Steve is correct! And the odds are there are other factors and subtleties that make such a comparison of questionable value anyway. But unless they let me benchmark it... :-)

The Z16 has plenty of local DRAM. That robot around 14m is inserting DDIMM (not DIMM) modules. They work with a transactional interface called OpenCAPI (aka OMI in a simplified version) that runs around 32GB/s full duplex compared to DDR4 25GB half duplex, and the Z16 chip has a lot of those channels (they are lower power and take up half the silicon on the CPU chip of a DDR4 channel). So that bandwidth is additional to the distributed "virtual L4" comprised of idle borrowed L2 capacity. As the L2 cache is SRAM it is likely significantly lower latency than DRAM even if it is in another CPU.

How does the Apple MacBook Pro M3 max compare

@@willd0gis that satire?

@@rolux4853 you got me. What’s up with using windows XP (embedded). I was told the world ran on Linux!

How do you get a job designing microchips ? just go for Electrical Engineering course ?

​@@monad_tcp ...It's more about electronics than electricals. The specific _course_ would probably be computer engineering. Let's wait for the original commenter to offer further advice.

@@monad_tcpCheck out Google sometime. They can answer inane questions like this. 45 seconds: Computer Hardware Engineering majors. Oh wow, look at that.

Total CHIP chad :)

Any chance of using CAPI / Open Interfaces?

I started on an IBM 1401 in high school then an IBM S/360-91 at UCLA. Joined IBM as a Systems Engineer. I installed 3033s, 3084s and 3090s, etc. It was an amazing journey!

The 360-85 was years ahead of it time. I assisted with hosting several corporate benchmark sessions at the Kingston plant in 1969 before it was withdrawn from the market because of its speed. Years later I had a occasion to work on a 3033 and realized it was just a 360-85 with faster logic chips, solid state memory and a CRT console. Like others here, I have worked on 1401 1410 1460 360 370 and 4300 systems. Plus 20 years on PCs from about 1990 to 2011. Now running Proxmox and Truenas in my retirement lab. Lots of GREAT memories!!!

Thanks for taking us ALL behind the scenes!

2024 is 60th anniversary of the IBM System/360.

I haven't touched a mainframe in a little over 20 years, but.... _"when it just HAS to work: mainframe"_

@@jovetj Because I got an earworm from that, I need to share it with you. I'm standing in your server room 'Cos I got kicked while playing Doom The lights are dim but I can see Your hardware is so quality I need to apply the patch So I can get back in the match But something's caught my eye A mainframe reaching to the sky

Did you catch the brief comment that they can upgrade the mainframe firmware/software in place without shutting down?!

hes an under utilized resource the universe hasnt tapped yet

@@AllenCavedo Yes. The system I worked on was a small business server, and I restarted it weekly, but if something went wrong with it we usually wouldn't even know, IBM would just show up _out of the blue_ to fix.

People have been predicting the death of the mainframe for a long time-3 decades, at least. It's not coming anytime soon. It's just too easy for people to poke fun at or write off what they do not understand.

​@@jovetj the problem is it's very expensive to move out. It's not a question of benefits, but expenses and risks 😅

​@@MaximZemlyanoyabout half of mainframe workload is now Linux. And much of the z/OS workload is now Java. It's time for this "it's top expensive to move" meme to go away

@@briansomething5987Linux is good, Java is 💩, Apache should use Zig and Bun, no need for a Dinosaur, Linux should be done on Zig for even more Performance

if only it were that simple lol@@briansomething5987

HOW MUCH SMOKE DID YOU SEE?????

@lisandro-lopez would like to know more about it from you , i wanted to start my career in Mainframe !!

Very cool!

What language? Legit question - would take a stab and say COBOL perhaps?

@@willd0g Assembler I expect. Operating System components were written in Assembler in my day, and IBM used PL/S a lot, and later PL/X. Both could GENERATE(inline assembler code).

why do you /ibm use win xp

@@maxpowerlive8852 Its for a machine. Empedded XP. This thing works and does one job and only one job. Why change it? Chances are, this thing doesn't even have a network connection - and IF it has, surely no Internet connection.

Now you're looking at 7 users with 4 tabs each on Chrome. How far we've come 😄😄

We had some baby 4300's--41, 61, 81. A 4341 could easily support 30 or 40 users in VM/CMS 370 without a sweat. We even bonded two 81's together with one as a "real system," and the other as a "test dummy," running loads and having the 2nd frame testing responses. Basically debugging one mainframe from another! LOL Our "big iron" was 3033s, 3081s, 3084s (replaced by 3090s). Cool toys! And we even have a baby 9370. I miss those days and the cold DC room.

Six sounds pretty few, we had 70 or so on a System/370 Model 168. Under VS2 Release one, all TSO users shared a single address space and swapping was painful (3330, 3350 DASD). When we got OS/VS2 (MVS), everyone had their own and things were fine. In Australia, the 43XX computers had special firmware available to allow them to run Fujitsu operating systems, OSIV/F4 and OSIV/X8.

@@oneeyedphotographer Interesting. Hadn't heard of OSIV

Company I worked at the 4381 was the local branch machine, anything really needing processing went to the remote 3090/600S. The VAX/780 was used for DTF and ascii/ebcdic conversions. Now I'm all nostalgic about loading tapes/carts. lmao

Better than hoarding

In my days as applications COBOL programmer, I used a 370-158.

32 bit 2 um is the same memory bandwidth as on a C64 from 1982. 512kB is as much RAM as in Atari ST from 1984 .

360? I think my dad worked on soviet clone of 360 when he was in university.

I started on a 360-30. IBM used to supply a Systems Engineer along with the machine back then. Lots has changed since then. It was quite an interesting field to be in.

Looks like they even changed their dress code. Interviewed with them back in the 90's and was rejected because my suit jacket wasn't the right color.

If you have access to the source code of Assembler H and scan for my initials, JCS, you might find evidence I was there. I also did work on PL/I and LE/370, but that was mostly testing and I didn't get to make my mark.

Yes. Mainframes are awesome, and they have been for 53+ years. IBM hasn't always been that awesome, but they have sure kept their eyes on the ball as far as their core product.

@@marred2277 Maybe that is one of the reasons they completely lost their dominance? Wrong focus, trivialities, not adapting?

@@marred2277 Yeah I interviewed a while back and I remember feeling it wasn't really up to snuff with the more modern companies, perhaps "progressive" is the word. In any event I wasn't impressed.

A Burroughs guy! Hello. The first computer I ever used was a B5700 at my uni in 1972, I loved that machine. My degree was Applied Math and Computer Science and my career remained in the Software Development area.

I started on Unisys A Series then operated B2900s. Your story looks similar to mine (see my top level comment).

@@rwh777 yes, I’m definitely an old Burroughs dude. In fact, my first exposure to a Burroughs was in my High School Data Processing class, where I wrote, compiled and ran my first ever COBOL program on a B80-31/131. That was 1979 and some years later, my lead op from work brought an old B80 to my parents house where we attempted to re-wire/resurrect it from the grave (it originated from a surplus auction that my supervisor’s friend/neighbor had procured). Long story short, it didn’t pass its IPL and the external hard drive (about the size a clothes dryer) made a horrendous grinding noise. Oh, it also nearly blew the entire circuit box for my folks house. Haha, those were some great times. No Internet, no cellphones and the only personal computers were the ones you either built from Heathkit or spent a small fortune on an IBM.

My first job interview was with Burroughs in 1981 where I had was shown a mechanical calculator and was required to explain how it worked. Thankfully I didn't get the job and was employed by IBM a month later. Great company to work but sadly they totally stuffed up the PC market opportunity in the early 90's.

I joined Burroughs in 1970 in the UK, starting on L/TC Assembler. By 1972 I was doing systems architecture and project leading/management work and the actual machine didn't matter all that much. I was working at the Burroughs World HQ in Detroit on and off from 1972 to 1975 and emigrated to Australia in 1976 working for Computer Sciences Corporation. I found myself working with B7700, IBM 370/168 Univac 1100, and a bunch of other stuff. I worked for Digital in the late 1980s. It's been fun.

It's no coincidence that this is being done in close proximity to Wall Street.

well, you look at 13:14 and now you know why. some things need to be really really reliable. what happens to a normal PC data-center when there's such an event ? the entire data-center needs to fail-over to another entire data-center. You can't just keep working because of annoyances like earth-quakes.

One trend I'm noticing in these comments is a lot of "was" and "now retired" people. What are the prospects of these systems going to look like in the near future when the next generation of techies coming up won't know F all about languages like COBOL, etc.? Who will maintain these applications in the future?

DONT get all excited they moved most manufacturing to Mexico I think

You are so welcome!

But why is Dave never in the garage?

@@knerduno5942 Because after Dave started making more money on his UKposts channel than his wife explaining these things we lie awake at night wishing we understood or knew someone in the business to keep us old farts up to date he is not no longer confined to the garage.

You working for blackrock?

@@mephInc nope, black rock owns damn near everything. But we work with black rock and at our annual RIA conference that my company puts on (and I owned and engineered the technology for the event for 7 years) over 800 desktops, 400 laptops, 2000 monitors, two racks of full mobile servers, blades, SAN, networking in a rack. Where in the rack we would run around 30,000 virtual desktops that would build on demand and deploy for items and destroy after. At any time, 3,000 vdi desktops running, about 30 virtual servers running the back end systems and network capacity of 4, 10Gbps fiber links fully redundant running from 4 different isps, and setup a funny NOC/SOC onsite, and lots more. Now I'm out and moved on from that stress, but they run a lot in cloud now that we've started some implementation into that area. But we are the largest financial institution by assets, as in customer money in accounts. With our combined Acquisition we made 2 years ago and just on the end of finishing integration, nearly 11 trillion under our management. Official is 8 trill because of they are not finished moved over yet and the old name we purchased to go bye bye in history. Been with this company since 2007 when I was 23. Started as a contractor doing 21" CRT to dial 15" LCD monitors in two buildings here of around 9000 employees. And moved to support, then took on engineering and built and designed our VDI deployment of 12000 virtual desktops for offshore and retired our old HP blade computers (like a rack mounted laptop where a chassis had like 25 of them. And moved on to desktop engineering and then to security when we formed our new security org, then to various parts where I owned the products and built security policies and procedures and deployed a full rdp/ssh/admin (secondary) account vaulting and session management, recording, keystroke logging, monitoring and review, etc... and then to lead architect. Literally started as the lowest on the tottem pole and in my 16 years there, moved up to sr manager and lead solutions architect of some of our more sensitive security infrastructure. Lol, and since day one... mainframes will be out in the next 2 -3 years, we are retiring them! But nope. There is just things that with active customer trading, world wide bank transactions, batch processing, and much much more, that a mainframe can do in a micro second with 100% accuracy that a computer or server can't. And when a milisecond is the difference of tens of thousands of dollars, they are worth their ungodly price tag. (Millions spent yearly on them for support and maintenance and IBM has staff onsite that their job is to sit and if something breaks, for it right away. And a new unit is multi millions when you factor in power, cooling, data center floor space, setup, configuration, migrations, fibers, interconnects, etc... it's mind boggelinging

@@MikeHarris1984 Nice to see you work your way to the top. Something most kids today can't comprehend lol.

Same at my financial comoany. MF is a dirty word. I think its cultural. The new sexy buzzwords get the attention, but all trades still clearing via the MF

Fascinating to me as a Career Microcomputer Architecture guy! I would love to see such things!

I remember a 1990s news story buried by most news services, US Navy outfitted a ship entirely with Windows XP, in control of everything. Maiden voyage and all computers blue screened leaving the ship dead in the water, they had to tow it back to port. Fortunately it wasn't in war time.

@@jrstf XP was released in 2H 2001. If you're referring to Windows 98 (first edition), that would be believable because it crashed with a BSOD during the full press demo of plug-and-play.

@@jrstf If I remember correctly this fiasco was caused by some early Windows NT. An incorrect manual data entry on a keyboard caused the complete ship to shut down and it had to be towed back to port. It was not possible to restart the engines.

If my deteriorating memory is right then a university professor found the XP system would crash at intervals because a 32 bit counter overflowed(?) I think it was about 7 days.

@@jrstf"Buried by most news services..." or, as an intelligent person would say, "not reported because it's not exactly newsworthy to the vast majority of people." But sure, it was "buried".

PWRDWNSYS😊and PWRONTIME(*SAME) PWROFFTIME(*SAME) I remember those good times😊

After decades big blue confirm great products with great engineering

@@neriozulberti1492 Yep. Also look at IBM Northpole AI/ML chip. 6x faster than NVidia A100/H100(new) whilst being more efficient.

> How much maintenance is required for one of these? Not a lot, they basically very rarely go wrong, and when something does it can be replaced in flight generally.

Sign and magnitude binary numbers, programmed in ESPOL.

In the late 70s I had access to a System/370 model 168MP. MP meant two full CPUs configured as a single computer. One evening I dropped in, and after a short conversation it was agreed I can have my own computer for a few hours. Without interrupting the workload, I showed an operator how to separate out a CPU, some RAM, a channel, some DASD, a 3270 controller so I could have a console for doing stuff. Then got out my 3330 disk pack and ran OS/VS1 for a few hours, doing a SYSGEN for our computers. When it was done, I showed an operator how to put it all back together. Back then a CPU alone occupied a whole box, and there was another to connect the two CPUs, and each channel also had an equally large box. I had to read the numbers on the sides to see what was what.

One thing Dave did not really cover (maybe because he didn't work on them himself) is that mainframes are a whole other world of computing. Sometimes you just have to forget what you think you know. Many of the same terms are used, but in different ways than you might expect. Partitioning is a key aspect of mainframe architecture. The bare-metal partitioning mentioned is called an LPAR. I'm sure you can find more details about that by searching for that term. A mainframe operating system can then soft-partition available memory to different tasks. This essentially contains each process. Each program running at any given time runs in its own [memory] partition. In the frame cabinet, behind the front door, there is one or two laptops that are affixed to the frame to be opened up and used. This is called the Support Element and is the lowest-level access to the complete machine's operation and status. Satellite computer(s) outside of the frame called the Hardware Management Console can do many of the things the SE can, but over the years the distinction has gotten less and less. If you wanted to turn on or reboot or partition the mainframe, you would use the HMC or SE. As mentioned, when failures happen, a hot backup instantly kicks into place, and IBM is automatically notified so a repair will be scheduled. Edit to add to my comment: The zSeries and z- mainframes are a complete 64-bit architecture. They also started out being fully backward-compatible even for old software written in the 60s. Some of that very-older comparability has been removed or minimized in newer z- models, but backward comparability is a huge assurance to customers.

Well i have worked for a long time on the software implementation end for IBMs os and other servers and know that they configure storage devices and use mount points directly to deploy oses, they are tricky to configure, I have not configured personally but I have worked on these mount points or something they called luns to setup our softwares, they even have replication directly from the storage itself

Get the Hercules (softlabs) emulator and you have a S/370, S/390 and Z-ish mainframe enough to run even z/OS and z/VM. Then go on a quest and hunt down the dasd images of whatever OS/390 or z/OS you can find. Then hunt down the documentation. Start reading the MVS and JCL book from Ranade publishing. Be prepared to read the IBM manual "Principles of Operation". This manual started in the sixties so start with the manual for S/360, that most amazing family of machines. You can also get started right away with a public MVS 3.8j distribution that is actively maintained by the community. Now you have enough to get started, you could start learning about JCL in about an hour and submit your first assembly job to MVS $HASP today.

The redundancy around the Storage is quite impressive. You can lose a disk array in one data centre and its replicating counterpart in another data centre can take over the workload without it impacting on normal IO on the mainframe... Just an increase in IO read write time due to the change in distance when going cross site.

I'm a current systems programmer, we run about 50/50 z/OS and RHEL. Most of the stuff you are familiar with is probably still around, just now you also have web servers and it's sort of slowly evolved into more of a unix type system. MVS is still what people call it, IBM uses that term now to refer to the non-unix part of the OS. So ya know everything with datasets instead of files basically. So z/OS is made up of z/OS Unix and MVS. It's still a very customizable OS but I think people have probably mostly moved towards the point of regretting some of the crazier customizations at this point as giant kernel level assembler programs become very difficult to maintain especially when you had the switch to 64bit and those sorts of challenges. Still plenty of retired systems programmer code out there in the wild causing trouble haha. Linux is Linux of course, the majority of shops are probably running most of their Linux on top of z/VM or possibly Linux on top of z/VM and then more Linux's on top of that KVM. I guess the only other thing to say is that ISPF is still probably the most popular interface on the MVS side (that's the green screen except it has more colors now for all you non mainframe folks), but you have more options now if you want them and can have a command line and vim workflow or a remote IDE.

Microcode, the MCL (on the SE) loads microcode (in POR) that creates the environment for the Image (LPAR). Your Image then has an environment with channel access to storage to load an OS from (IPL (Initial Program Load)). That's called IML (Initial Microcode Load), and until you've done that the Image environments don't exist. Once it's been done IPL for the specific Image environment becomes available. MVS has been 'rebadged' as is the current trend, to z/OS, it still looks and feels like MVS, so you'd recognise it. It's also had a whole lot of modification to make it more Java friendly and cloud capable. z/VM is the micro partitioning OS Dave mentioned, and is still quite common.

Yes that's a good point. It's still true that most business computing is I/O intensive rather than compute intensive. The architectural decisions in favor of better I/O is what generally makes mainframes feel snappier to users than giant clusters of cloud compute that should be much more powerful on paper (not that mainframes are slouches at compute, as far as single threaded workloads go they are probably the most powerful single core on the market as long as you ignore floating point, which is basically useless to businesses anyway). The cloud is like a distributed super computer when what business computing really is looking for is a distributed mainframe. But you can't have a distributed mainframe because network latency is just going to kill the snappy response no matter how many nodes you've got. That's why apps feel slower today than they did on a mainframe from the 80s and 90s with like 1/1000th the compute. The mainframe is actually one computer, with terminals connected and part of it. The network is a LAN not the internet. You cannot overstate that difference. It's also true that modern user interface design is complete trash and skilled analysts and accountants and the other people that may not be computer people but get really fast with the terminal based mainframe apps know it. Those folks really hate it when you come in with a flashy new web app that you think is doing them a favor by being less intimidating. It's totally disrespectful to people in other professions to treat them like children as soon as they need to use a computer. That's pretty much all they ever get with other platforms and with the way sysadmins tend to treat windows and the way webapps are inherently limited by network latency.

The die size is HUGE

Think about the absurdly low yield IBM has on those chips and how expensive they must cost. Low yield means a lot of refused chips when manufacturing because of defects, it must be absolutely perfect, which is very hard for a chip that big. They probably don't even do binning.

Different use cases, one is not necessarily better than the other. Epyc has smaller less powerful cores but more of them. Mainframes are designed intentionally for single threaded workloads and a lot of the I/O offload is specifically so that you can prevent interrupts from wrecking you with context switching. I'm also not totally convinced that 3D memory is ever going to be the correct way to go but that's a very long and complicated discussion. The mainframe would wreck Epyc in transaction processing but would get crushed for stuff like multimedia encode/decode and video games. But yeah, the Epyc you mention is also a newer design so the feature size on the chip is smaller.

@@monad_tcp That's why they have spares.

@@monad_tcp maybe they disable failed cores or cache segments?

I remember Microsoft claiming five nines. I made my reasoned rebuttal to the Trade Practices COmmission (Australia). The TPC accepted the MS explanation, but those ads ceased. My rebuttal was simple, take away minimal maintenance time for Windows (NT), there just wasn't enough time left. It's not better with Windows 11. Linux, I reboot to change kernels. Everything else I upgrade, but choose whether to restart, individually.

I started 20 years earlier. I recently spent a day scanning the latest equivalents to the manuals I used in the 70s. The assembler code I wrote then would still work today, i was going to say except for the loss of remote 3270 terminals, but I suspect those would still work. Provided I could cook up the appropriate (emulated) 270x controller. I had a 3704 back then.

Very nice! It's been about 20 years since I've touched a mainframe, but I was pretty much hooked.

My Burroughs B5500 mainframe (circa 1968) now runs as an emulator in my browser. It supports 8 tape drives, multiple card readers and printers, card punches, datacomms and head-per-track disk. It supports full virtualization of memory and can easily run half-a-dozen ALGOL and COBOL jobs simultaneously in 32K words.

@@pquirk99 48 bit words. The real thing didn't cope well with hordes of students trying to save their work in under 20 minutes.

One TB. Barely enough to store my photos for 2022.

I bet it looks considerably less attractive when the bill comes

@@TheGreatAtario They start at about $2.5 million and head on up. But if you consider the salaries of elite developers and sysops plus the hosting costs of a large cloud infrastructure plus the reputational costs of lost data, it obviously makes sense in a lot of scenarios. Imagine if you are a bank. Do you really trust your core account data to some exotically complex sharded distributed setup? Or do you prefer to stick it on a couple of massively redundant replicated super-computers? Again, if you are a high frequency trader, using dozens of GPUs and striving for microsecond latency. As we've seen with real examples, the cost of a glitch can run to $billions and break the company. You're surely happy to pay for the fastest, simplest, most robust solution...

@@tullochgorum6323 "Exotically complex sharded distributed setup"s come off the shelf from cloud providers and have for quite some time now. But it is indeed true that bankers prefer to throw money at problems

Yup, came to say the same myself.

he knows. its just to invite comments..

ELI5

its almost as if they're doing all the heavy lifting of the entire operating system via hardware. (I know they are, this is rhetorical)

That's true and one of the reasons it can be cheaper for a lot of businesses to pay such a huge premium for the hardware and licensing. If you go the distributed route but want the same snappy feeling and feature set and have everything replicated and in one place, you are looking at paying a shitload of engineers to maintain a very complicated Kubernetes setup and a bunch of apps to handle that complexity and make it reliable. That could all be provided by a couple of systems programmers with a mainframe and the operating systems, and then more resources can be devoted to defining business processes in custom apps specific to your business. It's a very different model than what you find in flashy high tech companies, but I think it may be fundamentally superior for a lot of industries.

So what you are describing is a SYSPLEX, a cluster of mainframe images (LPARs) running as a single system (the 50% split gave it away). The geolocation diverse line might make yours a GEOPLEX, so somewhere on your mainframe complex there will be a CF (Coupling Facility) that handles the Image to Image communications between the two physical systems (it might just be in an LPAR on one of the processors).

There is a browser available that can get around it. I can't name it directly as the comment will be auto deleted. However, its name is the opposite of "coward". It will fix your problem. I've been using it for over 5 years .

Hint: it would be a _brave_ choice. 😊

That is one of the most important features, and also one of the most important differences between mainframes and micros. The US government has a huge set of operating guidelines that must be met by the bank's IT departments, and I'm out of date on that stuff, but those generic CPUSs probably don't pass those guidelines for certain transaction processing.

I have been in an IBM Z team for a little over a year now as a sysprog. It was pretty foreign to me as I am only 23 and had only done a little bit of mainframes at university. I wish there was more young talent coming into mainframes but we're trying!

@@MitchMax5 ill come join. I jokingly tell people at work I would love to live in the terminal. Engineering man makes it look fun also

IBM continues to develop and build mainframes because there are customers who will pay for mainframe capabilities. If there were less-expensive solutions that worked for customers, customers would buy them instead.

@@danpatterson8009 There are less expensive solution, they are just not nearly as fast and powerful

@@willd0g My team is in australia. The engineering and sales of mainframes isn't really done here

It might be a licensing issue (like you pay for activated RAM and can upgrade later) or for redundancy. That way if a DIMM fails it can just swap in some of the spare 10TB.

I once had to deal with am IBM systems programming manager in the who was obsessed with conserving CPU cycles.