IBM z9 mainframe computer teardown. All components explained. Technical description: Google for "IBM z9 Redbook". Look at the book for the Enterprise Class model. it is more detailed.
You know something's over-engineered when a fan is called an "air movement device" -- I'll remember that the next time I get in my fractional petroleum distillate powered latitudinal movement conveyance device.
Great video! You don't get a chance to tear down such a beast every day ;) Actually, the Service Elements (the two ThinkPads in the front) are not only used for diagnostic purposes - they run a big chunk of the machine firmware as well. Yes - the books are indeed called like that because they are mounted like books in a shelf. Like another commenter already mentioned, since the zBC12 and z13 EC there are no more books and MCMs but rather drawers (guess why they're called like that ;) and SCMs to save costs. Although each SCM combines several of the chips on that z9 MCM on a bigger die. All components on the MCM are soldered BGAs (IBM almost exclusively uses flip chip packaging). The ESCON and FICON ports are used to connect to external peripherals like storage servers. ESCON is IBM's proprietary optical I/O connector (actually, it's standardized in the T11 SBCON standard). FICON is the successor to ESCON - this time based on the FibreChannel physical layer (and standardized in the T11 standards FC-SB-2 to FC-SB-4). Actually, all of the I/O cards have their own PowerPC processor handling the attachment to the mainframe channel subsystem (the mainframe uses channel I/O instead of port or memory mapped I/O).
ESCON is the older half duplex fiber interconnect for these. You could run ESCON up to 43KM! It had a speed from 10Mbit to 17Mbit before it was replaced by FICON. FICON could run up to 100KM at a speed of 1 to 16Gbit!
Thanks for taking the time to make these videos, it is really fascinating to see how machines in this price class are engineered (VERY well is how :) )
Me too. I'd like these to be more reusable, as "built to last", so modular design, to use parts of it in other uses... same way as people invent new use cases for washing machine engine, car parts and so on. I'd imagine there are use cases for reliable but low performance heavy in eg. non profit scene, but expertise and knowledge is needed. Otherwise closed system sell for couple of bucks in ebay, as those are so nonpractical, only for most enthusiast hobbyists.
Thanks for the video! It was very interesting. The strange decoupling capacitors you mentioned on the ceramic CPU are very similar to the AVX LICA series - basically it's a very low inductance decoupling capacitor array.
Very nice presentation and excellent explanation Thank you very much for taking the time and exerting the efforts to produce this video Keep up the great work 👍🏻
These machines blow my mind. Just watching this video just runs through my head the brilliance of the team of engineers that built this mainframe. From design of the case to the layout of the individual components to the software engineering. I should of stuck with computer science. I missed the boat.
+mikeselectricstuff On the CPU board itself are only caps and connectors. The DC/DC converters for the CPU are on the backside of the system. I will do another video with one of these. They have some interesting specs... for example 1V @ 800A!
Awesome video and it was immensely interesting. Saying that, it was like nails scratching on a black board watching you rip that poor thing [of beauty] apart.
4.5B transistors.. just as a comparison. Apple's previous generation SoC used in iPhones(A8) has a 2B transistor chip. The new A9x is believed to have 3+. Nvidia's GP100 GPU has 15.3 billion.
If you look at technology advances from decade to decade, such as 40s to 50s or 50s to 60s, there hasn't been much advancement from the 2000s to the 2010s. Seems like it's just a marketing game now. When's the last time you saw a processor spec advertised in a compute ad? The ads now are simply "Laptop! $229.99" with very little information. Another example are the processor model numbers. No indication of speed is suggested.
One reason is because there is a finite speed that the chips can function at, or rather it gets seriously difficult to get ever greater clock speeds out of computers. Also functionally there is enough computing power in something fairly basic that you can do what average people do with a computer at any clock speed and you don't need to talk about speed any more when mass advertising a computer on TV. Only enthusiasts are excited by GHz numbers anymore and almost no one cares about transistor count in their daily life. I'm typing this comment on a 2011 Lenovo ThinkPad because it does what I need, it didn't cost me much and it is indestructible.
i bet my mac book pro is really indestructible compared with a lenovo , i have one mac book which is completely destroyed from the outside but the computer is still running .
Thankyou very much for taking this apart and sharing it. It's a shame to see it go but amazing to see it up close, I hope to add a fully functional one to the rest of my collection one day. If you have time, could you please take and upload a high resolution photo of the MCM (Processor)? Set on a black background and lit just right, I'd love to have that as my wallpaper.
They call PCBs 'planars'... This thing would had been connected to synthetic AC power, and never directly to the grid. Very interesting that they have separate modules for the system clock. I wonder if the CPU microcode still has to be loaded from the Support Element on powerup. What replaced this machine?
10:19 Those red rings by the fuses are probably current-sensing magnetic field coils. They measure the current flowing through the large flat metal conductor passing through them.
31:05 "i hope you can see that" dude i can see the writing on those dies, your camera is badass. are you thinking of uploading at 4k in the future ? your videos and their macro shots could really benefit from it.
Wow. It wasn't that long ago (About two employers ago) I was working on Z9 and Z10 mainframes, FICON storage and big honking 3584 tape libraries. It's not very glamorous compared to VMware server farms, but it has a hell of a pedigree.
IBM has gone away from this type of CP book construction to make the units easier to manufacture. Each CP is now a separate unit which plugs into a tray; no more will you see MCMs in IBM mainframes. You'd be surprised how much manual work goes into those multi-chip modules; they definitely did assemble them by hand! They had a special team at the Fishkill semiconductor plant to assemble them; there was no way to do it reliably with machines. Of course they had machines to assemble the chips onto the modules, but everything else had to be done manually.
gee, I remember writing software on am IBM-360/30 with 48k yes KILOBYTES of main storage, (3) huge 5 Megabytes of disk, and (8) 9 track tape drives, card reader/punch unit, and a 1403 printer. I used to run BPS and DOS 26.2. To watch a tape sort was amazing. with DOS you could run 2 jobs at once. FG and BG. of course all the software written was in assembler. I loved it
Your vids are fascinating. If you know, it would be interesing to understand what processing tasks these machines were doing? Banking for this one maybe?
yea i hate to see such unique systems going to scrap yards... why i dis-liked the video. stuff should like this needs to go onto ebay for ppl who like to collect this sort of thing, the biggest thing in my collection so far is a sun microsystems sunfire v880..
How might one get ahold of these machines? If they're only worth throwing away how might someone go about getting their hands on the outdated equipment?
You have to know people who work with these machines. One day they will be replaced and then you have to catch the right moment. A bit of luck is always necessary.
What are those work loads that can't be off loaded onto a server cluster? Just curious, my assumption would be VoIP or a massive massive storage cluster.
we have a lot of those older machines running in the IBM data-center in Brazil, we still use some even older S class lol. This is wroth a lot of money here...
Thank you for the teardown. Man, this thing is extremely over engineered. It seems mainly just due to IBM wanting it to be too complicated for non-IBM employees to deal with it.
So many products are over engineered. Desktops and notebooks too especially from 2000-2015+. Only recently we've seen some sort of simplification of design, even then the jury would still be out
It's about reliability. Banks and airlines for example, want systems that perform very well, but more importantly, they want systems that never, ever, break. For IBM to produce that sort of quality, and ensure that it's delivered to every single customer, they need to engineer the shit out of it.
I wonder what the TDP of that monster CPU chip could have been? I like the idea of the replaceable socket as well. Odd construction of the individual pins though. I also would have thought each one would have been spring loaded pogo pins.
I don't have data for the z9, but I do for the most recent machine: the z13. IBM has moved away from Multi-Chip Modules and towards Single Chip Modules. In a z13 EC there are a maximum of 4 drawers, each having 6 Central Processor SCMs and 2 System Controller SCMs. Upto eight 2-way multithreaded cores are active on each CP SCM. The TDP of each CP SCM is just under 300W, with the SC SCMs rated at about 80W each. Going back to the z9, a fully-loaded box is rated to consume just over 18kW.
I would have thought as much. Most of everything you need for a system is practically on one die now. However I still think these older designs are a lot more interesting to look at. There's just more that you can see.
Ca 1985 the had a Data Migration Facility. It came in a nice box full of quality control stamps and a manual. Turned out to be a null modem cable to connect 2 PC together. But way to ask over 100 US$ for a piece of cable and two 25-D connectors. When I offered to just solder a cable from parts so we could get on with the job, the customer refused: "we only connect genuine IBM parts...."
I'd guess there was also some software in the package? Or only a cable? That reminds me when we had a problem with a SCSI device. A DEC technician came to fix it and complained about the "non-DEC" terminator. So I quickly changed it to an original one an the problem was still there. Now he started working on the device...
Good that the PGA spring pins are a separate unit ... a bend pin is already a nightmare on a consumer Intel motherboard , but this would be real DOOM ...
Does your work require you to scrap it rather than sell it? I know that it's too old to actually be useful in production, but I reckon something like this would sell pretty quickly if you stuck it on eBay for a token amount of money. Someone crazy like me would buy it.
The problem is this... people want stuff like that but when they realize of transportation cost, they suddenly lose interest. I'm in Switzerland and shipping such a beast to USA will cost a couple of 1000's of $$$. I once sold a similar system fully working and then the buyer came with questions of all kind.... lot of work for me... not worth the few bucks you can get.
Sad to see a awesome thing like this going to the scrapyard... but companies do that alott, and selling is either dificult, more expensive and/or the data should be destroyed anyway. still sad to see this die
Gold and copper recovery and other materials probably give back more money. But some people like old hardware just to run and collect them, even pay more than scrap metal prices. So yes, old hardware might be not cost effective to keep arround but dont just trow everything away, keep the collector in mind, maybe sell 1 of many servers to some collector and rest for pure metal recovery. Yeah, old hardware waste more power then they give processing power, so replacing them is most of the time smart.
Agreed! More people should think about at least looking for collectors who might want stuff before disposing of it. Speaking of big, special-purpose hardware, I just saved the rolling stock from a 1980's-vintage monorail system from the scrap heap, and also secured a bit of beam on which to run it. I was literally bidding against scrap dealers for it, and yes, I outbid them and got the trains for an absolute fraction of their original cost. (Party at my place, eventually, with one of the best backyard railroads in the world!)
The time needed to find someone who can take that thing away is money spent by the company who bought this unit to do something with it. They by now don't need it anymore, and getting it taken away ASAP is their preferred way. What happens afterwards is not much of their interrest. They are also not interested in getting US$1000 for this as booking this value is more trouble than worth it. Where I work we donate a lot of reusable things. Sadly that's mostly notebooks and desktops. Servers are not wanted. So those are scrapped.
Harald, yup, I've noticed that that's how it works. That's how I got the monorail. The business view is "time is money, just get it out of the way so that we can get on with our plans." That doesn't change that there is value for the community in a better attempt at recycling.
In two years no one has answered your question. Mainframes are no different to modern computers, they're just scaled up in performance and capacity and are based on a distributive architecture which allows it to be used for time sharing applications. Many years ago, mainframes were more relevant as an identity to high performance computing. These days hardware has become more accessible and it's possible to build your own cluster of systems and perform distributed workloads that way. Whether you call it a mainframe or not is up for debate.
Sometimes it's hard to see if a screw is hex or torx... Do you know why you get a general anestetic during surgery? So you cannot see their mistakes... :-)
Processing a very large amount of data at one time; they were and are used for things like airline ticketing...and never crashing in 10-20 years. Supercomputers are for extremely complex calculations.
Is it just me or do people buy these beasts thinking, we can upgrade/enlarge them later, but when the time comes around, it's cheaper to buy a completely new system?
Oh definitively! Computer manufacturers often advertise the "expandability" of their systems but most users buy them and leave them as they are. Around 2000 there was Hot-Swap PCI... everyone made systems with that. Until the customers found out that it is more or less useless. For most PCI cards you have to shut the system down anyway to replace.
Looks like a certain amount of over-engineering, like IBM have made it deliberately difficult to service without special tools & jigs. All those big fuses soldered into multilayer board, for example! Imagine having to remove one of those.
+ForestCat IBM has always made their stuff difficult to repair. I remember some power supplies where they used rivets instead of screws just to prevent someone to open the case. But we repaired them anyway ;-)
