> Cooling is one of the lowest priority problems when designing a mainframe CPU

That's not true at all. Next to power deliver, power removal--aka HVAC--is next in line of recurring costs (after one-time costs of building the facility and the initial server BOM).

For example: https://perspectives.mvdirona.com/2008/11/cost-of-power-in-l...

Unless by "mainframe" you mean specialty compute-server farm for a supercomputer (with massive IO) then perhaps.

IBM z chip is for mainframe and it's too expensive and normally not massively scaled out compared to cloud servers.

Also, not to beat a dead horse :), but saying "cooling is ... lowest priority" is misinformation so I want to put this here for future readers:

I spent most of my career working on power and cooling for CPUs and MCUs and have presented on the subject at ISSCC since the 90's. I can tell you with absolute certainty that modern CPU power concerns have never NOT been a problem. Power, area and timing are the three competing constraints that all CPU manufacturers struggle to balance. And it is not easy since fixing timing can hurt power and area, and vice versa for the other two components. Finding a solution that doesn't impact the other two is hard (and very, very valuable).

Recall the early 1990's when PowerPC dinged the Pentium for being a 16W TDP part ("TDP" didn't even exist as a term in datasheets at the time, that came along once Intel added throttling with the P4). Ahh the good old days.

> Recall the early 1990's when PowerPC dinged the Pentium for being a 16W TDP part

PowerPC...

Up until the series 9672 (1994) when IBM changed from bipolar to CMOS mainframes had water cooled CPUs (although the ES/9000 had air cooled options). Water cooling returned as an option with the zEnterprise in 2010 and the zEC12 CPU made it mandatory a few years later. These were not 16W parts. Not even 160W.

Interesting, I didn't know that. I have very little historical knowledge pre-1990 about mainframes, and by the time I started working in the field the term "server" replaced "mainframe" exclusively. In fact, during my graduate years, the mainframe & TTY terminals were replaced with IBM AIX workstations, and a big honkin file server, so I never really got to dive into it before it was gone.

They were not called big iron for nothing. They are not measured in racks but metric tons and such :) In 1991, a Hungarian university installed its mainframe by partially removing the roof and craning the thing in place. I am not kidding. Hard to find photos. http://hampage.hu/nosztalgia/netto/ursus/ursus.jpg here's a small section of it.

https://www.backblaze.com/blog/wp-content/uploads/2016/11/hi... look at that dainty hard drive in the back :D

Well, you still need to prevent it from melting. Coefficients of heat transfer are not infinite.

For a mainframe, you could probably get away with using liquid nitrogen if you needed to.

Sadly, IBM has not the same penchant for striking design as Seymour Cray had.

Well... Their quantum computer is quite striking. I wish they did that with the more mundane machines.