Fastest CPU? PC or Mainframe?

Dave is such a character sometimes. 😆 the thing at the end about demoing an IBM mainframe was very funny to me especially when he did the call me thing.

Dave, you are talking to me. I'm 75, started coding on an IBM 650 and I still love to code. I worked in disc drive and IC development and I'm flabergasted by the development I've seen. Love your presentation and research and your memories.

Hi, I own a IBM Mainframe from late 1990. Actually my boss gifted it to me when I started my apprenticeship. He even gave me the proper terminal, physical keys to unlock the beast and 17 drives (almost the size of my head). What a great guy.

Hi Dave, a fan of your channel and an IBM mainframe guy here - maybe you already know, but just to make this clear: the biggest difference between a PC and a mainframe in terms of performance is not the processor speed, but actually the I/O throughput and efficiency... so while the processor speed alone in terms of instructions per second may be just a small multiple of a high-core count desktop CPU, it will be able to do the work of a few hundred (at least) such machines in terms of number of transactions processed - that's where one of its main strength lie, besides security, redundancy, etc.

My work had a “minicomputer”, engineers had an account and our department was billed according to the amount of cpu time we used. Our lead developer was doing multi variable control system simulations and was our biggest user but he had was nothing compared to the one user who was using orders of magnitude more cpu time. This user was a managers secretary and was using the electronic office.

My first actual programming job was using Fortran (WATIV) on an IBM 360. I feel so old. Thanks for the trip down memory lane, Dave ;)

There's something genuinely cool about mainframes, all that power in one package makes it feel like "the ultimate computer". If they didn't cost an arm and a leg to own and run, I'd own a few.

Great topic. Back in the 80's I programmed Ratheon and Vax 11/750 -& 780 systems in assembly. My first PC was a Heathkit and I programmed it in assembly to communicate serially with the Ratheon. With it I was able to replace punch cards for inputs with ASCII text files. It was so cool.

08:42 - 1951 Univac 1 (2.25MHz) : 0.002 MIPS 09:12 - 1961 IBM 7030 (?MHz) : 1.2 MIPS 10:02 - 1965 CDC 6600 (10MHz) : 10 MIPS 15:27 - 1969 AGC (2.048MHz) : 0.04 MIPS 09:40 - 1975 MOS 6502 (1MHz) : 0.45 MIPS 09:33 - 1977 DEC VAX 11/780 (5MHz) : 1 MIPS 10:09 - 1988 Motorola 68020 (16MHz) : 10 MIPS 10:19 - 1991 Intel 860 (50MHz) : 50 MIPS 10:15 - 1993 Intel 486 (66MHz) : 25 MIPS 11:01 - 1994 Mips R4400 (150MHz) : 85 MIPS 11:13 - 1994 Motorola 68060 (75MHz) : 110 MIPS 11:27 - 1994 Intel Pentium (100MHz) : 188 MIPS 12:28 - 1994 PowerPC 601 (80MHz) : 157 MIPS 12:35 - 1995 PowerPC 603 (133MHz) : 188 MIPS 12:42 - 1996 PowerPC 603ev (300MHz) : 423 MIPS 12:49 - 1996 Intel Pentium Pro (200MHz) : 541 MIPS 13:33 - 1999 Intel Pentium 3 (600MHz) : 2'054 MIPS 13:58 - 2003 Intel Pentium 4 Extreme (3.2GHz) : 10'000 MIPS 14:05 - 2006 AMD Athlon FX-60 (2.6GHz) : 20'000 MIPS 14:09 - 2006 Intel Core 2 Extreme (2.6GHz) : 50'000 MIPS 12:57 - 2011 ARM Cortex A5 (800MHz) : 1'256 MIPS 14:15 - 2013 Intel i7 4770K (4GHz) : 133'000 MIPS 13:42 - 2014 ARM A53 RasPi (1.2GHz) : 5'000 MIPS 13:22 - 2014 Apple Watch S1 (520MHz) : 1'000 MIPS 15:03 - 2020 AMD 3990X Threadripper (4.35GHz) : 2'356'230 MIPS

Noble gas is a term used specifically for the group 18 elements, FYI! Halon in most fire-extinguishing applications was another name for tetrachloromethane. I figured you'd want to know, not meaning to nitpick :)

I have implemented the PrimeSieve in Lattice / SAS C-compiler on my Amiga 4000 with the Blizzard CyberStorm 2 rocking the 68060 at the base 50mhz. Have also implemented the sieve in Pascal, Comal and working on an assembler version - all on the venerable Amiga A4K, because I can. If interested, I can post numbers. I also have an A1200 with 68030, 68040 and 68060. As I am a geek of old, I’ve done the same implementations in Borland Turbo Pascal, Turbo C and IBM PC Comal on my Digital Prioris Pentium 90…. So sad, really - AND GREAT FUN! Thank you for all your content. The historic stuff tickles me the most. Regards Anders, Denmark.

You missed the Alpha CPUs introduced in 1992. It out performed MIPS (that was the whole point by DEC). The most beautiful CPU created as it has pipelining branch prediction and out of order processing. Which is now common place. I loved that CPU, and the two GS140s we had running with Tru64 and TruCluster. Clustering that was real clustering like DEC intended it. Migrating running processes to different nodes.

Dave, that "Halon abort" button is to prevent dumping Halon, not to cause it. IME most if not all machine room panic buttons are red, and must be pulled to activate. This is so an unintentional bump doesn't set it off. Pulling the panic button does three things: 1. it causes an emergency stop of the computer, 2. it activates the Halon system, and 3. it activates the building fire alarm. It's all done by one switch because better safe and simple than sorry. Mainframes typically have exotic power requirements, like 400 Hz 3-phase AC, not unlike what aircraft use. Utility power is converted to 400 Hz by a rotary converter, a motor attached to a generator. It's big and heavy, and because it spins, if you see one smoking or vibrating, shutting that down is really important. No matter what, removing power is Step 1. It's possible that just removing power will fix the problem, but if everyone runs from the building, nobody will be left to figure that out. For that reason, the fire suppression system is set by default to go off after a predetermined time to allow evacuation and reassessment. After the initial panic, the computer operator may decide to halt the automated process. Think about it, if the fire retardant is discharged, there's no more left in case of a future fire, so the computer can't be turned back on and the computer room can't be occupied until the fire equipment is refurbished and re-certified by local authorities. If there's no fire, you'd want to avoid that. No, Halon doesn't "suck the air out of the room" but it's not 100% safe to ingest either. The whole building may or may not be evacuated, depending on a large number of circumstances. In a commercial building there's someone whose job it is to make more complex decisions that automation isn't good at. That "bulletproof glass" may not be to protect the computer from armed intruders as much as it's there so it doesn't spray the computer operator or building engineer with shrapnel when they go to see if the computer really is on fire.

I always heard that the biggest difference between a mainframe and other CPUs wasn't the CPU power but the amount of I/O they could handle. I also have seen youtube presentation on the Apollo Guidance Computer that pointed out that it was a special purpose computer that had 3 processors running the code and would compare the results, so that if 1 processor came up with a different answer it would be ignored and go with the answer the other 2 gave. So kind of the 1st fault tolerant computer, way before Tandem.

I still recall upgrading the old disk cabinet on an S360 in the 1990's. The guy that recycled it got close to a grand for the scrap metal alone. There's gold in them thar drives! Swapping for a newer 360, the mainframe was so embedded into the building's power that they ended up having to shut power off to the five story office building at the street to remove and replace it. It was the equivalent of a liver transplant for a business. We lowly Netware engineers and admins were never given such luxuries or support, yet no one could connect to the S360 via our Netware SNA gateway without us. What still impresses me is how much computing access could be had with so little bandwidth. The satellite offices (where the actual products were manufactured) averaged about 200 terminals, yet they share a 19.2k tunnel.

Great video - I come from an IBM mainframe background and have worked on everything from a 1620 to the biggest 370. I learned ALC from a Green Card before there was a manual and the assembler software was on a tape. Years ago while working on a project for Royal Bank in England. I had a friend who was heavily into PC archtecture who was constantly bragging about how fast his PCs were. Finally, I took him over to a mainframe console and showed him the 256 threads running concurrently. He stopped bragging. Anyway, I finished up the project, the first PC to mainframe to PC using LU6.2 protocol and post wait logic and went home. He is still friends today.

So funny story. I worked on decommissioning a small data center running IBM Big Iron for a three letter agency as they moved to AWS GovCloud. All that hardware was only a few years old and basically got thrown out with the all of the storage (SSDs and HDDs) securely and destructively destroyed. It was really sad to see all this hardware getting thrown out and I couldn't take anything home to play around with :(

I’m 85 and started writing code in the 1960s. I’ve written code for the NCR Century 100, the GE PAC4000 series, the GE PAC30 and the PDP-11 in the late 1960s and early 1970s. I was at Motorola when Chuck Peddle lead the team that designed the Motorola 6800. He recruited a coworker of mine to write a cross assembler for the 6800 on a PDP-11. Most of my programming was done in assembly language. I now have a dozen Raspberry Pi computers and I’m learning to write Python scripts.

I had started my career in IBM Mainframe. Z14 excels in parallel processing like no other. Z14 can pause the execution while waiting for I/O and pick up any other Tasks. Z14s are multi tasking beast. While I admire the power of Big Iron, they are prohibitively expensive.

One of my all time favorite mainframe instructions is Move Character Long (MVCL). This instruction uses 2 register pairs. Each register pair contains an address and length of a memory area. One par for the source string of the move and the other for the target. Notice that the lengths are not required to be the same. This can result in either truncation when the target is shorter than the source or filling with a pad character when the target is longer than the source. One of the registers contains the pad character which is often time either a blank (x'40') or null (x'00'). This instruction runs in a virtual storage environment. It is possible that the source and targets span many pages but also the instruction itself might span across pages. The code would look like 1) LOAD R0 with the address of the source, 2) LOAD R1 with the length of the source, 3) LOAD R2 with the address of the target, 4) LOAD R3 with the length of the target. Both lengths are limited to 24 bits so this instruction cannot be used to move a string longer than 16 megabytes long. One of the length registers contains the pad character. I can no longer remember which register. It would be interesting to see a benchmark of a z processor versus a PC when moving a string from a source variable to a target variable when the string length is 16 megabytes. Remember, we are not using a mainframe to create cute web page graphics. We are doing DATA PROCESSING.