EEVblog Well that was an interesting 37 minute teardown of a Huawai, box of chips and shit. I still don’t know what it’s supposed to do but the MSRP is $175,000
Getting an accurate price for something like this will be difficult, i cant speak for the video market but as a system engineer for telecommunications when you buy something like this the hardware itself is only part of the price and generally only 40-60% of the cost. Throw this on the secondhand market and it would be worth about its weight in recyclable material thats it. Much like Cisco equipment the real money is in the total service contract, ie ongoing support, turn around time on replacement hardware in the event of failure and software updates etc. It's been years since i've dealt with Huawei personally but they were really aggressive on pricing compared to Cisco, Juniper etc. With an MSRP of 175k behind closed doors in a large company you would expect a quote for around half of that, so lets say 90k and of that i'd estimate the hardware itself would have only been about 50-60k with the remainder for a 3 year negotiable contract - it starts to get merky when you start talking negotiations though. What i found most interesting with the Huawei gear that i came in contact with is the CLI was in Chinese, ie all the characters were the typical chinese looking ones - we couldnt even make basic changes using it and had to rely on a Huawei employee to get the initial configuration set to then be able to use the web tools and external software to control things.. thankfully the project went in another direction :)
Not less then $20.000 from my experience depend on customer previous orders and connection with manufacturer. All prices in that business are floating and every hardware and software require price negotiation.
What device has a Chinese CLI? Used many different switches and routers from Huawei but always only seen English, perhaps devices made for PRC has Chinese as native CLI.
@@JoelBergmark It was DSL related equipment, and im not talking consumer or enterprise side this was proper telco grade to terminate customer connections - so long ago i dont recall a specific model number. Technically one could say it spat out ascii characters.. but it was just gobbledygook without a terminal program that supported the Chinese/Mandarin character set.
Yep. They even have recorder boxes for ip cameras with just a single hisilicon soc that can record up to 16 fhd streams. They even have ones that records 4k streams. Sure, it's not at 60fps but I still feel like this Huawei kit must've been outdated as soon as it came out. Or I just don't understand what exactly it does. Or maybe I'm remembering computers where better then they actually were in 2013. 😋
We do up to 6 streams of video+audio transcoding on an Intel Atom processor, with 5 Atoms per blade, up to 4 transcoding blades per system, for a max of 120 simultaneous transcoded streams. The blades go into a 16-slot backplane that uses naked 1Gbps SERDES (SGMII) to provide 52 fully switched ports for stream routing. We can pull down all the transponders from multiple TV satellites and dump any selected combination into multiple output formats, including HLS/DASH, QAM, DVB-T/T2, NTSC, ATSC, ISDB-T and more. And I'm one of the Systems Engineers. Yeah, it's fun!
Curious who the potential clients would be. Like big news channels for getting real time info around the world, or for the military to organize drone strikes 😏?
Hotels & resorts, convention centers, college campuses, hospitals, apartment buildings, small cable companies, etc. Oh, and barracks and homes on military bases. Basically everyone needing a satellite TV feed WITHOUT a tiny dish and a set-top box.
The hotel one is a good point. Quite a few large hotels on the Gold Coast use retransmission systems to avoid having Foxtel STBs in each room. They usual method I've seen is having 4 or 5 STBs in the comms room, each tuned to a different channel with the output being captured and sent to a retransmission system which grabs the video and audio outputs, muxes them into a single DVB-T transport stream and broadcasts it along with the regular FTA transport streams along the one coax distribution system. Seen similar systems for residential blocks of the mid 2000s, which have systems that retransmit Foxtel's DVB-C transport streams along the same coax as FTA. Both systems cut down costs, as the entire building does not need to be cabled up for either DVB-S2 (which requires two RG11 runs for V and H polarisations) or DVB-C.
Can you reveal the brand ?. I manage some Appear TV setups which is commonly used here in Northern Europa, your system sounds to be in the same ballpark ?
@@chadmed Yeah, the "stack of STBs" is actually the right way to go for *some* installations. But they send content "in the clear", which is a huge problem for premium services that must be encrypted at all times, right up to the viewer. A simpler approach is CSS (Channel Stacking Switch) systems that mix satellite feeds between the dishes and the STBs to provide custom lineups to viewers while maintaining the original encryption. The system I work on wraps a large CSS with a massive amount of processing (and tech) to ensure no signal is ever accessible in an unencrypted state. It's actually one of the main reasons we exist: It is impossible to pirate our streams via gaps in the system.
That weird metal pin around 6:00 is just for alignment during assembly/maintenance. See the taper on the pin? It's so you don't need to 'aim' the module and hope that the connectors will line up; the metal spike aligns it for you!
Similar units i've worked with sell for 30K or more, back in 2013 they'd probably be 70K+ USD. It's probably useless mostly today. These units are highly modular, and they'll reuse the same modules to scale them to the larger units to process more video calls. This kind of device enables a video conference, so it has to deal with processing and transcoding the video streams from the video conference devices. So you can have all your locations in your business in a single call and talk to each other. Assuming this came from emergency services, it was probably used for training and meetings between the locations for police/fire/EMS or something like that.
Why do you guys think this is for video conferencing? I believe it's a "broadcast encoder" used by BDUs (cable TV and sat TV companies) to re-encode video streams for broadcast distribution via cable or sat TV.
@@LudicInterface because it says so in the datasheet? e.huawei.com/pl/related-page/products/enterprise-network/telepresence-video-conferencing/infrastructure/vp9600/tpvc-mcu-vp9600 similar to eg Cisco TelePresence MCU 5300 So all endpoints connect to this device which decodes each audio/video, mixes them/does fancy layout and then re-encode for each participant Modern systems get rid of these complex boxes by just switching the encoded streams and leave the layout to the endpoint. (so each endpoint has to decode multiple streams instead)
Send the power supply to DIODEGONEWILD. That is the guy with the cute cat and European accent. He does some great reverse engineering schematics of power supplies.
wow! the power supply is made by my dad's company. they provide a large amount of power stuff to Huawei. I've just asked. the boom cost is under 100 dollars! Thousands of this model are still produced every month.
Wow, power supply part is under $100. Wonder what the rest of the unit costs (factory cost that is). Then it gets to market and its like $60,000+ ... holy crap.
I noticed that, too. It doesn't have the same ancillary components around it that the other 8 have; maybe some type of controller or interface for the 8 channels.
An MCU is used to connect multiple video conferencing calls together. So there would have been a bunch of meeting rooms each with a video conferencing device in there and if they all wanted to meet, they would dial into this MCU. The MCU mixes the audio and video streams, and then sends a stitched together image of the far ends to each video device. You can either have lots of small meetings or one large one, this device supported up to 24 1080p calls. These kinds of video conferencing devices and MCUs are mainly sold to big enterprises and they are not cheap. The room devices can be between $25k - $299k list and the MCU something like $8k a port. This sounds ridiculous, but remember that these are enterprise devices, one of the main selling points is they cut down travel, so it doesn’t take many transatlantic flights saved before you have paid for the devices in time and fares.
@@georgedemean2228 The description of Skype will tell you what's wrong with it in the context of a meeting room video conferencing solution. It's primarily for "computers, tablets, mobile devices". The devices used in this context will have high quality cameras designed to capture 3-20 people around a single table and they will use standards based SIP or H.323 protocols that allow them to interoperate with all the other video conferencing devices deployed by enterprises world-wide.
Well that was an interesting 37 minute teardown of a Huawai, box of chips and shit. I still don’t know what it’s supposed to do but the MSRP is $175,000
These connectors are called power blade, are from the company Amphenol ICC, are very good and support multiple connections to high current, not that you disconnect with the current flowing.
I visited Australia while the first Crocodile Dundee movie was still in the theaters. I saw it in the US just before I left (cultural study material), then watched it again with friends in Melbourne. While we laughed together most of the time, there were several times I was the only one laughing, and several times I had no clue why everyone else was laughing. We spent the rest of the evening explaining those scenes to each other. The final scene in the movie became my favorite, once I understood it.
On such a redundant device, I would expect ECC ram for the main controller and the network stuff, but not on the a/v side (no care would be given if a bit flip happens on a video feed), so at least 2 types of ram chips, the other might be because of timings or bus width ? don't think it would be ram shortage on such small runs.
Not really. Manufacturers have had multiple sources for DRAM for many years. (Shortages, design problems, ... earthquakes... plenty of things can mess with your supply chain.)
These MCUs (I'm used to different brands, so take with grain of salt..) would basically be a switch for A/V conferences.. You would "dial" in via SIP or H.323 and each person connecting would get it's own point to point connection back to this box, which would do all the transcoding for different bitrates, and all the audio mixing and echo cancellation. You would create connection points (basically a URL) with a unique number, which defined how many resources that conference would allow, so you could give a secretary a 3-party conference, and reserve a 20-party conference for the bosses big conference room, etc.. I haven't done anything like this for about 15 years, which might put this device on par with my old outdated experience, but I think all this is done with DSP resources in routers these days... not really sure..
That data connector on the backplane is used to connect boards together in the larger chassis version. There will be some backplane interconnect, which is not needed in single board version. That is also why there are so beefy spikes, they're used as guides when you slide these boards to the chassis.
The little 8 pin DIP on the backplane could be an EEPROM that stores inventory data, likely so that the backplane serial number, part number, revision etc can be identified within the OS running on the box. Quite common in telecomm'sy equipment. May be DIP if there are no surface mount components so they could avoid a reflow process for that board. Edit: spoke too quickly. 😋 Looks like a press fit socket, so no soldering at all on that board.
Huawei are one of the largest players in the VDSL DSLAM market which have their SFP modules on a 45 degree angle, protruding the front of the switch/router/modem.
I let some Chinese electrons out the other day. They stunk up the place for almost a day. Chinese voltage controller ( DC to DC ) became Chinese fire breathing dragon. Rated at 36 Volts, will tolerate 40 Volts, becomes smoke bomb at 45 Volts.
16:16 the data connection on the power board is only there to hold the board in place, the processor PCB is the same as in the modules that can be swapped into the larger chassis' where you can have multiple processor boards for redundancy. This is just another classic case of the same PCB for different versions of a tiered product line. :)
Nice! I never knew that dedicated video conference hardware was a thing. Until now, I assumed one always used servers for this, maybe with some GPUs for video transcoding.
@22:17 people might not think of Canada as a high-tech place (with all of us living in igloos and such) but keep in might that modern things like AMD Radeon video cards have their roots in Canada via the former ATI Technologies company. Great video! Love looking at enterprise/industrial design and what results when there are no bean-counters involved in the design process.
24:27 That would be Hisilicon, a fabless semiconductor company owned by Huawei who do their in-house processor designs (such as the Kirin series of SoCs in their smartphones)
One good reason to use press-fit for those connectors is that a PCB as thick as that backplane wouldn't get very good fill on the plated through holes as it is passed through wave solder. And it's too thick for pin-in-paste reflow. So press-fit is your only option in that case.
Yes, Huawei is (or *was*) actually replacing everything in datacenter after datacenter, all the old cisco, juniper, etc. routers, switches, fiber optic concentrators, you name it. Mainly due to their cheaper price, and excellent performance. Until the people who fully backdoor all Ciscos (I have seen a switch being rootkitted live) figured out: "Hey !!!! how come we can't spy on the customers internet traffic anymore ?" So they complained to their masters, and that's why the US govt. went so heavy against Huawei. Their reasoning was simple: we used to spy on the sheeple, so I guess China must be the ones spying now.
Exactly. Last time I was lucky enough to see a Huawei datacenter switch it was one of the best engineered things I'd ever seen. Even this teardown shows the quality of stuff. Of course no Altera FPGAs anymore, but they'll now throw everything at the problem to make equivalents or better. SMIC and others aren't too many process nodes behind and they're catching up. ASML already has sold them EUV tech before the sanctions came in. And you're right, products are cheaper with no NSA backdoors. Source is not fully open but 100% available to top end customers including governments, but everyone spouts this nonsense about Huawei spying on everyone with resultant sanctions, bans which only spurs them on to replicate the tech themselves. In the process AMD/Xilinx or Intel/Altera lose another customer for very high margin components - further affecting the shocking trade deficit.. Yes China make a lot of crap for an absurdly low price, or these days outsource it to Belt and Road partner countries, but they also have stunning research and some of the best engineered products on the planet. The fact Huawei employees (or US subsidiary Futurewei) make up a large percentage of the Terabit IEEE study groups, 5G standards committees and panels is testament to their research and development prowess. Yet we have a bunch of utterly technology clueless politicians and their military intelligence mates spouting nonsense. In the meantime they're running up even more unserviceable debt which will damage the western standard of living we all enjoy. This is not Long March China under Mao or during the Cultural Revolution. In some ways it's more capitalist than most socialist-leaning western countries. Anything to try to protect the dollar reserve when in the end it's pointless. Better to work with China and allies rather than treating them as pariahs. Ultimately the boot will be on the other foot and they have very long memories.
Those power connectors on the back of the power supplies and fan tray are Multibeam XL connectors... They are a standard off the shelf part made by TE Connectivity and I believe they are also licensed to other connector manufacturer's as well, so nothing custom there... Also, those "ground pins" on either end of the backplane card are probably not actually used for ground, they are most likely just alignment pins used for blind mating the daughter card. No idea what this would have cost, but a lot of familiar looking connectivity on that thing for the period... I'm an engineer for TE connectivity so this all looks familiar to me lol Also, you nailed the high speed mezzanine connector, made by Samtec and TE... you must be some kind of expert or something! EDIT: looks like this response wasn't needed lol, you pretty much said everything I did above later in the video.
@@BspVfxzVraPQ There's not much I can do when idiots cut holes in the walls. (these are the same geniuses that replaced the door+lock without giving _anyone_ a key -- including the building owners. Maintenance had to drill the lock to get us back in there; high security locks are not easy to drill. It literally takes two seconds to swap the core -- insert control key, turn left, pull core out.)
When I worked for a major US defense contractor, we built equipt to the same standards as this. Same manufacturing techniques and standards. The prime reason for it to exist was to last far longer than the designed lifetime. Reliability was the primary concerned. I recognize many of the cable manufacturing standards, and those were from the mid- to late-80's.
The large pins on the backplane would be for alignment. Doesn't look like it has guide rails for the module, so vertical alignment could be off if some yahoo is trying to shove the module in and quickly destroy both the backplane but the module's connectors. Replacing a backplane is a pretty big pain in the butt since they are almost always the last component to come out. I had to do one on a server cluster because a blade with a defective connector that would either fold or break off a few of the backplane's connector pins. Worse yet was that the damaged backplane port would cause the same damage to good server blades and repeat the cycle. 3 blades and half the slots were foobared before the damage was noticed. The blades were salvageable due to the pressfit connectors and donor pin slices from a dead blade. The backplane connectors were monolithic pressfit with no sourcable replacements even if they could be pulled out. Luckily used good backplanes on ebay were pretty cheap.
Checked it out on LBRY. Video quality on youtube is better than LBRY. They might be restricted to 2k (1080p) or their compression codec sucks. LBRY doesn't show you the res you're viewing so you can only go by appearance. The web version of LBRY just buffers forever, but the win10 app version will actually play videos.
I've an old Juniper MX series (biga$$ internet router) DPC line card I've wanted to send in for a teardown, but it's really big and heavy. Basically think the board you've got without the chassis around it. They're about the same size, very somewhat similar idea. 4x10G (ethernet) ports on the front of it. A bunch of custom ASICs and some FPGAs, a switch fabric to the backplane (fabric switch cards are connected to the backplane). A PPC on the linecard for management, housekeeping, and bringing up the ASICS and FPGAs. That talks to the chassis management (called a routing engine) via a couple of ethernet ports (on the backplane) using TNP (after TFTP booting from the routing engine!!!!). Very distributed architecture in the Juniper MX. MX240/480/960 support far more than just ethernet, but, the other types of physical interfaces are going away. The line cards basically do all the routing right at the line card, the routing engine connects to the peers, digests a routing table, and sends it to the management PPCs on the line cards. They do the dirty of making that into whatever formats/etc are needed by the ASICs on that line card (running code loaded at line card bootup from the RE!). As packets come in filters are applied, routing decisions are made, and the packet is kicked over to the outgoing linecard via the fabric, or, locally. Other end the line card does a bit of the reverse, packaging it back into say an ethernet frame (possibly making it a VLAN frame, or getting way moree complicated with MACSEC and non ethernet port types!) The line cards also handle a TON of routine housekeeping stuff so the main routing engine doesn't need to. They also protect the main routing engine from DoS attacks by applying rate limiting right there on the line card for anything they hand off towards the routing engine. From very simple/common things like TTL exceeded (too many hops in transit), to some more complicated things like the ping/pong of the BFD protocol.
Impressive. Even the salvageable parts on those boards are worth a day's effort of de-soldering. Love it! 220uF tantalum SMD caps are amazing, never seen one higher than 100uF. The 32.768 KHz OCXO is definitely a keeper, I would love to put it in my all TTL logic desktop clock, replacing the already pretty accurate TCXO. Those 52V SMPS's can be ideal for RuiDeng's new lab supplies. Please do a test of those planar transformer power converter modules.
You can also buy MLCC capacitors at 6.3V with 220uF capacity in a 1206 package. They are very expensive at 0.50$ per capacitor in bulk but they will generally outperform the tantalum capacitors!
If you want to see Huawei kit BT uses loads of it. Also BT uses similar kit in the exchanges to buffer BT TV offerings to keep the backbone clearer of traffic.
From Wikipedia: "Cortina is a fabless semiconductor company. It outsources all semiconductor manufacturing to merchant foundries. The company is based in Sunnyvale, California. It also has other research and development sites in Hsinchu (Taiwan), Ottawa (Canada), Raleigh (USA) and Shanghai (China). "
Huawei is making a looooot of stuff, not only LTE/5G but wired network (dwdm, enterprise routers, switches, access points, gpon so ftth), servers, power supply devices.
.... The Australian Government prevented the company from involvement in our 5g infrastructure. Citing security concerns as all Chinese companies are required to respond to any request from Xi Jinping.
@@robbieaussievic the Australian goverment and US are 100% uneducated idiots and everything is just a coverup lies that they spy,everything is just for money specially for us they just scared mostly for market and stupid apple cause they know Huawei is much better in quality and years ahead when they even co invented 3g and 4g. the real spy's are apple
@@robbieaussievic So? Germany or UK didnt do that. Ot course Huawei is connected to Chinese goverment, just like Cisco, Juniper are connected to USA, you can choose which country backdoor will you have.
Why do I get the feeling this was used in something for state or federal government ( not security to be fair ) and was removed precisely because it's Huawei ?
I've got a customer that I do IT consulting for who has a similar video conferencing system from Cisco that cost around a million USD to install. But it is an incredibly trick setup.
4 ปีที่แล้ว
@EK89 this takes H.264 streams from video conferencing systems and combines them into an output stream that shows all the pictures or one of them based on speaker detection. It's receiving the video streams of the various sources and encodes the output for the other parties to view in realtime.
Huawei has been making routers and very high end telecom stuff like this for decades now. A lot of it is potentially stolen or reverse engineered from companies like Cisco. I know more than that, but I'd rather not talk about it in a public comment.
In a former life, I worked with and deployed several realtime mpeg4 transcodey things in the early days of ip video. Tandberg, RGB, motorola, scientific atlanta. This brings back the memories. Likely it was before Huawei was on the scene, but all built up to a price. Advancements/features, to be supported in the codecs, were so uncertain that fpga's pretty much needed to be used for transcoding. We also rolled our own picture in picture transcode service, using commodity hardware. 24 streams per U if I recall. Was a fun project in the way back times.
24 SD streams and 8 HD streams in 2 or 3U was what Moto and SciA used just about the same board layout and sold to cisco with a powerpc and not an intel. 88 grand seems to be the list price for 8 HD HDMI in/out with 2 1 Gig copper and 2 1/10Gig Fiber . I would bet some of gear circa 2006 for the second wave of HD channels is being torn out this year. Nothing has changed in the industry, RF ingress and Drop and Insert and content management on/off a multicast seems older than rope. The content file collection I am sure is better, no KENCAST of 20 or 50 Gig movie files and 60-400 meg commercials, just pick them up off AWS and cache. Some bastard ordered the our KENCAST servers with 1 Gig Inflow and 100 meg interface to the content store. The input was 10 times larger than the output to storeage so all humor was lost as the HD filled up. We inherted the whole thing from another telco, so worst was Our on-demand system was always out of sync, The last issue I remember was about before leaving was missing breaking bad episodes. Now you would just put everything in elastic storage and the head ends pick up what you need at the locals minutes before being put on the channel. The on demand movies can be anywhere at any of the headends with a meshed IP network. Soon enough most cable boxes will just put whatever commercial you will respond to inserted right into a NFL game.
@@christopherwhull lol. I had to write our vod ingresser for our "super headend". I see we're kindred. Nice to meet you Chris. Lifetimes ago. Was fun stuff.
@EEVblog the spikes are for blind mating boards, they're pretty common for modular designs like this. Grounded more for safety than anything else, it's really just a mechanical feature.
The ports are angled for ease of cable management. Lets you route the fiber into cable management on the side of a server rack without needing extreme radius bends that can damage the fiber.
As retired test engineer, the testhardware boards has 8 to 12 layers, a tickness of 3.2 mm and gold finish on outsider layers. Test machine connectors ends always whit a bed of pogo-pins. You do not mention the jtag bus on this system with is full whit high pincount bga devices mostly designed in the US.
I was surprised to see the Davinci dm8168 in there. I have spent a lot of time over the last few years working with the little brother to that part (the dm8148), and they can only be described as a hot mess. TI has effectively dropped all support for the part and there are some really nasty silicon bugs to contend with, but for a part that was released in 2010 they have an amazingly fast hardware h.264 encoder built in that is still as fast as almost anything you can buy today. The soul of that part lives on in the Sitara AM57xx series, but hopefully with fewer silicon bugs.
One thing that strikes me: i's not very old! Well, it could be defective - but such expensive devices should run for many years. Not even 10 years old, and now in the bin. That is terrible!
True, not very old and now binned. Why it was replaced could have been for a number of reasons (most likely financial but could also be political). They could have at least tried to sell it 2nd hand and recoup some of the cost but the effort was probably going to cost more than it would have sold for 2nd hand...sad.
@@stormkhan4250 if it was a 5 eyes government agency or contractor, the getting rid of it and replaced with gear built in vietnam or sir lanka is all paid for. The NSA will send any government a unit out of their warehouse at a deep deep discount.
Used from a Canadian reseller they go for 12400 US dollars per single unit, so a rack of ten would be 124000 US Dollars plus mounting and rack hardware costs
There are circular diamond saws designed to drill into ceramic tiles. Put one in a drill press, then drill around one of those pins from the power supply, and then you can desolder that bit of pcb on there much easier.
Actually you WANT turbulent flow around heatsinks. It seems a bit paradoxical, but if you have smooth laminar flow, you end up creating a boundary layer that acts like an insulator. Turbulant flow generates vorticies that bring unheated air into contact with the heatsink. That is why the fins are serrated instad of solid.
Very similar to the approach of server blade and routing chassis from the same era. You're spot on with the comment that cost is no object. It's all about continuity of service delivery, hot swapability/live servicing, throughput and heat control in dense racked environments. What's even better is when you design the server rooms and racks for these things, you need to worry more about BTU output and airflow than anything else. You DON'T want hot spots or no-flow zones or you get real problems.
Lots of good salvage parts here, power modules, good quality caps, top quality inductors, lots of top quality mosfets, a lifetime supply of low profile heatsinks, a home defence weapon (backplane)
It's mind-boggling to consider all the coordination and man-hours that went into the design and creation of this hardware. It's like watching you disassemble a skyscraper.
I like all the added datasheets and extra info added to this video. It's better than watching your hands wave around a plastic stick while you make educated gueses about what something might be :)
Those board-to-board connectors could be Samtec QSTRIP/QPAIRS or something from the Molex mezzanine line. I got to know them when I was hand soldering a few of the former brand onto test fixtures for testing just a CPU+MEM control board assy on high-end test and measurement gear. It was quite a challange for the first time!
As long as the XTZ accelerator can keep up with the QN rate when overclocked and the local MOSFET thermal reactance controllers don’t max out when power vector polarity morphs from linear non-phase reactance to non-standard decoder channel proxy configurators especially when positronic differentiation differs from radial expansion levels-I’m cool with it.
The BOM cost must indeed be enormous! Dang, I wonder how much you could make by selling the components of that thing... I'd happily pay a a few bucks for some of them ;)
HiSilicon develops SoCs based on ARM architecture. Though not exclusive, these SoCs see preliminary use in handheld and tablet devices of its parent company Huawei.
Notice that on the audio board, where there are 9 chips for 8 channels--the center chip does not have two little chips peeking out from under the heatsink. Some kind of controller?
Those pens on the front and underneath are also for alignment. That way when you are inserting the board or module it’s properly aligned so that pins don’t get Miss connect or vent.
PSU at 31:00 - yes, they have a custom AC input filter on its own pcb, but DOH, still need to add an "on cable" common mode ferrite to pass EMC on the AC power cable that leaves that filter pcb and jumps over to the main pcb........ ;-)
.... The Australian Government prevented the company from 'High Level' involvement in our 5g infrastructure. Citing security concerns as all Chinese companies are required to respond to any request by Xi Jinping.
Just wanted to say thanks for your recommendations. It's my birthday this month and I got myself a "Yihua" 937D soldering station - you know, like the cheapest one you recommend worth buying which is exactly my price bracket. It replaced a shoddy plug-in adjustable-on-the-handle iron. ;-) thanks a lot and have a nice one, Greetings from a German viewer.
I subbed to lbry but will stick with youtube mostly because it seems to take hours for videos to show there. Plus after looking around, there is not really much else there (I'm mostly a lets play / minecraft + tech video viewer).
Oh no. You mean you have to wait hours for a video?!? That's completely unacceptable. What could you possibly do in those hours while you're waiting? Almost as bad as the people that complain when a video is uploaded during the time they're sleeping. Like it will disappear by the time they wake up. LOL.
4K version still rendering BTW
Should of let this monster do the transcoding for ya :-)
Peter Stevens should HAVE
Software upscaled?
EEVblog Well that was an interesting 37 minute teardown of a Huawai, box of chips and shit. I still don’t know what it’s supposed to do but the MSRP is $175,000
@@JesusisJesus WTF, I knew it was expensive, but $175k!
I’ve been waiting days for the Mailbag video and then my own video shows up! 😂
Man of the half hour!
Cool piece thanks for sharing
Thanks for sending it in!
Do you know what it does exactly.
SidneyCritic ComedyHound Dave pretty much sums it up in the video.
When you see that "HLi" logo, that is HiSilicon, Huawei's fabless subsidiary that makes all their custom ASIC.
Getting an accurate price for something like this will be difficult, i cant speak for the video market but as a system engineer for telecommunications when you buy something like this the hardware itself is only part of the price and generally only 40-60% of the cost. Throw this on the secondhand market and it would be worth about its weight in recyclable material thats it.
Much like Cisco equipment the real money is in the total service contract, ie ongoing support, turn around time on replacement hardware in the event of failure and software updates etc.
It's been years since i've dealt with Huawei personally but they were really aggressive on pricing compared to Cisco, Juniper etc.
With an MSRP of 175k behind closed doors in a large company you would expect a quote for around half of that, so lets say 90k and of that i'd estimate the hardware itself would have only been about 50-60k with the remainder for a 3 year negotiable contract - it starts to get merky when you start talking negotiations though.
What i found most interesting with the Huawei gear that i came in contact with is the CLI was in Chinese, ie all the characters were the typical chinese looking ones - we couldnt even make basic changes using it and had to rely on a Huawei employee to get the initial configuration set to then be able to use the web tools and external software to control things.. thankfully the project went in another direction :)
Plus, gear like this doesn't have a published price list.
Not less then $20.000 from my experience depend on customer previous orders and connection with manufacturer. All prices in that business are floating and every hardware and software require price negotiation.
What device has a Chinese CLI? Used many different switches and routers from Huawei but always only seen English, perhaps devices made for PRC has Chinese as native CLI.
@@JoelBergmark It was DSL related equipment, and im not talking consumer or enterprise side this was proper telco grade to terminate customer connections - so long ago i dont recall a specific model number. Technically one could say it spat out ascii characters.. but it was just gobbledygook without a terminal program that supported the Chinese/Mandarin character set.
@@sirtra Thanks for the info! That makes more sense to me! :)
That 24:20 mystery logo is the Huawei HiSilicon semiconductor company. They do ARM, Mali, and IC designs
Right, they are also known for video equipment; as many of the smaller IP cameras for home use are based on a SoC by HiSilicon.
Beat me to it. 😝
@@gbraadnl HiSilicon SOC inside at lease 30% of all CCTV cameras produced in China, including mid to high range commercial products.
Yep. They even have recorder boxes for ip cameras with just a single hisilicon soc that can record up to 16 fhd streams. They even have ones that records 4k streams. Sure, it's not at 60fps but I still feel like this Huawei kit must've been outdated as soon as it came out. Or I just don't understand what exactly it does. Or maybe I'm remembering computers where better then they actually were in 2013. 😋
@@mopk47 I am aware. /me lives in China ;-).
We do up to 6 streams of video+audio transcoding on an Intel Atom processor, with 5 Atoms per blade, up to 4 transcoding blades per system, for a max of 120 simultaneous transcoded streams. The blades go into a 16-slot backplane that uses naked 1Gbps SERDES (SGMII) to provide 52 fully switched ports for stream routing. We can pull down all the transponders from multiple TV satellites and dump any selected combination into multiple output formats, including HLS/DASH, QAM, DVB-T/T2, NTSC, ATSC, ISDB-T and more. And I'm one of the Systems Engineers. Yeah, it's fun!
Curious who the potential clients would be. Like big news channels for getting real time info around the world, or for the military to organize drone strikes 😏?
Hotels & resorts, convention centers, college campuses, hospitals, apartment buildings, small cable companies, etc. Oh, and barracks and homes on military bases. Basically everyone needing a satellite TV feed WITHOUT a tiny dish and a set-top box.
The hotel one is a good point. Quite a few large hotels on the Gold Coast use retransmission systems to avoid having Foxtel STBs in each room. They usual method I've seen is having 4 or 5 STBs in the comms room, each tuned to a different channel with the output being captured and sent to a retransmission system which grabs the video and audio outputs, muxes them into a single DVB-T transport stream and broadcasts it along with the regular FTA transport streams along the one coax distribution system. Seen similar systems for residential blocks of the mid 2000s, which have systems that retransmit Foxtel's DVB-C transport streams along the same coax as FTA. Both systems cut down costs, as the entire building does not need to be cabled up for either DVB-S2 (which requires two RG11 runs for V and H polarisations) or DVB-C.
Can you reveal the brand ?. I manage some Appear TV setups which is commonly used here in Northern Europa, your system sounds to be in the same ballpark ?
@@chadmed Yeah, the "stack of STBs" is actually the right way to go for *some* installations. But they send content "in the clear", which is a huge problem for premium services that must be encrypted at all times, right up to the viewer. A simpler approach is CSS (Channel Stacking Switch) systems that mix satellite feeds between the dishes and the STBs to provide custom lineups to viewers while maintaining the original encryption.
The system I work on wraps a large CSS with a massive amount of processing (and tech) to ensure no signal is ever accessible in an unencrypted state. It's actually one of the main reasons we exist: It is impossible to pirate our streams via gaps in the system.
That weird metal pin around 6:00 is just for alignment during assembly/maintenance. See the taper on the pin? It's so you don't need to 'aim' the module and hope that the connectors will line up; the metal spike aligns it for you!
Similar units i've worked with sell for 30K or more, back in 2013 they'd probably be 70K+ USD. It's probably useless mostly today. These units are highly modular, and they'll reuse the same modules to scale them to the larger units to process more video calls. This kind of device enables a video conference, so it has to deal with processing and transcoding the video streams from the video conference devices. So you can have all your locations in your business in a single call and talk to each other. Assuming this came from emergency services, it was probably used for training and meetings between the locations for police/fire/EMS or something like that.
Why do you guys think this is for video conferencing? I believe it's a "broadcast encoder" used by BDUs (cable TV and sat TV companies) to re-encode video streams for broadcast distribution via cable or sat TV.
@@LudicInterface because it says so in the datasheet?
e.huawei.com/pl/related-page/products/enterprise-network/telepresence-video-conferencing/infrastructure/vp9600/tpvc-mcu-vp9600
similar to eg Cisco TelePresence MCU 5300
So all endpoints connect to this device which decodes each audio/video, mixes them/does fancy layout and then re-encode for each participant
Modern systems get rid of these complex boxes by just switching the encoded streams and leave the layout to the endpoint. (so each endpoint has to decode multiple streams instead)
Send the power supply to DIODEGONEWILD. That is the guy with the cute cat and European accent. He does some great reverse engineering schematics of power supplies.
yes please
That would be awesome
did someone say cute cat? i'll vote for this just for the cat even if they annoy dave!
Yes! The guy is absolutely brilliant.
Yea he’s good but his voice though 🙈
There aren’t 9 audio processors! There are 8! The center one is different, has no RAM attached to it, it looks like.
wow! the power supply is made by my dad's company. they provide a large amount of power stuff to Huawei.
I've just asked. the boom cost is under 100 dollars! Thousands of this model are still produced every month.
Your dad makes impressive power supply units! 😎
Thats some nice craftsmanship , you should tell him to be proud of the work they do. And you should be too ! :-)
Wow, power supply part is under $100. Wonder what the rest of the unit costs (factory cost that is). Then it gets to market and its like $60,000+ ... holy crap.
Impressive
20:30 "Tight as a nun's nasty" LOL
Thats right up there with "tighter than a dolphins butt, watertight!"
LOL. A nun's flange. LOL.
Second hand pricing via google from a reasonable source puts the price at ~USD$122,000
But Alibaba has them listed at ~USD$1500
@@ihavenopenisandimustscream My mate Abdul says he can do them for $500, no questions asked, you ain't seen me right?
So, Dave just tore apart a $122,000 video processor? He could have bought 2 new Corvettes for that money.
It goes for 60k new here actfornet.com/store/vp9630-8-ac-universal-transcoding.html
@@ihavenopenisandimustscream $233 on Wish.
On the audio processing board, the IC in the middle does not have memory next to it. So you do have 8 audio channels and an extra IC.
I noticed that, too. It doesn't have the same ancillary components around it that the other 8 have; maybe some type of controller or interface for the 8 channels.
PCI switch
Maybe the extra is for the CCP 😂
China spy chip?
@ Jaco, I was going to say the same thing.
An MCU is used to connect multiple video conferencing calls together. So there would have been a bunch of meeting rooms each with a video conferencing device in there and if they all wanted to meet, they would dial into this MCU. The MCU mixes the audio and video streams, and then sends a stitched together image of the far ends to each video device. You can either have lots of small meetings or one large one, this device supported up to 24 1080p calls.
These kinds of video conferencing devices and MCUs are mainly sold to big enterprises and they are not cheap. The room devices can be between $25k - $299k list and the MCU something like $8k a port. This sounds ridiculous, but remember that these are enterprise devices, one of the main selling points is they cut down travel, so it doesn’t take many transatlantic flights saved before you have paid for the devices in time and fares.
Huh I thought it was a broadcast encoder for BDUs.
Hmmm...what's wrong with skype ?
@@georgedemean2228 The description of Skype will tell you what's wrong with it in the context of a meeting room video conferencing solution. It's primarily for "computers, tablets, mobile devices". The devices used in this context will have high quality cameras designed to capture 3-20 people around a single table and they will use standards based SIP or H.323 protocols that allow them to interoperate with all the other video conferencing devices deployed by enterprises world-wide.
@@corce9631 Awesome, Thanks Pete
Now that's quite a different kind of "made in china", Dave
Well that was an interesting 37 minute teardown of a Huawai, box of chips and shit. I still don’t know what it’s supposed to do but the MSRP is $175,000
wow
Its a chinese spy device.
It would make a sweet Plex server.
Jesus, it's dedicated hardware "skype," basically...
The two metal spikes are only for mechanical alignment before the connector fingers mate / get damaged.
I don't understand a word of it...
but I watch every minute of it.☺
These connectors are called power blade, are from the company Amphenol ICC, are very good and support multiple connections to high current, not that you disconnect with the current flowing.
WOW Dave, your enthusiasm has exhausted me, wish I had just 10% of it, interesting tear down.
"That's not a knife! That's a knife!" - Do every Australian have such a knife like Crocodile Dundee?
So true! lol
Yes, yes we do.
Heck yes, you need it to fight for toilet paper.
I visited Australia while the first Crocodile Dundee movie was still in the theaters. I saw it in the US just before I left (cultural study material), then watched it again with friends in Melbourne. While we laughed together most of the time, there were several times I was the only one laughing, and several times I had no clue why everyone else was laughing. We spent the rest of the evening explaining those scenes to each other. The final scene in the movie became my favorite, once I understood it.
How else can you open a 10 Kilo box mate! LOL
I even keep the screws and hardware when I tear down stuff like this. Good quality hardware is expensive and very useful!
Interesting how a single device has Nanya, Samsung, Hynix and Micron RAM.
On such a redundant device, I would expect ECC ram for the main controller and the network stuff, but not on the a/v side (no care would be given if a bit flip happens on a video feed), so at least 2 types of ram chips, the other might be because of timings or bus width ? don't think it would be ram shortage on such small runs.
Indeed, fascinating. RAM is a commodity now and makes sense for different teams to use different brands
and obviousl, I should wait to be awake before posting, because the ecc would be handled by the controller and not the chip itself
Not really. Manufacturers have had multiple sources for DRAM for many years. (Shortages, design problems, ... earthquakes... plenty of things can mess with your supply chain.)
Global supply chain management at its best!
There must be 500 bucks in tantalum's alone in the BOM lol That thing is awesome,..
Thank you for the great content Dave!!! Your great guides for engineering interviews helped me a lot getting a job as an electronics engineer.
These MCUs (I'm used to different brands, so take with grain of salt..) would basically be a switch for A/V conferences.. You would "dial" in via SIP or H.323 and each person connecting would get it's own point to point connection back to this box, which would do all the transcoding for different bitrates, and all the audio mixing and echo cancellation. You would create connection points (basically a URL) with a unique number, which defined how many resources that conference would allow, so you could give a secretary a 3-party conference, and reserve a 20-party conference for the bosses big conference room, etc.. I haven't done anything like this for about 15 years, which might put this device on par with my old outdated experience, but I think all this is done with DSP resources in routers these days... not really sure..
Huawei technologies definitively blows my mind.
That data connector on the backplane is used to connect boards together in the larger chassis version. There will be some backplane interconnect, which is not needed in single board version. That is also why there are so beefy spikes, they're used as guides when you slide these boards to the chassis.
The little 8 pin DIP on the backplane could be an EEPROM that stores inventory data, likely so that the backplane serial number, part number, revision etc can be identified within the OS running on the box.
Quite common in telecomm'sy equipment.
May be DIP if there are no surface mount components so they could avoid a reflow process for that board.
Edit: spoke too quickly. 😋 Looks like a press fit socket, so no soldering at all on that board.
Huawei are one of the largest players in the VDSL DSLAM market which have their SFP modules on a 45 degree angle, protruding the front of the switch/router/modem.
Probably mentioned 100 times Dave, that solid state drive was a good old Compact Flash card.
what a beast!...
the "audio" board had 8 channels...the centre one didnt have memory around it...so its differernt..probably a countroller/multiplexer
Another PCIe switch perhaps?
You better start wearing a foil hat, the electrons that fell out of that huawei device will be watching you from now on. ;)
I let some Chinese electrons out the other day. They stunk up the place for almost a day. Chinese voltage controller ( DC to DC ) became Chinese fire breathing dragon. Rated at 36 Volts, will tolerate 40 Volts, becomes smoke bomb at 45 Volts.
16:16 the data connection on the power board is only there to hold the board in place, the processor PCB is the same as in the modules that can be swapped into the larger chassis' where you can have multiple processor boards for redundancy. This is just another classic case of the same PCB for different versions of a tiered product line. :)
Nice! I never knew that dedicated video conference hardware was a thing.
Until now, I assumed one always used servers for this, maybe with some GPUs for video transcoding.
@22:17 people might not think of Canada as a high-tech place (with all of us living in igloos and such) but keep in might that modern things like AMD Radeon video cards have their roots in Canada via the former ATI Technologies company. Great video! Love looking at enterprise/industrial design and what results when there are no bean-counters involved in the design process.
24:27 That would be Hisilicon, a fabless semiconductor company owned by Huawei who do their in-house processor designs (such as the Kirin series of SoCs in their smartphones)
One good reason to use press-fit for those connectors is that a PCB as thick as that backplane wouldn't get very good fill on the plated through holes as it is passed through wave solder. And it's too thick for pin-in-paste reflow. So press-fit is your only option in that case.
Yes, Huawei is (or *was*) actually replacing everything in datacenter after datacenter, all the old cisco, juniper, etc. routers, switches, fiber optic concentrators, you name it. Mainly due to their cheaper price, and excellent performance.
Until the people who fully backdoor all Ciscos (I have seen a switch being rootkitted live) figured out: "Hey !!!! how come we can't spy on the customers internet traffic anymore ?"
So they complained to their masters, and that's why the US govt. went so heavy against Huawei. Their reasoning was simple: we used to spy on the sheeple, so I guess China must be the ones spying now.
Exactly. Last time I was lucky enough to see a Huawei datacenter switch it was one of the best engineered things I'd ever seen. Even this teardown shows the quality of stuff. Of course no Altera FPGAs anymore, but they'll now throw everything at the problem to make equivalents or better. SMIC and others aren't too many process nodes behind and they're catching up. ASML already has sold them EUV tech before the sanctions came in.
And you're right, products are cheaper with no NSA backdoors. Source is not fully open but 100% available to top end customers including governments, but everyone spouts this nonsense about Huawei spying on everyone with resultant sanctions, bans which only spurs them on to replicate the tech themselves. In the process AMD/Xilinx or Intel/Altera lose another customer for very high margin components - further affecting the shocking trade deficit..
Yes China make a lot of crap for an absurdly low price, or these days outsource it to Belt and Road partner countries, but they also have stunning research and some of the best engineered products on the planet.
The fact Huawei employees (or US subsidiary Futurewei) make up a large percentage of the Terabit IEEE study groups, 5G standards committees and panels is testament to their research and development prowess.
Yet we have a bunch of utterly technology clueless politicians and their military intelligence mates spouting nonsense. In the meantime they're running up even more unserviceable debt which will damage the western standard of living we all enjoy. This is not Long March China under Mao or during the Cultural Revolution. In some ways it's more capitalist than most socialist-leaning western countries.
Anything to try to protect the dollar reserve when in the end it's pointless. Better to work with China and allies rather than treating them as pariahs. Ultimately the boot will be on the other foot and they have very long memories.
Loved it. My first EE job was at Compaq 35 years ago, working on heat dissipation.
Those power connectors on the back of the power supplies and fan tray are Multibeam XL connectors... They are a standard off the shelf part made by TE Connectivity and I believe they are also licensed to other connector manufacturer's as well, so nothing custom there... Also, those "ground pins" on either end of the backplane card are probably not actually used for ground, they are most likely just alignment pins used for blind mating the daughter card. No idea what this would have cost, but a lot of familiar looking connectivity on that thing for the period... I'm an engineer for TE connectivity so this all looks familiar to me lol
Also, you nailed the high speed mezzanine connector, made by Samtec and TE... you must be some kind of expert or something!
EDIT: looks like this response wasn't needed lol, you pretty much said everything I did above later in the video.
What engineering, wow.
Was there anything on the CF card?
I'm surprised you didn't check because I was thinking the same thing as well.
Probably just system stuff
Super cool to see a (relatively) modern device that isn't just a PC in a custom box. Some beautiful engineering in there.
Greetings: Considering the lack of dust (especially on the fans) I suspect this was a spare unit or an early failure.
Or it was run in a very clean room. (all of my gear looks like it's been left of the surface of the moon from construction in two offices.)
If your datacenter is dusty you are doing it wrong.
@@BspVfxzVraPQ There's not much I can do when idiots cut holes in the walls. (these are the same geniuses that replaced the door+lock without giving _anyone_ a key -- including the building owners. Maintenance had to drill the lock to get us back in there; high security locks are not easy to drill. It literally takes two seconds to swap the core -- insert control key, turn left, pull core out.)
He did say that someone may have been wrong before the power supplies didn't he?
I'd say crib death as a v e Kohl's that type of thing.
When I worked for a major US defense contractor, we built equipt to the same standards as this. Same manufacturing techniques and standards. The prime reason for it to exist was to last far longer than the designed lifetime. Reliability was the primary concerned. I recognize many of the cable manufacturing standards, and those were from the mid- to late-80's.
The large pins on the backplane would be for alignment. Doesn't look like it has guide rails for the module, so vertical alignment could be off if some yahoo is trying to shove the module in and quickly destroy both the backplane but the module's connectors.
Replacing a backplane is a pretty big pain in the butt since they are almost always the last component to come out. I had to do one on a server cluster because a blade with a defective connector that would either fold or break off a few of the backplane's connector pins. Worse yet was that the damaged backplane port would cause the same damage to good server blades and repeat the cycle. 3 blades and half the slots were foobared before the damage was noticed. The blades were salvageable due to the pressfit connectors and donor pin slices from a dead blade. The backplane connectors were monolithic pressfit with no sourcable replacements even if they could be pulled out. Luckily used good backplanes on ebay were pretty cheap.
Checked it out on LBRY. Video quality on youtube is better than LBRY. They might be restricted to 2k (1080p) or their compression codec sucks. LBRY doesn't show you the res you're viewing so you can only go by appearance. The web version of LBRY just buffers forever, but the win10 app version will actually play videos.
I've an old Juniper MX series (biga$$ internet router) DPC line card I've wanted to send in for a teardown, but it's really big and heavy. Basically think the board you've got without the chassis around it. They're about the same size, very somewhat similar idea. 4x10G (ethernet) ports on the front of it. A bunch of custom ASICs and some FPGAs, a switch fabric to the backplane (fabric switch cards are connected to the backplane). A PPC on the linecard for management, housekeeping, and bringing up the ASICS and FPGAs. That talks to the chassis management (called a routing engine) via a couple of ethernet ports (on the backplane) using TNP (after TFTP booting from the routing engine!!!!). Very distributed architecture in the Juniper MX. MX240/480/960 support far more than just ethernet, but, the other types of physical interfaces are going away. The line cards basically do all the routing right at the line card, the routing engine connects to the peers, digests a routing table, and sends it to the management PPCs on the line cards. They do the dirty of making that into whatever formats/etc are needed by the ASICs on that line card (running code loaded at line card bootup from the RE!). As packets come in filters are applied, routing decisions are made, and the packet is kicked over to the outgoing linecard via the fabric, or, locally. Other end the line card does a bit of the reverse, packaging it back into say an ethernet frame (possibly making it a VLAN frame, or getting way moree complicated with MACSEC and non ethernet port types!)
The line cards also handle a TON of routine housekeeping stuff so the main routing engine doesn't need to. They also protect the main routing engine from DoS attacks by applying rate limiting right there on the line card for anything they hand off towards the routing engine. From very simple/common things like TTL exceeded (too many hops in transit), to some more complicated things like the ping/pong of the BFD protocol.
Impressive. Even the salvageable parts on those boards are worth a day's effort of de-soldering. Love it! 220uF tantalum SMD caps are amazing, never seen one higher than 100uF. The 32.768 KHz OCXO is definitely a keeper, I would love to put it in my all TTL logic desktop clock, replacing the already pretty accurate TCXO. Those 52V SMPS's can be ideal for RuiDeng's new lab supplies. Please do a test of those planar transformer power converter modules.
You can also buy MLCC capacitors at 6.3V with 220uF capacity in a 1206 package.
They are very expensive at 0.50$ per capacitor in bulk but they will generally outperform the tantalum capacitors!
If you want to see Huawei kit BT uses loads of it. Also BT uses similar kit in the exchanges to buffer BT TV offerings to keep the backbone clearer of traffic.
From Wikipedia:
"Cortina is a fabless semiconductor company. It outsources all semiconductor manufacturing to merchant foundries. The company is based in Sunnyvale, California. It also has other research and development sites in Hsinchu (Taiwan), Ottawa (Canada), Raleigh (USA) and Shanghai (China).
"
Huawei is making a looooot of stuff, not only LTE/5G but wired network (dwdm, enterprise routers, switches, access points, gpon so ftth), servers, power supply devices.
.... The Australian Government prevented the company from involvement in our 5g infrastructure.
Citing security concerns as all Chinese companies are required to respond to any request from Xi Jinping.
@@robbieaussievic the Australian goverment and US are 100% uneducated idiots and everything is just a coverup lies that they spy,everything is just for money specially for us they just scared mostly for market and stupid apple cause they know Huawei is much better in quality and years ahead when they even co invented 3g and 4g. the real spy's are apple
@@robbieaussievic So? Germany or UK didnt do that. Ot course Huawei is connected to Chinese goverment, just like Cisco, Juniper are connected to USA, you can choose which country backdoor will you have.
Why do I get the feeling this was used in something for state or federal government ( not security to be fair ) and was removed precisely because it's Huawei ?
I've got a customer that I do IT consulting for who has a similar video conferencing system from Cisco that cost around a million USD to install. But it is an incredibly trick setup.
@EK89 this takes H.264 streams from video conferencing systems and combines them into an output stream that shows all the pictures or one of them based on speaker detection. It's receiving the video streams of the various sources and encodes the output for the other parties to view in realtime.
Huawei has been making routers and very high end telecom stuff like this for decades now. A lot of it is potentially stolen or reverse engineered from companies like Cisco. I know more than that, but I'd rather not talk about it in a public comment.
In a former life, I worked with and deployed several realtime mpeg4 transcodey things in the early days of ip video. Tandberg, RGB, motorola, scientific atlanta. This brings back the memories. Likely it was before Huawei was on the scene, but all built up to a price. Advancements/features, to be supported in the codecs, were so uncertain that fpga's pretty much needed to be used for transcoding. We also rolled our own picture in picture transcode service, using commodity hardware. 24 streams per U if I recall. Was a fun project in the way back times.
24 SD streams and 8 HD streams in 2 or 3U was what Moto and SciA used just about the same board layout and sold to cisco with a powerpc and not an intel. 88 grand seems to be the list price for 8 HD HDMI in/out with 2 1 Gig copper and 2 1/10Gig Fiber .
I would bet some of gear circa 2006 for the second wave of HD channels is being torn out this year. Nothing has changed in the industry, RF ingress and Drop and Insert and content management on/off a multicast seems older than rope.
The content file collection I am sure is better, no KENCAST of 20 or 50 Gig movie files and 60-400 meg commercials, just pick them up off AWS and cache. Some bastard ordered the our KENCAST servers with 1 Gig Inflow and 100 meg interface to the content store. The input was 10 times larger than the output to storeage so all humor was lost as the HD filled up. We inherted the whole thing from another telco, so worst was Our on-demand system was always out of sync, The last issue I remember was about before leaving was missing breaking bad episodes.
Now you would just put everything in elastic storage and the head ends pick up what you need at the locals minutes before being put on the channel. The on demand movies can be anywhere at any of the headends with a meshed IP network. Soon enough most cable boxes will just put whatever commercial you will respond to inserted right into a NFL game.
@@christopherwhull lol. I had to write our vod ingresser for our "super headend". I see we're kindred. Nice to meet you Chris. Lifetimes ago. Was fun stuff.
@EEVblog the spikes are for blind mating boards, they're pretty common for modular designs like this. Grounded more for safety than anything else, it's really just a mechanical feature.
The ports are angled for ease of cable management. Lets you route the fiber into cable management on the side of a server rack without needing extreme radius bends that can damage the fiber.
As retired test engineer, the testhardware boards has 8 to 12 layers, a tickness of 3.2 mm and gold finish on outsider layers. Test machine connectors ends always whit a bed of pogo-pins. You do not mention the jtag bus on this system with is full whit high pincount bga devices mostly designed in the US.
'Backbone for a video conferencing system for emergency purposes' Oh boy this is like the 007 touch slide table display stuff.
On the board with 9 chips, the center one is not like the others. Should pop that heatsink off and check underneath it.
The audio board around 24:20 has 8 chips laid out the same but the center one has different components around it.
I was surprised to see the Davinci dm8168 in there. I have spent a lot of time over the last few years working with the little brother to that part (the dm8148), and they can only be described as a hot mess. TI has effectively dropped all support for the part and there are some really nasty silicon bugs to contend with, but for a part that was released in 2010 they have an amazingly fast hardware h.264 encoder built in that is still as fast as almost anything you can buy today.
The soul of that part lives on in the Sitara AM57xx series, but hopefully with fewer silicon bugs.
One thing that strikes me: i's not very old! Well, it could be defective - but such expensive devices should run for many years. Not even 10 years old, and now in the bin. That is terrible!
True, not very old and now binned. Why it was replaced could have been for a number of reasons (most likely financial but could also be political). They could have at least tried to sell it 2nd hand and recoup some of the cost but the effort was probably going to cost more than it would have sold for 2nd hand...sad.
@@stormkhan4250 if it was a 5 eyes government agency or contractor, the getting rid of it and replaced with gear built in vietnam or sir lanka is all paid for. The NSA will send any government a unit out of their warehouse at a deep deep discount.
Those Altera FPGA's are not just "housekeeping" PLD's. Those are serious big boy FPGA's.
Used from a Canadian reseller they go for 12400 US dollars per single unit, so a rack of ten would be 124000 US Dollars plus mounting and rack hardware costs
sure, it might not be built down to a price, but you'll certainly get a deep discount depending how valuable your company's trade secrets are.
You have no clue what are you talking about and watching the news doesn't help you much either
There are circular diamond saws designed to drill into ceramic tiles. Put one in a drill press, then drill around one of those pins from the power supply, and then you can desolder that bit of pcb on there much easier.
Actually you WANT turbulent flow around heatsinks. It seems a bit paradoxical, but if you have smooth laminar flow, you end up creating a boundary layer that acts like an insulator. Turbulant flow generates vorticies that bring unheated air into contact with the heatsink. That is why the fins are serrated instad of solid.
Very similar to the approach of server blade and routing chassis from the same era. You're spot on with the comment that cost is no object. It's all about continuity of service delivery, hot swapability/live servicing, throughput and heat control in dense racked environments.
What's even better is when you design the server rooms and racks for these things, you need to worry more about BTU output and airflow than anything else. You DON'T want hot spots or no-flow zones or you get real problems.
Lots of good salvage parts here, power modules, good quality caps, top quality inductors, lots of top quality mosfets, a lifetime supply of low profile heatsinks, a home defence weapon (backplane)
Wonderful stuff! First class engineering meets first class enthusiasm - it doesn't get much better.
It's mind-boggling to consider all the coordination and man-hours that went into the design and creation of this hardware. It's like watching you disassemble a skyscraper.
I like all the added datasheets and extra info added to this video. It's better than watching your hands wave around a plastic stick while you make educated gueses about what something might be :)
I at beginning didn't like this channel but now I'm such a fan x thankyou
Excellent teardown. Really interesting. Treasure trove of devices
Those board-to-board connectors could be Samtec QSTRIP/QPAIRS or something from the Molex mezzanine line. I got to know them when I was hand soldering a few of the former brand onto test fixtures for testing just a CPU+MEM control board assy on high-end test and measurement gear. It was quite a challange for the first time!
As long as the XTZ accelerator can keep up with the QN rate when overclocked and the local MOSFET thermal reactance controllers don’t max out when power vector polarity morphs from linear non-phase reactance to non-standard decoder channel proxy configurators especially when positronic differentiation differs from radial expansion levels-I’m cool with it.
21:55 You can bend it in horizontal, but you can`t bend it in vertical. It works like a stiffener.
Please say you are taking precautions Dave. We need you
The Jianghai Chinese electrolytic caps are really good and low ESR. We use them. Jianghai used to make some of the Panasonic electrolytics
The BOM cost must indeed be enormous!
Dang, I wonder how much you could make by selling the components of that thing...
I'd happily pay a a few bucks for some of them ;)
HiSilicon develops SoCs based on ARM architecture. Though not exclusive, these SoCs see preliminary use in handheld and tablet devices of its parent company Huawei.
The optical ports are probably angled so the rack door can close without putting too tight of a bend on the fiber optic cables.
20:37 "Tight as a nun's nasty." Love it!
Liking the swift, ultra-pro editing on this mailbag. Keep it up!
Got to respect the engineers efforts to construct that piece of kit! Very cool indeed! 😎
Excellent video
24:28 HiSilicon ic which makes processors for phones and other stuff. It's mainly used in Huawei smartphones from what I see.
Notice that on the audio board, where there are 9 chips for 8 channels--the center chip does not have two little chips peeking out from under the heatsink. Some kind of controller?
Those pens on the front and underneath are also for alignment. That way when you are inserting the board or module it’s properly aligned so that pins don’t get Miss connect or vent.
PSU at 31:00 - yes, they have a custom AC input filter on its own pcb, but DOH, still need to add an "on cable" common mode ferrite to pass EMC on the AC power cable that leaves that filter pcb and jumps over to the main pcb........ ;-)
One of the power supplies is for backup if the otherone fails.
yep, still blows my mind how skookum industry/server gear is compared to commodity consumer grade stuff
how old were you when you realised your voice was going to stay like that and you would sound like Timmy mallet forever 😁😂🤣 awesome vid
"tight as a nun's nasty" haha, took me a second, never heard that one
China watched this episode before you uploaded it to youtube.
.... The Australian Government prevented the company from 'High Level' involvement in our 5g infrastructure.
Citing security concerns as all Chinese companies are required to respond to any request by Xi Jinping.
Just wanted to say thanks for your recommendations. It's my birthday this month and I got myself a "Yihua" 937D soldering station - you know, like the cheapest one you recommend worth buying which is exactly my price bracket. It replaced a shoddy plug-in adjustable-on-the-handle iron. ;-)
thanks a lot and have a nice one, Greetings from a German viewer.
I'm still not sure what this thing does. It seems way too complicated (and expensive) to be a video transcoder.
I subbed to lbry but will stick with youtube mostly because it seems to take hours for videos to show there. Plus after looking around, there is not really much else there (I'm mostly a lets play / minecraft + tech video viewer).
Oh no. You mean you have to wait hours for a video?!? That's completely unacceptable. What could you possibly do in those hours while you're waiting?
Almost as bad as the people that complain when a video is uploaded during the time they're sleeping. Like it will disappear by the time they wake up. LOL.