Allright, here's my quick review of the b660 Deskmini.
I'll make comparisons to the x300 at each step.
1) Appearance, Accessories and I/O
Appearance and packaged accessories are identical to the x300. Identical 120w Brick. Front I/O is identical. The difference is the rear I/O, offering 2 more USB-A and 1 more USB-C Gen 3.2. Internally it has the same 2x Sata and 2x M.2, but supports 4th and 5th gen Pci-E.
Winner: b660 hugely due to much better I/O and 4th/5th gen Pcie
2) Internal Board Layout and VRM
No it did not come with that copper bit...
The board layout is funky because it has to accommodate the b660 chipset. The RAM and M.2 slot locations are swapped compared to the x300. The other differences are the fan headers are not next to each other, they are on opposite corners of the board ( this is a good thing - so you can use the best located fan header). You'll also notice that funky battery sticking out.
Asrock did not provide any chipset heatsink and it gets quite hot. It could use something. But you cannot use a large chipset cooler because it will be blocked by the edge of the m.2 card on one side and by the CPU socket keep-out zone on the other!
I used a small Gpu RAM sink made by Enzotech here. This is sub optimal but forgivable. With a little creativity you could solve it. It will be getting active cooling in this area anyway.
This is all I could do here
Now the VRM area... it's TINY and the heatsink is about 1/4th the size of the one on the x300 (which wasn't that large to begin with). Possibly because there was less board real estate, and also because this is a b660 board.
However, Asrock advertises compatibility with 12th gen processors up to the 12900
non-K. This is sort of misleading, because there is no way this board will be able to deliver the potential performance of most of the 12th gen lineup with that wimpy VRM (in fact, as you'll find out soon, it cannot even deliver enough power for the 12300). Whereas the x300 - it can actually deliver 100%+ of the potential performance of the 5700g for a brief period (until it thermal throttles)
Winner: x300, due to no cramming of the board for a Chipset, and having larger VRM and VRM heatsink
3) CPU & Memory Performance
Processor: Intel 12600 Non-K
Memory: 2x 8GB DD4 - Corsair CMSX32GX4M4X3800C18 @ 4000mhz CL-17-17-17-34 1T, 360 tRfc cycles (~180ns tRfc)
Cooling: Intel Stock
GPU: igp, UHD 770
The Corsair memory used is identical to the x300, it's from the same quad-channel kit, with identical main timings + tRfc (except for some minor sub timings). However memory latency was coming in much higher on Aida64 latency testing. About 65-69ns on the b660 vs as low as 51-55ns on Ryzen 5700g. 69ms memory latency is comparable to what the 5700g has @ 3200mhz. I am unable to explain why, but it is not a good result. Read and Write performance however was great @ 60GB/s.
Cinebench r23 Single Core
The 12600 scored approximately 1800 in CBr23 Single core
, which is ~20% more than my 5700g. This is an excellent score, however I suspect it is less than what the typical score should be by a noticeable amount. The 12600 scores I was seeing on google and Hwbot with different m-Itx / ATX b660 boards is closer to 1900
. The i5-12400 is supposed to score =~1700, i5-12500 =~1800, and i5-12600 =~1900.
A little penalty for for being on this tiny form factor might be expected, but this seems like a larger discrepancy indicating poor optimization. Especially because the x300 doesn't suffer nearly any "form factor tax".
For proof of this you can compare my x300 CBr23 single core result in this list against the others with similar clock speed
and see that people on the m-Itx / Atx b550 boards didn't have any better frequency scaling than the x300 did.
Cinebench r23 Multi Core
This is, unfortunately, where things start to fall apart for the DeskMini b660. The b660 has a hard limit of 75w power for the entire CPU socket
(and, less than 75w if the iGpu is doing anything by the way).
This isn't that horrible by itself, as it is somewhat close to the limit of what the best thermal solution can cool for an extended period (~85 watts is really the most you want to put into a DeskMini for a sustained period, IMO).
What really hurts is that the board will not allow more than 75w power draw even for very short durations.
The 12600, kneecapped to a 75w limit, only can manage about 10K in Cinebench multi
, so only about 71% of it's expected score (14,000). Average clock speed was approximately 4.1ghz (& without Power Limit would be 4.8ghz). Thus, any chip higher than a 12300 will be wasted, as it can't even turbo past this limit for a few short moments.
The efficiency of the turbo algorithm trying to behave itself within this 75w limit is just terrible. To test this, I normalized the 5700g to the same core and thread count, and normalized the clock speed to 4000mhz to yield the same ~10k Cbr23 score. Same Cores, same Threads, same performance
. The 5700g only drew 45 watts while the 12600 drew it's cap of 75 watts (66% more power draw) for the same performance.
Is Alder Lake really 66% less efficient for the same performance? I don't know
. But you won't want to run a high spec 12th gen processor on this thing and hope it will scale down gracefully. It will not. The way that it is scales down the voltage / clocks to stay within the 75w TDP limit is sub optimal. The idea of someone putting a 12900 in this thing is ludicrous.
3) iGPU Performance
The 12600 has an Intel 770 UHD igp which offers about ~60% of the performance of the 5700g. I did not bother running extended benchmarks but just loaded Heaven and Witcher 3 for a few moments to verify this. Considering that that Ryzen APU's already struggle to deliver 1080p performance in most modern games, the Intel igp is basically obsolete for all modern games as it won't even be able to deliver acceptable 720p performance. Running 4000mhz CL17 memory didn't seem to help much.
4) dGPU Performance with 4th Gen Pci-e m.2 Riser
Processor: Intel 12600 Non-K
Memory: 2x 8GB DD4 - Corsair CMSX32GX4M4X3800C18 @ 4000mhz CL-17-17-17-34 1T 360 tRfc (~170ns)
Cooling: Intel Stock
GPU: Evga 3060ti XC
Riser: ADT R43UH-4.0 Riser
Pretty much the reason why I wanted to test the b660 Deskmini in the first place is because it has 4th gen Pci-e, which in theory can unlock another 5-10% performance with a dGpu (depending on the gpu used) compared to 3rd gen. 3rd Gen M.2 Riser is equivalent to "Gen 1.5 x16
" and 4th gen M.2 is thus equivalent to "Gen 2.0 x 16
" or Gen 3.0 x8 -
so approaching only a tiny loss compared to Gen 4.0 x16 with the 3060ti.
Unfortunately despite having 20% better single core performance than my 5700g, and being on a 4th gen Pci-e riser, the b660 failed to produce any performance gains over the x300 with the Gen3 riser (it was within the margin of error).
That might be normal.
The 3060ti is not really bandwidth limited in the first place. It is unknown if the b660 system would produce better results at 4K, or how much more it would improve with a more bandwidth hungry card such as a 3080, which will increase the significance of the Gen3 vs Gen4 difference.
Note- Average GPU clock speed is actually the same within ~10mhz despite what is indicated.
There is potential here as a budget system, or file server.
The Deskmini b660 is potentially a nice little system for low budget general use and can take advantage of 12th Gen intel processors with excellent single threaded performance such as the 12100 ($129 for ~1650 cbr23 Single core perf -- it's higher than a 5600x overclocked). With the Gen4 storage it could be a nice platform for a file server or something like this. So the use case is ultra low end, with a 12100.
However at the time of this writing it costs about $60 USD more than the x300 Deskmini ($299 vs $240). So any savings on a cheaper processor are going to be a wash, since you could just get the cheaper x300 with the more expensive 5600g ($178). So it doesn't present a great value.
The x300 is better value overall, right now.
The x300 is also much much better for igp gaming, multi threaded workloads, and better supports modding, due to the ability to undervolt for better efficiency or overclock the iGPU for up to ~35% higher gaming performance over stock. And also it has a nice VRM - which can output as much as 140 watts, compared to the b660's crippled 75 watt limit. So you can unlock even more performance from the x300 for free
with a larger cooler.
If going the dGpu route, both are about even. Slight edge to the b660 as it should scale better, and possibly improve as the microcode improves further.