even using default voltage. This same
GPU on LN2 will require more memory voltage to keep the same memory overclock
stable when GPU is being pushed 1850+. That’s how it works.
Keep in mind that the cards memory is calibrated when it initializes and the boots
windows. Sometimes when raising memory voltage for example to 1.8v at idle desktop, it
can cause the GPU to lock and artifact. This is normal and happens because the memory
has already been calibrated at lower voltage, long before you got into windows. A simple
reboot after desired voltage is set, and everything will be working 100% fine on next boot
into windows with raised memory voltage. Running very high voltages on your cards
memory for extended periods of time may degrade the IC’s slightly, so be careful not to
overvolt when not necessary. Spend some time finding out exactly what memory voltage
your cards needs to run at different temperatures, so it’s not set too high for the clock and
all should be good. I run 1.82-1.85 V on cards on LN2 temps always. For air cooling,
usually I don’t go over 1.8 V and it seems to be able to handle it so far. The memory can
scale very high on some card using higher than 1.90+ V, but please do so carefully. I have
seen some cards IC’s degrade at voltage over 1.94 V/air and running extremely high
frequencies (750+ offset on memory, = 2250+)
Card overclocked with LN2 and running 1800+MHz can take some serious power, so
ensure your PSU is adequate for the job. We highly recommend running system (CPU,
peripherals, fans, motherboard) from one PSU, and VGA card from separate second
PSU, with at least 1200W of power on +12V rail.
To show what card is capable of, here are measurement results of +12V current
consumption at 8+8+6pin plugs during 3Dmark FireStrike Ultra runs:
Idle power during benchmark loading screen, ~30A taken, near 400W, 1800MHz 1.65V
Running GT1 test, ~71A taken (850W), max peak is 125A (1500W), 1800MHz 1.65V
Running GT1 test with higher clocks&voltage, ~92A (1100W), max peak is 135A
(1625W!), 1950MHz 1.74V
Running GT2 test with same settings as above, ~77A (927W), max peak is 133A
(1603W).
All results were captured with GPU cooled to -100°C, using Tektronix TDS5034B scope
and TCPA303 current probe.
Alternative functions
If you are a first timer to the extreme world, you might not have lot of experience about
what hardware mods to do on your card, or what traces you need to cut to disable
overtemperature/overcurrent protections.
No worries, you can use KPE just as it is for the ALL your LN2 sessions. You only need
voltage control tools, which could be either EVBOT, SW tool such as classified voltage
controller, or even fancy Raspberry Pi. As already mentioned, 980 Ti KINGPIN is
designed so there are no hardware modifications or soldering required to reach maximum
clocks with your GPU and VMEM, just like all other Kingpin edition cards.
However, If you want a little more fun than just max clocks and performance, we have
included a few extra tricks/mods to help you out and get even more from the card with
some easy soldering.
The point of LN2 overclocking on a GPU is to freeze the GPU and nothing else.
Everything else on the card, such as discrete components in display output area, is much
better working at ambient temperatures. And for many years extreme overclockers were
using heatguns to keep that area warm, as there are no parts which dissipate much
power and without external heating front area of PCB get cold really quick. Warm PCB
keep the GPU signaling correct and reduce icing on critical areas.
So is there a solution without external heatgun? It’s possible with 980 Ti KINGPIN cards
to use dedicated design feature that keeps PCB layers always above negative
temperatures. It also keeps the memory IC’s from bugging out due to cold. While
Samsung is the least buggy of GDDR5, it still runs best when it is not frozen solid. This
PCB heating feature is needed only for hard-core LN2 overclocking, so we fully
de-activated it by default in stock condition.
That’s for single-card power ONLY. This doesn’t account for CPU, peripherals, any
high-speed high-power fans, etc.
Integrated monitoring
This feature can be used by connecting any popular TTL UART to USB converter to serial
port on card. Server crowd sure knows about serial consoles, as these used widely to
remotely control rack mount hardware. This is a new thing on gaming VGA cards
however. If you don’t have serial dongle, you can get all-ready-to-go solution here or little
DIY module here . Both are pretty cheap, and likely to be less than 5$ at sites such as
ebay. Pay close attention that you need UART-USB adapter, not RS232-USB (RS232 is
same protocol, but much higher voltage).
Interfacing is done via onboard header located near SLI edge ports.
You have access to this port with both stock fansink operation and watercooling/LN2.
Connect UART cable to VGA’s serial port like this photo below. It could be little tricky to
get cable between fansink cover, so use twizzers to fix cable in position.
USB end of serial cable goes to your PC or smartphone via OTG, or any other device
able to capture serial stream at 115200 8N1 speed.
We use the popular putty , and if everything is connected properly, you should see this :
4. Set RGB data, which will let you configure LED RGB setting, just like Precision X does.
7. Monitors key voltages
8. Monitors GPU and PCB temperatures every second. Both GPU and PCB temperature
can be measured in range from -64°C to +100°C. Works until any key press received
9. Monitors voltages, temperatures, and status 5 times a second, until any key press
received
10. Self-explanatory setting, just like EVBOT, can be used to change GPU voltage.
Entering value 0 will switch back to automatic control, 1-799 will return to menu without
reverting to automatic control.
11-12. Power and heat monitoring
15. Activate or deactivate LN2 XOC heater coils. (works only if enabled by PCB mod)
NVVDD, GPU Voltage – IR3595A digital controller with 14-phases using latest
generation IR3575M PowerIRStage
FBVDD, Memory voltage – IR3570B digital controller with 3-phases using latest
generation IR3575M PowerIRStage
PEXVDD, Digital-controlled IR 10A SupIRBuck
1VPVDD, Digital-controlled IR 10A SupIRBuck
Each phase with IR3575M can deliver ~45 Amps of current with operation temperature
+80 °C. And it’s unlikely to get so hot, as loading is spread across all phases. Thanks to
exposed metal top of IR3575M, thermal heat is dissipated not just thru PCB copper, but
also directly to VRM heatsink plate. It helps to keep VRM cool, and loss less power, so
that improve margin on power target limit used by NVIDIA Boost function.
For voltage monitoring purposes EVGA GEFORCE GTX 980 Ti KINGPIN have 10-pin
2mm pitch PH-type connector near top PCB edge.
Pin definition as below:
This pinout is same since GTX 680 era, so you still can use your ProbeIt setup if you had
it done before. Also helpful labels are located on bottom side of PCB, near each
connector pin to aid connection:
You can use bundled ProbeIt adapter to your usual multimeter probes, or in case you
want to have custom cable connection to your specific meter, you can grab connector
separately, for example here from Digikey.com . You will also need these contacts. If you
want less pins (for example only 4 to monitor GPU and MEM voltage) you can get needed
housings right here .
In addition to monitoring, this time around KPE card provides also extra hardware-based
controls:
V-Tune switches
Top switch near ProbeIt port dedicated to LN2 operation, to eliminate possibility of voltage
shutdown when benching with GPU temperatures below -100 °C. First switch position
disable thermal protection for memory VRM, and second switch position does same for
GPU voltage VRM.
Another switch on bottom lower side, can be used to boost GPU voltage +25 or +50mV
above current setting. It’s different to software or EVBOT override control, as offset will be
applied any time, even if GPU uses stock mechanism to control voltage. It could be useful
if you wanted little bit extra voltage for everyday gaming, without any use of EVBOT or
other voltage override methods. Each switch position at ON will add +25mV, so both will
give +50.
Two switches near bottom corner dedicated for Vdroop fine tune. It’s can be used in
ADDITION to EVBOT’s Vdroop enable/disable control, giving you at least 8 different
Vdroop settings for GPU Voltage and three Vdroop settings for memory voltage.
EVBOT way
For lucky people who still have EVBOT controller, there is P55 EVBOT Firmware for
EVGA GeForce GTX 980 Ti KPE
You can update your BOT with EVBOT Flash tool supporting Kepler & Maxwell GPU
This is it, these settings and ranges are more than enough to max out GM200 GPU, even
on LN2 cooling.
DIY way
Same as 980KPE, there are pads for commonly used 3296W-type trimpots.
To activate this mod, put two shorts in area marked on photo below.
Shorted pads in cyan box enable memory voltage trimmer, while red pads short enable
GPU voltage trimmer.
Adjusting resistance down to 550 ohm will increase memory voltage to ~1.7V.
Software way
FTW12345’s little tool 2.1.2 also working on 980 Ti KPE.
There is one drawback for software, as settings can be adjusted any time, and will keep
but will be reset on power cycle (card power turn off/turn on).
So if your system shutdown, then all voltages and OCP/Vdroop setting will be reset to
default.
That’s one of reasons why other ways, like EVBOT/RPI or hardware mod are prefferable
for hardcore LN2 benching.
Raspberry Pi way
You can use widely available RPI to set voltage via EVBOT port on the card. To do so you
will need connect three wires from Raspberry Pi to VGA’s EVBOT connector.
Pin 1
Pin 3 SCL Pin 4
Pin 5 SDA Pin 6 GND
After connection, check that all pins are correct and try to detect card on RPI.
root@raspberry-pi:~# i2cdetect -y 1
0 1 2 3 4 5 6 7 8 9 a b c d e f
00: -- -- -- -- -- 08 -- 0a -- -- -- -- --
10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
30: -- 31 -- -- 34 -- -- -- -- -- -- -- -- -- -- --
40: -- -- -- -- -- -- -- -- -- 49 -- -- -- -- -- --
50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
60: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
70: -- -- 72 -- -- -- -- --
Correct output is shown on terminal capture above.
Now, you can run download tool here to your Pi and run it to set GPU voltage.
root@raspberry-pi:/home# ./tikpe_vcore
/******** Manual VID setting tool :) **********/
Voltage monitored = 1000 mV
Enter voltage, in millivolts (800...1850): 1000
Voltage monitored = 1000 mV
Enter voltage, in millivolts (800...1850): 1100
Voltage monitored = 1100 mV
Enter voltage, in millivolts (800...1850): 1200
Voltage monitored = 1197 mV
Enter voltage, in millivolts (800...1850): 1300
Voltage monitored = 1293 mV
Enter voltage, in millivolts (800...1850): 1400
Voltage monitored = 1388 mV
Enter voltage, in millivolts (800...1850): 0
Incorrect voltage input!
Here voltages in 100mV steps were entered, and actual readout values for input, output
voltage after each change also reported back.
To connect LEDs, we use same connector type as ProbeIt , but with 4-pins only.
Following simple schematics as shown, connect your custom RGB LEDs with 20-47 Ohm
resistor in series. I had no 5mm RGB LEDs, so used 5050 SMD ones instead.
Now you can use Precision X to adjust LED settings. As example, to set rainbow effect,
cycling all colors, use Demo2 setting:
References