I Tried to Break a Million Dollar Computer - IBM Tour

[James]

IBM invited us to do a facility tour showing off the mainframes they sell to high frequency trading customers and the like. You want to see wacky, custom hardware? This is it!

 

 

 

[DANK_AS_gay]

My Uncle works for the Pentagon (in Ohio, but is technically a Pentagon Employee), and he told me that the Power architecture sucks balls. It's incredibly hot, it's many times slower than other architectures, and is outrageously expensive. Plus the way it handles memory is unfamiliar and difficult to use properly, which isn't good given the application. They actually ended up buying a massive number of A100s, and handing down the V100s to the Air Force. If I remember correctly they have Power 9 and Power 8 machines still. This is coming from a department that has a silo for water. In fact, a pump failed recently, and they had to have fire trucks pump in cold water to keep the hardware from melting itself. THE ENTIRE FIRE TRUCK TANKER COULD ONLY KEEP THE THINGS COOL FOR 8 MINUTES.

[Borhork]

I would love to see a follow up on the other systems IBM makes. The cognitive power line with hybrid cloud would be very cool to see. 

[Zodiark1593]

"I don't need SAS from my camera operator..."

 

Get it? SAS!

 

*insert mic drop here*

[MultiGamerClub]

Suprised IBM is still that big..

[Results45]

Word on the street is that the DoD is going to use exascale supercomputers with hardware like this (Probably Intel's Aurora & AMD's Frontier) to simulate nuclear missile failure scenarios in extreme detail to inform the appropriate upgrades and maintenance efforts.

 

I doubt Russia has nothing like this to work on their nukes. China's maybe, but they're probably still playing catch-up. Japan is certainly not going to lend Fugaku to either of them.

[Nystemy]

The reason why the "ECC" is done outside of the RAM module is incase the whole module dies. After all, DRAM is quite likely to fail when one aims for the upmost in uptime.

 

And yes, DRAM is really slow from an access latency standpoint, especially if we want to change row address. So going to another CPU's cache could be faster if there is enough available bandwidth and our connection is sufficiently low.

 

Some people might though be surprised at it "only" having 256 cores at a maximum. But to my knowledge IBM's Z systems do have a lot stricter error rates than Intel/AMD puts onto X86. After all, Intel/AMD markets peak performance since that is what their customers mostly are after, while IBM markets peak reliability. _{(And I wouldn't be the slightest surprised if some applications on a Z16 actually computes everything as two separate instances and then cross compares the results to check for errors.)} So peak performance is a second _{(likely even lower)} priority. After all, it is very much inexcusable for a bank to make a transaction error. _{(And having read some whitepapers about PowerPC from IBM over the years they sure do like talking about how to capture compute errors, and the pros and cons of various ways of responding to such a capture.)}

[HenrySalayne]

Are they flooding the facility with helium (something like Trimix) or is Linus' Mickey-Mouse voice a side-effect of the noise removal in post? It sounds way off.

[Ultraforce]

1 hour ago, MultiGamerClub said:

Suprised IBM is still that big..

IBM is still a super massive company it's just most of their offerings are now more so enterprise focused rather then for consumers. Back in November one of my 4th year classes had as a guest lecturer a lady who had been at IBM for 40 years working on their mainframes and had moved to a consulting firm that heavily deals with COBOL.

[Vishera]

Linus was having so much fun screaming in the test facility that he lost his voice

32 minutes ago, Ultraforce said:

IBM is still a super massive company it's just most of their offerings are now more so enterprise focused rather then for consumers.

IBM sold it's consumer division,now days it's called Lenovo.

[dogwitch]

2 hours ago, DANK_AS_gay said:

My Uncle works for the Pentagon (in Ohio, but is technically a Pentagon Employee), and he told me that the Power architecture sucks balls. It's incredibly hot, it's many times slower than other architectures, and is outrageously expensive. Plus the way it handles memory is unfamiliar and difficult to use properly, which isn't good given the application. They actually ended up buying a massive number of A100s, and handing down the V100s to the Air Force. If I remember correctly they have Power 9 and Power 8 machines still. This is coming from a department that has a silo for water. In fact, a pump failed recently, and they had to have fire trucks pump in cold water to keep the hardware from melting itself. THE ENTIRE FIRE TRUCK TANKER COULD ONLY KEEP THE THINGS COOL FOR 8 MINUTES.

that part of the reason why both cell/ 360 cpu ran hotter then normal.

they truly have smart people at ibm. but they tend to get over ruled. due to greed.

[An0maly_76]

IBM = Incessantly Bumbling Morons.

[Leiska]

Why was the Z on the Z16 blacked out for the opening but shown seconds later?

[HenrySalayne]

4 minutes ago, Leiska said:

Why was the Z on the Z16 blacked out for the opening but shown seconds later?

Congratulations, you found an oversight. Watch the news, it might give an idea why they did not want to have this symbol in their video.

[Ultraforce]

1 hour ago, Vishera said:

Linus was having so much fun screaming in the test facility that he lost his voice

IBM sold it's consumer division,now days it's called Lenovo.

Oh, didn't realize Lenovo was previously related to IBM.

[tkitch]

28 minutes ago, Ultraforce said:

Oh, didn't realize Lenovo was previously related to IBM.

That's why lenovo makes Thinkbooks, Thinkpads, etc.  Those are all the old IBM names, and they were purchased / spun off.

[faketruth]

good job, ibm is trying to promote and make their image looks positive after all of the age discrimination lawsuit that have been plaguing them. and using LTT is their way to do it.

 

and for anyone who doesnt know about it, just search using keyword "ibm age discrimination lawsuit".

 

wonder if LTT just dont do enough research about the controversy of a company that they are covering/sponsored with or just blatantly ignore it  straight away at this point and just apologize later.

[Ultraforce]

18 minutes ago, tkitch said:

That's why lenovo makes Thinkbooks, Thinkpads, etc.  Those are all the old IBM names, and they were purchased / spun off.

Ah, I only really was exposed to home computers and laptops really in the later half of the 2000s  so by then the ThinkPad was already Lenovo. I had some good experiences and some bad ones with the Thinkpads. They were available for school laptops and worked okay but I had one at home which had on the side a grill for the heat to escape from and that one keyboard frame started coming apart near the end of my time with it.

[tkitch]

1 hour ago, faketruth said:

 

*reads username*
Sounds about right.

[Needfuldoer]

2 hours ago, HenrySalayne said:

Congratulations, you found an oversight. Watch the news, it might give an idea why they did not want to have this symbol in their video.

Maybe there's a small detail (like an address or serial number) written on it that's illegible in the wide shot?

[LeLuc]

6 hours ago, DANK_AS_gay said:

My Uncle works for the Pentagon (in Ohio, but is technically a Pentagon Employee), and he told me that the Power architecture sucks balls. It's incredibly hot, it's many times slower than other architectures, and is outrageously expensive. Plus the way it handles memory is unfamiliar and difficult to use properly, which isn't good given the application. They actually ended up buying a massive number of A100s, and handing down the V100s to the Air Force. If I remember correctly they have Power 9 and Power 8 machines still. This is coming from a department that has a silo for water. In fact, a pump failed recently, and they had to have fire trucks pump in cold water to keep the hardware from melting itself. THE ENTIRE FIRE TRUCK TANKER COULD ONLY KEEP THE THINGS COOL FOR 8 MINUTES.

I believe z and power are different systems…

[faketruth]

51 minutes ago, tkitch said:

*reads username*
Sounds about right.

so whats your point? i guess you represent how ltt forum "discuss" thing, looking at the other username and judge based on it instead of offering a counter argument to what i said. /shrug

 

and what i said still stand, ibm is not getting a good image right now, and they just use ltt coverage to promote/bend their image into a better one. this is the same tactic used by far right just by making memes of nazi stuff fun to attract and slowly make people thinks "its okay, its just a meme".

am i thinking too far into this? maybe - but i want to highlight that ltt have that much of influence to shift opinion of public on a company simply by doing coverage for them - especially on controversial company they are covering/getting sponsored with.

 

[a tech newbie]

at what point does he try to break it?

[Markypizz]

This was super cool! I worked at IBM for 2 years during the development of the power 10 systems, specifically working on memory bring-up. We worked on those exact DDIMMs (differential DIMMs) that Linus was holding @ 11:25. The code we wrote for Power ended up in Z, so those systems are running code that I wrote! He briefly mentioned the power management chips (PMIC) in DDR5 that were on these DDR4 modules. This was super brand new at the time and we were on the ground floor of it, getting an early head start having those modules on our DDR4 dimms before DDR5 came out. The code running on P&Z to initialize those was my main focus while working there, and parts of our team had a hand in architecting the JEDEC standard for how those PMIC chips should be operated. (https://www.jedec.org/document_search/field_keywords/pmic-3786) Super cool to see this blast from the recent past. Thanks for the great video Linus and crew!

[yakovlev]

9 hours ago, Nystemy said:

The reason why the "ECC" is done outside of the RAM module is incase the whole module dies. After all, DRAM is quite likely to fail when one aims for the upmost in uptime.

 

You're right, but I'm pretty sure Linus didn't quite get the facts straight on the details of ECC vs. RAIM in these systems.  To do RAID you have to have some way of detecting errors on individual drives, and I would expect the same is true on the individual modules in a RAIM scheme.  IBM Z memory architecture is somewhat unique, so I suspect that may be part of the reason for the unusual DIMM configuration.