• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Help Choosing Between 2 Different CPU's for Mini Itx build

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
I guess I should post it before. I pushed inside the Deepcool Assassin IV cooler, but in the end, I went back to Noctua NH-U12A.
This is still a photo before some changes. Now there is 7800X3D, 64GB Kingston DDR5-6000, Corsair SF750 PSU, ASUS B650E-I Gaming, and RTX4070 FE.
As long as you use SFX PSU, then cables can be hidden behind the PSU. If you use ATX PSU, then there is not much space on the back.
If you feel that everything is too hot, then the top, bottom, and back support 120mm fans, but it is better to use slim fans (and normal/25mm thick won't fit with some other devices). Since the graphics card sucks the air from the bottom and it has that "pass-through" design, then I'm not using fans there. I still installed 2x 120mm slim Noctua fans on the back and top just to force the airflow (not in the photo).

1697201768148.jpg

So, any thought on this mess I am about to start putting together? Would the 13000k be a better performer than the 13900hx?
I've read that the 13900hx outperformed the 13700k in multicore geekbench, but that was a one off comparison.

I found this YT video, which shows that the 13900HX is sometimes better than the 13700K and sometimes worse -
It's pretty much the only video that shows some results on the AR900i. All others are like unpacking and "this is great", but no real info.

The AR900i is now available (there was a problem with availability in the EU and delayed shipping everywhere else when this thread was started), but I still have some mixed feelings about it. It's great for the price but still requires SODIMM, and it's limited to the included cooler as the mounting is non-standard. I also haven't seen any BIOS updates for these mobos (the AMD version is available for a couple of months). I wanted to grab the AMD version, but I really wish for something more than 8 cores, and the 16-core option is not available. I may decide on the AR900i for a bit weird project, we will see.
 
Last edited:
I wasn't sure about it, and I actually had some points against buying one of those motherboards, but I ordered AR900i. It can be a stupid idea, but we will see. I may ask for some SODIMMs for reviews once I get it and check if it works well. It seems like there is barely any laptop or PC that supports more than DDR5-5600.
I have no idea when it will arrive as there is no confirmation. The only info is that it will be shipped from Hong Kong at the end of December. To solve all the problems with customs and taxes, they send it to Germany (at least for the EU), and from there, they send it to other countries in the EU without additional costs, duties, taxes, etc.
I'm not happy about the ~150W TDP and spending money on another toy, but it will replace the Ryzen 7600, which I can sell to cover some costs. For a "mobile" PC, I would still pick something from AMD because of the lower wattage. I'm not sure why the 16-core AMD mobo is not available, and if I'm right, it's not in the plans for the next weeks, even though it's listed. The PC from post #21 is my gaming 24/7 PC (it has actually run 24/7 since I built it), and it will stay like that for a while ... probably at least until the next AMD gen so it may take some time.
 
Well, the build is done.
The 13600k is a nice choice for this build I think.
The editing goes smoothly with the Arc770, it's quiet enough.

Scores about average on Cinebench, and temps wont go above 80 in a tiny case.
Got two portable monitors, an 18 inch 16x10 and 16 inch 16x9 to pair with it.
Whole kit weights about 18 pounds.
This is about 8 pounds heavier than the laptop I as considering but much more versatile.

I did mess with some overclocking for a few min, of course, and got what I thought was stable at 5.5GHZ all P cores.
Passed Cinebench just fine with a ridiculous 27k score. Then promptly crashed during a Davinci Resolve benchmark.
I could have added a bit more vcore but temps were 94. I'll mess with that later on.
This is why we tell folks that Cinebench is NO STABILITY TEST.

Thanks guys for your help, next is my desktop build, which will be a 14900k build.

Attached a obligatory pic.....
Post magically merged:

Woomack, that is sweet and clean. I need to do some cable management, but that case fought me EVERY step.
One thing out of place and all has to come out again.
I am very curious about your ar900i adventure.
Please post updates on that.
If it looks good, I might use these innards on my desktop and follow what you do.
 

Attachments

  • 20231223_051911.jpg
    20231223_051911.jpg
    1.2 MB · Views: 2
Last edited:
It's weird that this case has not been available in EU Amazon (PL, DE, UK, etc.) since I bought it. I saw some pictures of upcoming Deepcool cases, and I think they will release something similar under their brand. This is also funny, as I saw the same case under multiple "no-name" Chinese brands first. On the other hand, cases like Lian-Li O11 were first on Aliexpress under some Chinese no-name brands, too, but barely anyone noticed that. Now people are already bored of that design as nearly every brand has something similar.

If your desktop build is not needed anytime soon, then maybe wait for the next-gen Intel. It should be in a couple of months. I just say that as I'm not really impressed with what Intel offers regarding performance, wattage, temps, and motherboards. There will also be a new socket and probably better support for additional devices.
Whether it's for gaming or work, I would pick AMD right now or just wait for what Intel will bring in the next-gen. Lower chips still seem fine, but i7/i9 are like pushed to the limits and only packed with e-cores to beat AMD in synthetic tests. These mobile chips seem much better balanced, as maybe they have many e-cores too, but the performance per watt is much better than on desktop chips.
 
Last edited:
:attn: Nice little build there!!

This is why we tell folks that Cinebench is NO STABILITY TEST.
Perhaps it's just ignorance on my part, but these two apps do two different things, right? Cinebench is a good test for heavy rendering-type loads, but photo/audio/video editing works differently. For you, maybe loop the UL Procyon Video/Photo editing tests?

No single test is The Gospel, anyway. P95, OCCT, etc. We've all stress tested for (number hour hours here) amd been unstable somewhere else. In the end, you need to do what you do with your PC to ensure its fully stable for your uses.
 
It seems lately that everyone you see is overclocking and using cinebench r23 to prove they are stable. But after just 5 min of P95 they would most certainly crash. I think that cinebench is a good benchmark, but useless for stability testing.
But you are correct, if it's stable for its intended purpose, its good I reckon.

I remember the days of running P95 for 24 hours to prove stability, which I look back on and realize the overkill in that LOL.
I wonder if anyone ever ran a completely stock system like that, and I wonder how often there would be errors out of the box? hmmm...
 
But after just 5 min of P95 they would most certainly crash.
Perhaps. Depends on the test they use. I'd gather small fft would pass though. Cinebench is def. not useless as a stress test...loop it. Running a 30s benchmark isn't stress testing on any level, but especially if its not focusing on your specific tasks.

Have you tried CB24? That's even more painful than 23, lol. It even has a duration option for stress testing.

I dont remember those days because I wasn't one of the people pushing that 24 hour nonsense, lol. A few to several hours is plenty for most* home users. My weapon of choice is AIDA64's stress test (CPU, FPU, cache, and memory checked off). That does great for me and my uses.

I'm sure plenty tested at stock (it's how you know your stuff works out of the box, lol), and perhaps some errored out... I wouldnt think it's a problem, however. :)
 
if you do small fft, the size needs to be set to fill the size of the L2. im still sad the old posts got purged, i thought someone did the test after i mentioned that. the got it on the verge of unstable yet it was the data set size in prime95, whether it was stable or not.
 
Cinebench is a pretty good stability test, but it's more like all at once. It crashes quickly if there is any problem, but you have to run it long enough to keep the CPU and everything else stressed enough. The typical "throttling" test is 10 mins. It's not long enough for stability, but long enough to see the peak temps on most CPUs. On the other hand, as it was already mentioned, there are tests that will run for 3h+ without errors. Next, you start a game, and it will crash in 5 mins. If you know what you test and how, then usually, 1h is enough, but it is better to use multiple tests.

Some CPUs are not acting as expected in some tests. For example, try to run AMD X3D in AIDA64 CPU (no AVX) and FPU (AVX) tests. In theory, the AVX test is more demanding. On these CPUs, the CPU test heats up more and crashes faster. It's because of how the CPU is designed and that in the AVX test, voltages and clocks go down, and you can't do much about it on most motherboards.
 
I am very curious about your ar900i adventure.
Please post updates on that.
If it looks good, I might use these innards on my desktop and follow what you do.
They will ship the motherboard in about 3 weeks from Hong Kong to Germany. Then repacking and will be shipped to me.
The same to other countries. I thought they will ship it at the end of Dec, as there was info about it on their website.
It’s really better to buy their products from Amazon. I ordered EM680 mini pc as there was a huge price drop for Christmas, and arrived in 3 days.
 
Hey everyone,

I really want some opinions. There isn't allot of info out there on the performance comparison between the 13900hx and a desktop CPU such as 13700k.
I am replacing my laptop that I use for video editing and content creation, and I have a fundamental choice to make.
I can either go with a traditional mini ITX board and get the 13700k for the exact same price as getting the Minisforum AR900i.
For those not familiar with that the link is here:

The cost for both choices is the same. I am probably going with a 8L case like this one:

(because of the portability, remember I am replacing a laptop that travels 5 times a week. Any better suggestions are welcome)

So the two things I am weighing are heat and performance.
I will not be gaming much, if at all, but simply editing 4k video in Davinci Resolve and After Effects.
I will need to fit a reference length GPU.

The MB with the 13900hx will obviously be easier to cool, and it has just about all of the features one would ever need.
And it has a two year warranty. But of course, there is no separating the CPU in case of MB failure. (which in all honestly, is rather rare).

But the 13700k with an Asus itx board is also an option. The CPU fan would most likely be limited to a low profile option.

So, any thought on this mess I am about to start putting together? Would the 13000k be a better performer than the 13900hx?
I've read that the 13900hx outperformed the 13700k in multicore geekbench, but that was a one off comparison.

And BTW, nice to see everyone after an extended hiatus ;)
That AR900i and the case would be my choice.
 
Last edited:
@Stilletto Just to let you know how Miniforums orders work. I still have no confirmation about shipping, and I wait for a month. At this speed, when I get the mobo, then the next generation will be released :p I'm joking, but the AMD 16-core version is expected to be shipped in mid-February, so in reality, it can be mid-March, and the 8000 series APUs will be released sometime then (at least wider availability of computers). At least in the case of AMD, there are barely any changes, and the early tests say that the performance is similar, too.

I'm just saying that picking the 13600K wasn't a bad choice when you wanted to build that PC in Dec.. If I knew it would take so much time, then I probably wouldn't have ordered it in the first place.

I like weird hardware, so I wanted to build a PC with their mobo, but I was thinking a couple of times about canceling the order. Really, if anyone wishes their products, then better order them from Amazon as then it will show the real shipping time. Their products from Amazon are shipped in 2-3 days (when it says available).
 
Woomack,

That's what I was afraid of. However, I still with you and have an interest in unique hardware that fits unique needs. I am still interested, even with the delays.
But for this itx build, I couldn't be happier (unless it was a wee bit smaller).
 
The AR900i mobo is supposed to arrive today (at least, this is what the tracking says).

The AMD version seems better for SFF because of the lower TDP, and it supports 2x M.2 PCIE 5.0 SSD. The 16-core CPU is available right now (it wasn't available about a month ago). Shipping in mid-Feb. I mean 16 regular cores, not a C+c mix, so it can be better than the 8P+16E cores that the Intel version has.

There is no CPU from the AMD 8k series that is as fast as the Ryzen 9 7945HX. I mean, so far, there is no 16-core 8k APU listed, and 8k IPC is about the same (differences in direct comparison are barely visible at the same number of cores/threads). If you are interested, then you can browse some more as I can be wrong in something (most are early news or leaks, including the SKU table), but I guess that 8k is only a refresh with a bit different options, like higher frequency or different graphics (still, 780M is the fastest for 8k APUs).
I'm just saying that the CPU is 7k series, but it will be mixed with 8k in the upcoming laptops as there is nothing better for Zen4.

If Intel mobo will be good, then I will consider AMD, too. Both seem not much different. Max clock and other things are about the same, a bit more or less threads, 4x M.2 PCIe 4.0 for Intel, 2x M.2 PCIE 5.0 for AMD, both have PCIe 5.0 x16 slot, both have the same size and general design. AMD is a bit cheaper.

Both versions have CPUs with unlocked multi, so they can be overclocked. I have no idea if Intel XTU or AMD Master supports them. It should work, but officially, they're not listed in the software specs (it's old, so it can be wrong).

Edit:
The motherboard arrived ... and as usual, with my luck ... it's DOA, or it doesn't support my RAM. I have 2x16GB G.Skill 5600 CL40-40-40 1.1V. I know it doesn't support XMP, and this RAM has only an SPD profile, so it should work.

Edit2:
I made it work. I have no idea what the problem was, but after I cleared CMOS maybe 15 times, it started to work. Another thing is that I wasted maybe 3h trying to figure out what was wrong.

Edit3:
The manufacturer said that this motherboard doesn't support XMP or manual memory settings. It supports XMP and can adjust main timings, voltages, and frequency.
 
Last edited:
Once I figured out everything with clearing CMOS and BIOS settings, then the AR900i doesn't seem so bad. I still have to check many things, and I am stuck at testing RAM ;)

I reached 6600 CL38-40-40 on my G.Skill 5600 SODIMM kit. Looks like the motherboard supports higher voltages, but only if they're in the XMP. The used kit has RENESAS PMIC, and no matter how I set voltages manually, it starts at 1.10V. Clearly, the motherboard is not any higher gaming series and doesn't support various PMICs. I'm still surprised it works at 6600 and 1.10V.
The only SODIMM with XMP and 1.35V that I found is Kingston Fury Impact 6000/6400. However, I'm not sure if 1.35V works when I enable custom memory settings mode. When the XMP mode is enabled, then it is impossible to change timings, frequency, or voltages.

I will try to run stability tests and maybe post some screenshots.
 
@Stilletto

MF AR900i / 13900HX / G.Skill 2x16GB 5600 CL40-40-40 @ 6400 38-39-39 1.10V

This CPU runs at much lower frequencies than expected when power plans are power-saving or balanced. You can see that the single-core result in Cinebench R24 is 105 when a balanced power plan is in use - with a minimum and maximum processor state set to 100%. When the high-performance plan is in use, then the score is 127. Multi-core is about 1480 vs 1540 average.
The CPU goes up to 5.4GHz only when a high-performance power plan is in use. I have no idea why balanced with 100% CPU state can't reach 5.4GHz. It reaches 5.2GHz. CPU-Z also shows a CPU ratio range up to x52, but you can see that hwinfo64 reports ~5.4GHz max.

RAM is not passing stability tests at 6600, but I could set 6400 CL38, and it passed 32 thread HCI memtest. It's still more than I expected at 1.1V.

The CPU supports overclocking. However, the motherboard doesn't support CPU ratio or bclk adjustments. Intel XTU software says that overclocking is locked even though it also suggests that this CPU should offer manual overclocking - the max turbo ratio should be adjustable. Some older mobile HX CPUs are on the XTU compatibility list, so I think that Intel forgot to add the latest generation.

The motherboard with power limits unlocked to 250W, works without issues. There is no coil whine during high CPU load or any other issues. The CPU goes up to ~155W (on the screenshot is ~152W). At this wattage, it's at the edge of throttling - 94°C with a Noctua 120x15mm fan. This is the max that will fit most small ITX/sandwich-size cases. A 25mm thick fan would be a bit better - about 2-3°C less. During mixed-load tests, the CPU is barely passing 80°C while the fan is at ~700-800 RPM. In a closed case, I can say it can be nearly silent with the tested Noctua fans.

Command Rate 1N doesn't work. When I try to enable it, it runs at 2N anyway.

It takes about 3-4 minutes for the motherboard to boot up after clearing CMOS.
It takes about 3-4 minutes for memory training after each RAM settings change.
There are no LEDs on the motherboard, not to mention a diagnostic display. Not even any power LED. Power/storage LEDs work only on a PC case, when connected to the front panel header. If not, then you know that the motherboard works when any of the fans is spinning.


mf2.jpg
 
Yesterday, the Crucial 2x48GB 5600 SODIMM kit arrived, so there is a little update. This is a 5600 CL46-45-45 1.10V kit. It also doesn't work at higher voltages - Richtek PMIC, but it looks like this mobo doesn't support anything without XMP. Maybe one kit with XMP will arrive in a couple of days, but it's not confirmed.
Either way, 6200 CL42-42-42 1.10V works fine. 6400 doesn't even post. 6200 performs not much worse, even though frequency and timings are worse than the G.Skill kit from the last post. Crucial is a dual-rank kit, so it should be faster in some applications.
Btw. the motherboard sees the SPD profile wrong. It sets Crucial memory at 5200 CL42-42-42 1.10V, so to make it work as specified, we have to set frequency and main timings manually.
Btw2. memory training takes ages ... literally a couple of minutes. The 96GB kit takes even more time.

mf3.jpg
 
wow, i thought memory training was only on amd or is that something different?
 
Memory training is on everything. It just takes a couple of seconds on older motherboards and most Intel series. On AM5, it's weirdly long, and it's normal. Also, higher capacity causes it to take more time. 128GB on Gigabyte B650E takes a good 3 minutes for the first training after clearing CMOS.
If the motherboard manufacturer fails something, then it takes longer, too. This Miniforums motherboard acts like some laptops - ~5 minutes of training of new RAM, so you sometimes wonder if the motherboard even works (it doesn't have any LEDs, so you don't know at what stage it is).
 
yea some kind of light or led read out would be nice for that. first time boot up with new ram taking up to 5minutes blows my mind right now. i would prolly lose my S if i paid money and i didn't boot. that would be like putting gas in your car then having to wait 5minutes for it to test the gas to figure out the octane level. you know how many people would lose their mind! lol
 
Back