5800 MHz Intel Tiger Lake with Liquid Nitrogen
The world’s fastest Intel Tiger Lake processor runs at 5800 MHz with the help of liquid nitrogen and an 尔英 (Ěr Yīng) motherboard. I also show benchmark results and will share my thoughts on overclocking Tiger Lake.
Let’s jump in!
Table of Contents
Tiger Lake XOC Preparations
In preparation for this video, I finished a SkatterBencher guide demonstrating how to overclock this Core i9-11980HK processor. It is certainly unusual to be able to overclock Tiger Lake because it never made it to the DIY desktop market. Intel famously opted for Rocket Lake instead.
The long story short is that 尔英 (Ěr Yīng) puts these mobile BGA-package CPUs onto desktop-size motherboards. Since BGA processors don’t sit on top of a socket, their Z-height will be much lower than a typical desktop processor, so 尔英 put a unique heat spreader on top of the BGA CPU to maintain compatibility with most desktop coolers.
After I finished the SkatterBencher overclocking guide, I pinged Elmor from ElmorLabs to see if I could spend a day in his office testing the CPU with liquid nitrogen.
Why Overclock Tiger Lake with Liquid Nitrogen?
I’m sure you’re thinking: why would anyone want to overclock Tiger Lake? Isn’t that like a mobile-only and old architecture? Yes, it is. There are two reasons why I wanted to overclock Tiger Lake and push it to the limit with liquid nitrogen.
First, because I never overclocked Tiger Lake. So, it’s an exciting experiment to see how it stacks up against Comet Lake, Rocket Lake, and Alder Lake in terms of performance and overclocking capabilities.
Second, Tiger Lake is made using the 10nm SuperFin process. This is the third generation of Intel’s famously plagued 10nm, following Cannon Lake and Ice Lake, and the predecessor of Alder Lake’s 10nm Enhanced SuperFin which we now have to call Intel 7.
Since Alder Lake, the maximum frequency of Intel’s CPUs has been increasing rapidly, culminating in the world’s first 9 GHz CPU. So, I’m curious if Tiger Lake also has any of that high-frequency potential in it.
What are My Objectives?
My main objective was to have a legitimate claim to the World’s Fastest Intel Tiger Lake processor title. I figured I needed two data points for that: one, obviously, to achieve the highest frequency ever on Tiger Lake, and two, to achieve a number one spot in a benchmark.
For the CPU frequency record, I rely on the CPU-Z database. The highest CPU frequency I could find was 5187.31 MHz by X17 on an Alienware laptop . I could even beat that by using regular water cooling, achieving 5.4 GHz already.
So, my CPU frequency target was two-fold:
- The highest validated CPU-Z frequency, and
- The highest validated XOC CPU-Z frequency.
The difference between both is easily explained. In non-XOC mode, CPU-Z uses a heavy all-core workload before recording the CPU frequency. In XOC mode, it just records the frequency. So, the former will be lower than the latter.
For the benchmark objective, I checked HWBOT to find the most used SKUs and benchmarks with Tiger Lake processors. Obviously, there aren’t many. The Core i7-11800H has the most submissions with 5,100 results, the Core i5-11400H has the second most with about 670 results, and the 11980HK is the third most with about 650 results.
Then I looked for the most used benchmarks, and it turns out that’s XTU 2.0. The highest Tiger Lake score was 5503 by The_Crusher with a Core i7-11800H at 4.6 GHz. Ironically I was pretty confident I could crush this score with liquid nitrogen.
Tiger Lake XOC Setup
Next is setting up the system. There are a couple of things I wanted to get right.
First, I swapped out the thermal paste between the CPU die and the heat spreader with Thermalright TFX Extreme. This paste does pretty well under extreme cooling conditions.
Second, I use the ElmorLabs Volcano LN2 container. That’s the same pot used for the 9 GHz CPU Frequency world record. I figured if it can do that, it’ll suit my Tiger Lake well too.
Third, I use an ElmorLabs HOT300 Heater Controller and HOT300 Heater Backplate. This is a nifty device that heats up the back of the motherboard. This helps prevent the cold of the nitrogen from spreading across the PCB, which could create patches of condensation.
Potential Challenges and Expectations
Regarding potential challenges ahead, there are several things to look out for.
First, Tiger Lake uses FIVR to power the CPU cores. Among the extreme overclocking community, FIVR is known to create a lot of difficulties under liquid nitrogen due to cold bug. To put it simply: FIVR usually stops operating below -100 degrees Celsius.
Second, this motherboard has no auto-rules to assist with extreme overclocking. Boards like the ROG Apex or Aorus Tachyon have a bunch of BIOS auto-rules that help set the proper voltages and settings for pushing the CPU in XOC conditions. It might be difficult to diagnose or work around specific issues without the auto-rules.
Third, while we exchanged the thermal paste between the CPU die and heat spreader, it’s still a sub-par thermal transfer mechanism. So, we can expect some difficulties in high-load scenarios.
Lastly, there’s very little information available on how to overclock Tiger Lake. While I have some experience with other Intel architectures, there may be Tiger Lake-specific quirks or oddities that we don’t know. We’re going into uncharted territory.
Tiger Lake XOC Results
Let’s get on to the results.
As I showed at the beginning of the video, the highest frequency I achieved with Tiger Lake was 5816.91 MHz with 1.45V. The LN2 container temperature was -50 degrees Celsius. The CPU was pretty unstable at this point.
Strangely enough, it was impossible to run the 56X ratio, and I had to resort to BCLK overclocking to achieve the result. As you can see from the CPU-Z validation page, the result is marked as “unchecked” because it was achieved with XOC mode enabled.
The highest stable frequency I achieved with Tiger Lake was 5486.58 MHz with 1.43V. CPU-Z marked this result as “validated” because it ran in non-XOC mode. That means CPU-Z ran a heavy workload before recording the frequency.
The highest XTU 2.0 benchmark score achieved with Tiger Lake was 6301. The processor was clocked at 5486 MHz with 1.43V. The memory was clocked at DDR4-4000. While the LN2 container was at -70 degrees Celsius, the Core temperature spiked to +19 degrees Celsius under load. This score was sufficient for P1 across all Tiger Lake submissions at HWBOT.
I also scored 17390 in Cinebench R23 Multi-Core and 1638 in Cinebench R23 Single Core with those same settings.
In terms of operating temperature, Tiger Lake did surprisingly well. I could keep the CPU running at -100 degrees Celsius, though I ran it at -50 degrees Celsius most of the time.
Discussion & What’s Next
The good news is that we achieved everything we set out to achieve. But could we do any better? I think so, yes.
The first and most crucial bottleneck for achieving better overclocking results is what I believe to be a PLL issue. Think about it. We can run everything stable at 55X but cannot even set 56X. If that sounds like a familiar problem, you’re right. It’s pretty much what we saw on Sandy Bridge!
The solution introduced on Sandy Bridge was an option called PLL Override. By increasing the PLL voltage, you’re able to get higher frequencies. Increasing the PLL voltage is standard practice and an XOC auto-rule on many enthusiast motherboards.
PLL overvoltage is also available on Tiger Lake and in this motherboard’s BIOS, but there’s a catch. For PLL overvoltage (PLL_SFR_OC) to succeed, the PLL input voltage (VCCSFR_OC) must be at least 150 mV higher than the target PLL voltage. This input voltage is an external package pin, and unless it is specifically implemented by the motherboard engineers, is not available. However, on some motherboard designs, the VCCSFR_OC voltage is connected to the VddQ power rail. Unfortunately, it didn’t seem that increasing VddQ helped increase the overclocking headroom.
Aside from the PLL issue, it also seems that the voltage scaling isn’t that great. Increasing the voltage beyond 1.5V didn’t yield much higher frequencies.
In terms of temperature scaling, there’s some improvement. Going from +40 to -50 degrees Celsius gives us about +600 MHz at 1.4V. That’s not too shabby. That’s mostly at idle, however. Under load, it’s quite a different story. Whether it’s the transistor density or the poor thermal transfer due to the heat spreader, the thermals spike up pretty quickly whenever we throw a heavy all-core load at the CPU. One way to address this would be to remove the heat spreader and run direct die. However, that’s risky, and will have to wait for a future video.
Last but not least, I feared the FIVR would induce a terrible coldbug that didn’t seem to be the case. I could quite easily run below -100 degrees Celsius.
Conclusive Thoughts
Overall, this was a pretty cool project to undertake. Between the novelty of overclocking the Intel Tiger Lake architecture, using a non-Taiwanese, non-enthusiast motherboard, and seeing the scaling of Intel’s 10nm SuperFin process, I had a lot of fun. As I said, there are still ways we could potentially improve the benchmark results and maybe push for higher frequencies, but that will be for another time.
I want to say a big thanks to ElmorLabs for the hospitality. Of course, I also want to thank you for watching and the Patreon supporters for supporting my work. If you have any questions or comments, please drop them in the comment section below.
See you next time!
