News Intel Working on New Cooling for Chips up to 2000W

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

rtoaht

Reputable
Jun 5, 2020
102
107
4,760
It seems most people in the comments assumed this is aimed for consumer chips. Either the article didn't communicate the news properly or this is not the right forum to publish server technology related news. Otherwise, it might mislead people into thinking something that is untrue.
 
Last edited:

truerock

Distinguished
Jul 28, 2006
299
40
18,820
Since the recommended max for a US household outlet is 1500W, I guess your next Intel PC will go in the garage next to the tumble dryer. :p
Exactly what my first thought was.

I'm in worse shape than that. My study (where my desktop PC is located) is wired with just one 15amp/120v circuit. That powers 4 ceiling lights, also. What was I thinking when I built this house!?

If I built a house (in the US) today, every outlet would be wired 20amp or 40amp and I would have a dedicated circuit for my desktop PC. And, of course I'd have two 4,800 watt outlets in the garage for charging electric cars.

I had an apartment above my garage re-built about 4 years ago. We put an electric water heater in it that used four 40 amp circuits... 19,200 watts!! The only reason we were able to do that was because the garage apartment was built on top of where the mains-electric entered the property. I'd like to replace my 15-year-old gas-fired water heater in the main house with an electric water heater - but, there is no way to get four 40 amp circuits to that part of the house.
 

bit_user

Polypheme
Ambassador
Single CPU, correct with the exception of certain entire wafer chips.
Those are even more exotic than the mainframe CPUs I made an exception for.

However edzieba said devices. So 4+ CPU, raid of SSD / M.2 / U.3 drives , GPU/Compute cards, multiple 100gb+ networking cards / connections, 1TB+ of Ram and the cooling to cool all that (plus any other add-in cards for specialty purposes) you can easily get over 2kW per 2-4U server (aka a single device).
I never debated that point. I just said he's confusing machine power with CPU power. The article is talking about 2kW CPUs! Yet his post was ridiculing everybody in this thread for thinking this was a new development, but it is!

As for servers burning > 2kW, it's mostly the GPUs which push it over the line. Otherwise, it's things like blade servers that could get into that range.

Intel also said "devices" rather than chips in their press release
Given we've had > 2 kW servers for a while now, it's clear that they weren't talking about that kind of device. They probably were just speaking inclusively of CPUs and GPUs, without getting bogged down in the distinctions.
 
Last edited:

bit_user

Polypheme
Ambassador
Software licensing is based on what the customer is willing to pay for the service. For Engineering software it is often based on the number of users and their location.
For databases, it does tend to be tied to CPUs/cores, from what I've heard.

This is one reason the company I worked for moved it's engineering work to The Philippines and PC motherboard design to Taiwan.
: (
 

Kamen Rider Blade

Distinguished
Dec 2, 2013
1,280
810
20,060
Software licensing is based on what the customer is willing to pay for the service. For Engineering software it is often based on the number of users and their location. For example, it costs $20,000/user/yr for a user in the United States or Europe and is $2000/yr for a user in China, Taiwan, Malaysia or The Philippines. This is one reason the company I worked for moved it's engineering work to The Philippines and PC motherboard design to Taiwan.
That sounds real scammy to have such a difference in pricing based on region.
 
That sounds real scammy to have such a difference in pricing based on region.
But it's not. At least you can't say that that easily.
IISC_EqualityEquity.png
 

edzieba

Distinguished
Jul 13, 2016
438
431
19,060
I never debated that point. I just said he's confusing machine power with CPU power. The article is talking about 2kW CPUs! Yet his post was ridiculing everybody in this thread for thinking this was a new development, but it is!
Because the article talking about 2kW CPUs is simply incorrect, and misread the Intel press-release.
People railing about something that has not actually happened deserve ridicule, at the very least for failing to check the original source despite being linked directly. All outrage over imagined occurrences can and should be taken to task as a matter of principle.
 
  • Like
Reactions: KyaraM

bit_user

Polypheme
Ambassador
Because the article talking about 2kW CPUs is simply incorrect, and misread the Intel press-release.
I can understand you're embarrassed to be corrected, when you were trying to be the one doing the correcting. Try thinking about this logically, if you can.

From the Intel source:

"Intel researchers are developing novel solutions to support the power and thermal management needs of next-generation architectures, including devices up to 2 kilowatts."

As you already stated, 2 kW servers are nothing new. And they said up to 2 kW! If they were talking about servers, then they wouldn't only be concerned about next-generation architectures and they wouldn't stop at 2 kW. It's obvious that what they meant by "devices" is something like a CPU or GPU package or perhaps even a module, like Nvidia's SXM cards or OCP's OAM cards.
 

edzieba

Distinguished
Jul 13, 2016
438
431
19,060
I can understand you're embarrassed to be corrected, when you were trying to be the one doing the correcting. Try thinking about this logically, if you can.

From the Intel source:

"Intel researchers are developing novel solutions to support the power and thermal management needs of next-generation architectures, including devices up to 2 kilowatts."

As you already stated, 2 kW servers are nothing new. And they said up to 2 kW! If they were talking about servers, then they wouldn't only be concerned about next-generation architectures and they wouldn't stop at 2 kW. It's obvious that what they meant by "devices" is something like a CPU or GPU package or perhaps even a module, like Nvidia's SXM cards or OCP's OAM cards.
Or that they are targeting density and disaggregated devices (e.g. split your 6U DGX into several discrete networked 1U or blade devices) rather than monolithic devices. Monoliths are already well served by immersion cooling, targeted cooling is better suited to disaggregation.