[nextpage title=”Introduction”]Last December, we received a rather special motherboard from a brand little known in Europe, Maxsun. The brand sent us their top-of-the-line model, the ICraft Z890 Vertex. A board with an unusual name in the community, since it bears the same name as myself, tester for overclocking.com and extreme overclocker in my spare time. A card whose name leads me to ask the most of it in terms of its capabilities. I’d also like to say that I benefited from the help of the Maxsun dev team. The icing on the cake is that this is a 2-dimm motherboard that is clearly intended for overclocking, even if it lacks a few essential features for very advanced use under liquid nitrogen, for example.
In this article, we’ll be talking about its current capabilities in extreme overclocking under Dry Ice, before a more comprehensive test to follow. This session will take place as part of the G.skill OC World Cup, at 12pm GMT on Wednesday 19/03/2025 the top 9 will be determined before the judges pass over each score.
Thanks to Maxsun for providing us with this sample and for the support given over the last few months.
[nextpage title=”The Maxsun i-Craft Z890 Vertex”]
Preview: some photos of the Maxsun ICraft Z890 Vertex
We repost the unboxing of the board available on the overclocking.com Youtube channel for your information. A great bundle with goodies and a bench table included!
And here are a few images of the card before the full test. Clearly, this is high-end packaging and a high-end bundle for a high-end motherboard. And yet, this board is at the same price point as the competitors’ Elite, TUF and Tomahawk ranges. Note the presence of Slim SAS ports, enabling the addition of 8 extra SATA SSDs or 2 Intel Optane SSDs in U.2 PCIe 4.0. And, on top of that, an OLED screen, a well-stocked IO and two PCIe 5.0 ports with bifurcation. One wonders why Maxsun isn’t in Europe yet, but I know a few who’d be willing to buy this particular model.
[nextpage title=”Preparing the configuration”]
We now move on to preparing the configuration for the dry ice run. We start with the RAM, a G.skill Tridentz 7200CL36 2×24 Gb kit equipped with Sk.Hynix Mdie chips. A little peculiarity here: the chips are not stamped SK Hynix but Gskill, which is unusual and we’re likely to see more and more of them.
Once delid, the kit is placed in nickel-plated copper Iceman heat sinks, which are then fitted with a RAM Frostvolt bucket.
For the motherboard, we use a Heater HOT from Elmor, a foam cut-out to accommodate it will support the motherboard and insulate it from below.
We then place the motherboard on its support, and place insulation around the socket. The insulation is thick, but crushes very well under the weight of the bucket.
All that remains now is to apply the thermal paste and install the Reaktor 4 LN2 pot from Burning OC. Here we use an in-house fastening system based on an M4 threaded rod, 4 springs from Waterblock Heatkiller IV and 3D-printed parts.
With the CPU LN2 Pot now installed, we can move on to the memory part. We install the memory sticks, then put foam around and between them. We’ve also added foam to the M.2 slot.
With the sticks installed, we now move on to the RAM LN2 pot, taking the opportunity to place some foam between the RAM and CPU LN2 pot.
All that’s left to do is finish insulating any exposed parts with foam, towels or absorbent cloths, and you’re ready to embark on an extreme overclocking session.
Now that we’ve reached the end of the assembly process, we’ll just need to warm up the bucket with a few light benchmarks under Windows and check that everything’s running smoothly.
For contact between the bucket and the dry ice, you need to use a liquid capable of withstanding such low temperatures and stay liquid, personally, I’m going for acetone. It’s a noxious liquid, but it performs very well. You’ll need to protect the configuration during the descent and ascent phases, which can cause splashes and stains on components.
[nextpage title=”Under cold”]
We start this session with a quick survey of what’s left in dry ice, around 10 Kg out of the 20 Kg left after the previous day’s session on Raptor Lake. Here’s a photo of the 20 Kg parcel, well underway with less than 4 Kg in the middle of the Y-cruncher session.
So we’ll start the Session on Y-cruncher, which is, along with 7-zip, one of the two benchmarks to be done for this competition under Arrow Lake. Why Arrow Lake? Because the core limit is set at 6, and without hyper-threading, Arrow Lake outperforms Raptor Lake on multicore benchmarks.
The first step is to switch Ycruncher to 6Ghz on the CPU with RAM in ambient. This may seem straightforward at -70°C, but not in the case of a very bad Ultra 7 265K on the core. Impossible to set 1.42V or more, under penalty of severe downclock, and impossible to benchmark at less than 1.425V during initial tests. This lasted well over an hour, enough time to test BIOS parameters and make a few adjustments under Intel XTU. In the end, the bucket was a little cold. Adding acetone dropped it from -55 to -64°C (yes, that’s not much, but strangely enough, on ARL the bucket is warmer, around -71 on RPL). With a voltage of 1.415V and a lower temperature, we manage to get through Y-cruncher without any problem and without any loss of performance, but it’s still borderline and it will be difficult to get much out of this CPU on other sessions.
The memory can now be cooled down, so we don’t think too much about it, taking it down to -70°C in a matter of seconds thanks to the highly responsive aluminum LN2 pot.
First benchmark: Y-cruncher 2.5B at 6GHz
We start with the Benchmark that will have taken us the longest, the most severe for IMC and CPU. If a profile passes here, it goes on 7-zip. We start with ambient memory, 8800c36, then push it to 9000, 9200 and finally 9400C36. With memory at -70°C on a motherboard that’s not up to the level of the Apex, OCF and Tachyon, this is quite a feat all the same. The only downside is the train time: I clocked up to 3 min 15s to train the profile to 9400C36.
Once the RAM is pushed to its maximum, we push a little more D2D, rather limited to 34-35 because of the cold and a VNNAON locked at 0.82V. The NGU is up to 38 and the cache to 51. With such high frequencies, the system is on the verge of stability, but a respectable score of 73.656s is obtained in the end.
Second benchmark: Z-zip at 6GHz
After perhaps 9 hours in the cold on Y-cruncher, pushing to the max and learning the platform (IMC, SA, memory voltage, CPU timings and limits), we now move on to 7zip. A few tests were carried out beforehand to see if performance was acceptable, but it’s time to push a little and validate a score.
On 7-zip, we were able to push the memory frequency to 9400C36, D2D to 34, NGU to 38 and cache to 51. I was also able to run the benchmark at 9494C36 by pushing the SOC’s BCLK to 101 MHz. With this memory profile, the final result is 82135 pts on 7-zip, some OS and Bios tweaks are involved, but we won’t reveal the magic recipe.
Latest benchmark: Max Freq in DDR5, target 11000 Mt/s
Just before this session, I made a post on X, teasing the session and giving me a goal of 11000 Mt/s in memory frequency, let’s see if the goal is reached.
I’m born for this moment??? Can’t wait https://t.co/CXjdAmy0Qo
— Maxsun Official (@MaxsunOfficial) March 17, 2025
So we start with less than 1 kg of dry ice left, if not less. Fortunately, the CPU is down onto 2 cores, and frequencies and voltages have been reduced, so there’s no risk of overheating. We start with a profile that had worked in ambient with 5350 MHz in CL48. Here, it’s impossible to go beyond CL48, so we’ll make do, with ram voltage at 1.45V, SA and IMC at 1.5V.
We therefore manage to train 10800 Mt/s (5400 MHz) in CL48, then on Intel XTU, we increase the BCLK of the SOC to 104 MHz, reaching a frequency of 11200 Mt/s (5600Mhz ), which is the target.
At the end, with the last remaining grams of ice, I manage to reach 11067 Mt/s, but it’s impossible to run Windows. Temperatures rise again, and it’s almost 4 a.m., time to call it a day.
[nextpage title=”End of session and conclusion”]
End of the extreme overclocking session on this Maxsun ICraft Z890 vertex, with the power supply disconnected. Then, once it’s hot enough, we remove the Godet and RAM. The board is dried and then blown to remove the moisture accumulated after 11 hours in the cold.
What can we say about this Maxsun ICraft Z890 vertex after this first cold session? Firstly, this is a motherboard that performs better than we expected, even if there’s still a lot of room for improvement. We regret the fact that there’s no safe mode or retry button, which means we have to use the clear cmos regularly. As for Arrow Lake, it’s a nice surprise that the platform is really interesting to overclock for RAM. This is due above all to the BCLK SOC not linked to the BCLK CPU, which makes it easier to overclock RAM with BCLK. This is also due to its ease of frequency upgrading, so you’ll need to invest in a Mdie kit to make things even better.
All in all, it’s a very good experience, and we’ll be back soon with some liquid nitrogen and more time.