EXPIRED fanless Intel® i3 Kaby Lake mini PC for $74.83

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
I will say the build quality is pretty darn good and thanks to the enormous heatsink, it's pretty hefty too.
it is - I actually ran some CPU load/Hysterisis testing to see how good it is when it came to wicking away excess heat - even when loaded up to a CPU load of 12 (in Debian) and with no airflow the hottest I’ve ever seen it was 67C under 2 hours of testing, which made it warm to the touch but not unpleasant. The t640 would’ve been at 80C and above by now with a heat sink that would have been at least 55c on the surface. Heat recovery from that was also very fast…which is something not surprising since they are expected to be hidden inside booths on show floors.

Also, that freaking thing is like 2300 USD retail.

EB13F706-6245-4028-A1A8-D2C88C212F1A.jpeg
 
Last edited:
  • Like
Reactions: Fritz

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
What....quicksync issues on Win10? Maybe I lucked out on getting one after all.
whoops. Had to run some later testing - it looks like QSV isn’t that fast with the Kaby Lake i3u when tasked with Down-converting 4K/2160p30 content (well, it is a cheap Ultrabook SoC in a passive enclosure).

Here’s some idea of its performance - do keep in mind that Kaby Lake QSV can only do up to 2160p60 4K transcoding For both h264 and h265, so this is more of a worst case scenario for it:

Source: Jellyfish-400mbps-4K-UHD-HEVC-10
bit, (HEVC Part 2, 3840x2160 @29.997 fps)
Destination: 1080p30
H264 x264 default, CPU based: 3.2 to 3.7 fps
H264 QSV 10 Bit: 3.6 to 6.7 fps
H264 QSV: 7.5 to 10 fps

That being said, I am seeing decent QSV figures going from 1080p30 H264/H265 to 720p30 H264/H265 (45fps on QSV instead of 7-11 fps on x264/265), so maybe it’s just a matter of the hardware simply being rate limited transcoding in the higher resolutions. I’ll need to grab some 2560x1600 h264/h265 videos to see what it does…
 
Last edited:
  • Like
Reactions: zack$

zack$

Active Member
Aug 16, 2018
606
242
43
whoops. Had to run some later testing - it looks like QSV isn’t that fast with the Kaby Lake i3u when tasked with Down-converting 4K/2160p30 content (well, it is a cheap Ultrabook SoC in a passive enclosure).

Here’s some idea of its performance - do keep in mind that Kaby Lake QSV can only do up to 2160p60 4K transcoding For both h264 and h265, so this is more of a worst case scenario for it:

Source: Jellyfish-400mbps-4K-UHD-HEVC-10
bit, (HEVC Part 2, 3840x2160 @29.997 fps)
Destination: 1080p30
H264 x264 default, CPU based: 3.2 to 3.7 fps
H264 QSV 10 Bit: 3.6 to 6.7 fps
H264 QSV: 7.5 to 10 fps

That being said, I am seeing decent QSV figures going from 1080p30 H264/H265 to 720p30 H264/H265 (45fps on QSV instead of 7-11 fps on x264/265), so maybe it’s just a matter of the hardware simply being rate limited transcoding in the higher resolutions. I’ll need to grab some 2560x1600 h264/h265 videos to see what it does…
Thanks for this.
 

thetoad

Active Member
Feb 10, 2021
153
64
28
Well, that's 1080p30, and even with 2560x1600p30 H264 clips the transcoding rate off the QSV is around 45 fps, which is useful enough.
UHD remux would be 3840x2160 in hevc, just capped at a bitrate of around 100mbps (not 400 like your test). I have one anime movie that peaks over 100 multiple times, most UHDs are much lower
 

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
UHD remux would be 3840x2160 in hevc, just capped at a bitrate of around 100mbps (not 400 like your test). I have one anime movie that peaks over 100 multiple times, most UHDs are much lower
Well, more testing is needed, then. The lowest bitrate UHD HEVC on Jellyfish Bitrate Test Files (the archive that I use) is at 120Mbps, and I am not sure if the Kaby Lake QSV is limited by resolution or bitrate.
 

thetoad

Active Member
Feb 10, 2021
153
64
28
I can't believe they can't accelerate in spec UHD Blu ray playback, but who knows. It would seem like a big oversight. Though uhd spec also supports 60fps (see Gemini man, or don't, it wasn't particularly good)
 

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
I can't believe they can't accelerate in spec UHD Blu ray playback, but who knows. It would seem like a big oversight. Though uhd spec also supports 60fps (see Gemini man, or don't, it wasn't particularly good)
Eh, playback is supported (in fact, the machine shows a CPU utilization of around 7-9% playing back that 400mbps UHD/HEVC video). I am talking about Quicksync transcoding (aka going from UHD HEVC to h264 720p content). Transcoding is both decoding and re-encoding at the same time, and that can be highly dependent on the format / bandwidth used, driver support and specific ASIC versions. The Kaby Lake GT2/QSV transcoder “should” support up to 2160p60 videos, but at what rate is the important question. In a perfect world it’s in multiples of the video playback, but for a passively cooled low voltage i3u, it might be hampered by thermal throttling or whatnot.
 

thetoad

Active Member
Feb 10, 2021
153
64
28
ok, that's weird. I'd assume that if it supports a transcoder "input" of X and a transcoder output of "x" it should just work (and bitrate shouldn't matter, as its bitrate shouldn't matter once decoded, as its just a frame in memory and is the same size no matter what the underlying bitrate was). But I guess it could be using the same execution units so wont, also understand why you are testing transcoding, not just playback, but plex type streaming (which in truth is also why I care about it).
 

Fritz

Well-Known Member
Apr 6, 2015
2,423
645
113
67
Well, the memory finally arrived today. Amazon thought it might be lost and I had to wait until tomorrow to request a refund. Pissed me off. Anyways, a 256Gb NVMe M.2 drive and 16Gb of ram and I'm in business. Just loaded Mint Cinnamon 20.2 and Kodi. I now have another little puter that's truly useful. :)
 

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
ok, that's weird. I'd assume that if it supports a transcoder "input" of X and a transcoder output of "x" it should just work (and bitrate shouldn't matter, as its bitrate shouldn't matter once decoded, as its just a frame in memory and is the same size no matter what the underlying bitrate was). But I guess it could be using the same execution units so wont, also understand why you are testing transcoding, not just playback, but plex type streaming (which in truth is also why I care about it).
Intel changed the architecture of their QSV transcoder several times, and on anything Coffee Lake and above, it’s one ASIC component that does the codec specific stuff, another that does the resampling, and the video streams are divided into various slices to be worked on in parallel by the GPU's execution unit arrays. The issue here is that when you have a higher resolution you'll have much more data shooting through the system, and those EU array data processing rates are based on the thermal limits, cache size, stuff like that.
 

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
Okay, so this is a bit of a follow-up. So in the interest of trying to figure out the history of the Moderro machine I feel that it is important to look at its old sister (the Cisco branded IEC4650)…there’s a few things of interest to the ghettoMiniMicro crowd (aka people who are too cheap to buy the ex-corporate mini boxes and is keeping their eyes on the killawatts in the house).

So I ordered an IEC4650 just to see what it does.

20109EC5-6847-40B0-BA6B-BE0009D8749A.jpeg

The more details will come later, but here’a a quick preview…

a) Broadwell i3-5100U based
b) DDR3/3L SODIMM based RAM, 2 slots
c) M.2 SATA drive + MiniPCIe slot
d) WiFi+BT included (Azurewave AW-CE123H), Intel i211at integrated NIC
d) A major surprise on the underside of the board
 
Last edited:

thetoad

Active Member
Feb 10, 2021
153
64
28
so, why would someone get this (say at $75) vs a 5070 (with double the ram an double the storage) at $85 (or less). I see a few possible reasons

1) higher single core speed (though it seems going full blast, the 2 cpu's are fairly equivalent)
2) more execution units on gpu (though in reality, is anyone using this gpu for anything besides plex/htpc type workloads? and does the more execution units matter then?
3) fanless (I guess it makes less noise than the 5070, but I can't imagine the 5070 making a huge amount of noise. perhaps in a bedroom it make a difference. I also wonder without a fan how ful bore performance would be impact on this guy. i.e. what happens once the heatsink has soaked up al the heat it can, how will the heat throtling impact the i3.
4) more ram, but again, depending on one's workload, more than 8GB might have not much value.

so people who have both this at the 5070, what do you see in the value in this over the 5070? did I miss anything? do you disagree with my analysis at all?
 

thetoad

Active Member
Feb 10, 2021
153
64
28
The issue here is that when you have a higher resolution you'll have much more data shooting through the system, and those EU array data processing rates are based on the thermal limits, cache size, stuff like that.
btw, I understood that. My Q was more wondering why an initial higher bitrate should matter for transcoding if it can handle decode/playback with ease.

i.e. for the encode portion, a 4k media at an initial bitrate of 400Mbps should be no different than a 20mbps initial bitrate (the frames that have to be encoded are the same size). The only things I can imagine are that as you said, while it might be using just 4% "cpu" to decode the 400mbps media, its actually using 100% of the execution units to keep up with the 400mbps video, leaving nothing for the encode stage. It could also be, as you said, thermal limits, while it can stay within thermal bounds decoding the 400mbps video, that actually pushes it to the thermal limit, again leaving no room for the encode stage.

What I wonder is if its possible to quantify this. Which might also relate to my Q above re the 5070 and would one make a better plex/htpc pc than the other (this being a fanless htpc is a win, though being the 5070 designed to be opened and upgraded is a win for it)
 

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
so, why would someone get this (say at $75) vs a 5070 (with double the ram an double the storage) at $85 (or less). I see a few possible reasons

1) higher single core speed (though it seems going full blast, the 2 cpu's are fairly equivalent)
2) more execution units on gpu (though in reality, is anyone using this gpu for anything besides plex/htpc type workloads? and does the more execution units matter then?
3) fanless (I guess it makes less noise than the 5070, but I can't imagine the 5070 making a huge amount of noise. perhaps in a bedroom it make a difference. I also wonder without a fan how ful bore performance would be impact on this guy. i.e. what happens once the heatsink has soaked up al the heat it can, how will the heat throtling impact the i3.
4) more ram, but again, depending on one's workload, more than 8GB might have not much value.

so people who have both this at the 5070, what do you see in the value in this over the 5070? did I miss anything? do you disagree with my analysis at all?
I have both, a Cisco IEC4650 and an HP t640 with the Ryzen embedded R1505G, so here’s the assessment/comparison between the first 2.

Keep in mind that the 85 dollar version of the 5070 is typically the Celeron J4105/4GB/16GB soldered eMMC variant with no WiFi/BT. The 8GB/64GB version is usually 100+, the Pentium J5005 version usually starts at 120 with the 4GB RAM/16 GB flash…which is close to what I paid for the t640 with the 8GB RAM/64GB Flash/WiFi/BT setup.

1) It depends on what circumstances where you would want 4 14nm Goldmont Atom cores versus 2 Kaby Lake i3 cores. The former has speedstep support so it runs slightly more efficiently in terms of power usage, while the latter has AVX/AVX2 support and a faster GPU. In terms of HTPC/transcoding…eh, the Atom idles slightly lower while the Kaby gives you better h264/265 throughput in Quicksync and x264/265…that being said, for HTPC applications in a living room situation? That’s where you go with a t640 - better margin for growth - you can pull off AV1 software decoding on the Ryzen but it’ll simply take up nearly 70% utilization.

2) Retrogaming - the Coffee Lake GT1 in the base 5070 is much slower on DX12/OpenGL than the Kaby Lake GT2 in the IEC4660. We are talking about half the clock rate with half the EUs. Note that in either case it’s not nearly as good as the Radeon Vega 3 setup in the t640.

3) I actually performed full thermal hysteresis testing on all 3 - the IEC4660 has a full (and rather heavy) metallic case that acts as a heat sink, so even after loading it up for 2 hours the highest the temp will go is around 67C, while on the Wyse 5070 it usually hovers around 72C or so due to its smaller heat sink. This is actually much better than the 84C I typically get on the mostly plastic t640. The thermal mass on the IEC4660 also means that once the CPU load disappears the temp drops quickly. This actually makes sense since the IEC boxes are designed to run inside trade show booths with little or no air flow for days at a time. I mean, sure, it’s not that easy taking apart the IEC box, but it’s actually upgradable once you get into the chassis.

4) Once again, depends on the use case. If it’s an HTPC running either Windows 10/some Linux variant with 8 or 16GB, it’s roughly equal. But on the Wyse 5070, 16GB is the max in Win10 (anything beyond and the machine acts weird) 32 is the hardware limit in Linux/ESXi, and since it has a Realtek NIC embedded, it can’t do ESXi 7 out of the box (the same also applies for the HP t640). The 4650/4660 are Intel i211/217 equipped so it can run ESXi out of the box.

In the case of the IEC4660 and the t640, both will run happily with 64GB of RAM and can use NVMe and SATA drives (the HPs can also use the Sandisk Mothim SD7 cards off the M.2 key M port). As I recall the Wyse 5070s are soldered eMMC and SATA M2 only. So if the use case calls for running a quiet low duty hypervisor somewhere inside a garage, I’ll tape an IEC4650 or 4660 to a wall somewhere and it’ll do quite well. The Cisco Moderro IEC4650 (which I didn’t have a chance to document yet) is nice because you can conceivably feed it a bunch of IT department leftover parts (DDR3L RAM, M.2 SATA, mSATA) and it’ll still be somewhat useful for the foreseeable future.
 
Last edited:
Aug 19, 2015
31
8
8
The more details will come later, but here’a a quick preview…

a) Broadwell i3-5100U based
b) DDR3/3L SODIMM based RAM, 2 slots
c) M.2 SATA drive + MiniPCIe slot
d) WiFi+BT included (Azurewave AW-CE123H), Intel i211at integrated NIC
d) A major surprise on the underside of the board
I'm quite interested to see the underside of the board.

The included quickstart guide references the 466xx, 566xx, and the 766xx models. Looking at Moderro's "datasheets", the only differences I note between the 4660 and the 5660 is memory and storage. The three 766xx models appear to sport an i5, per this guide. I can't seem to find any info on the 766xx, models on their site or elsewhere. Has anyone seen these in the wild?
 

thetoad

Active Member
Feb 10, 2021
153
64
28
thanks for the breakdown. from what it sounds like, for my use case, where basically I just want something to move my plexserver to (for hw transcoding), it sounds like the 8GB/64GB $85 wyse that I bught is good enough.
 

WANg

Well-Known Member
Jun 10, 2018
1,116
696
113
43
New York, NY
btw, I understood that. My Q was more wondering why an initial higher bitrate should matter for transcoding if it can handle decode/playback with ease.

i.e. for the encode portion, a 4k media at an initial bitrate of 400Mbps should be no different than a 20mbps initial bitrate (the frames that have to be encoded are the same size). The only things I can imagine are that as you said, while it might be using just 4% "cpu" to decode the 400mbps media, its actually using 100% of the execution units to keep up with the 400mbps video, leaving nothing for the encode stage. It could also be, as you said, thermal limits, while it can stay within thermal bounds decoding the 400mbps video, that actually pushes it to the thermal limit, again leaving no room for the encode stage.

What I wonder is if its possible to quantify this. Which might also relate to my Q above re the 5070 and would one make a better plex/htpc pc than the other (this being a fanless htpc is a win, though being the 5070 designed to be opened and upgraded is a win for it)
The Quicksync transcoding tests were done using Handbrake, and handbrake has extensive logs which gives you an idea of what it is doing. The transcoding pipeline is essentially a couple of ASICs handling format specific stuff, and the EUs on the GPU are used to do the actual frame section processing in slices (don't forget that the GPU still has to do stuff like redraw the screen even as things are being done). Sometimes the tasks uses up all the EUs and the read stream needs to process at the same time as the write stream, and if the quicksync pipeline is over a certain latency threshold the CPU itself might be tasked to handle parts of the process. If you have less EUs and your CPU has less horsepower per thread you might end up with dropped frames. As for how to quantify it - do what I do, transcode videos at various formats/resolution/bitrates and see how the machine behaves in terms of CPU usage, GPU usage, heat generation/power usage and then see if you are okay with what you see. For me the HTPC tests are only one facet of the IEC4660's performance envelope. My real interest is how well it does as a low wattage, ESXi7 compliant hypervisor.

thanks for the breakdown. from what it sounds like, for my use case, where basically I just want something to move my plexserver to (for hw transcoding), it sounds like the 8GB/64GB $85 wyse that I bught is good enough.
Probably - if you got a J4105/8R/64F version for only 85, then that’s usually good enough for basic HTPC applications.
 
Last edited:

Markess

Well-Known Member
May 19, 2018
729
360
63
So I ordered an IEC4650 just to see what it does…
d) A major surprise on the underside of the board
A cliffhanger! Could it have something to do with the mystery thermal pads that didn’t connect with anything on the 4660 (pictured back in post #35)?

I actually had an IEC 4650 sitting in my eBay watch list (along with a ton of other stuff stuff I don’t really need). So, now I’m curious too!