Cooler Chips Incoming: Intel's Investments
ฝัง
- เผยแพร่เมื่อ 28 ก.ค. 2023
- There's been a spree of Intel news lately...
[00:36] Better Cooler Chips at 2 kW
[04:07] Building monolayer 2D transistors on 300mm wafers
[07:00] Assembly and Test facility in Poland
[08:26] Increased investment in Magdeburg
[10:38] Tunnel Falls: Quantum Computing
[13:30] Selling a stake in a nanofab business
[15:45] Installing Aurora and Ponte Vecchio
-----------------------
Need POTATO merch? There's a chip for that!
merch.techtechpotato.com
more-moore.com : Sign up to the More Than Moore Newsletter
/ techtechpotato : Patreon gets you access to the TTP Discord server!
Follow Ian on Twitter at / iancutress
Follow TechTechPotato on Twitter at / techtechpotato
If you're in the market for something from Amazon, please use the following links. TTP may receive a commission if you purchase anything through these links.
Amazon USA : geni.us/AmazonUS-TTP
Amazon UK : geni.us/AmazonUK-TTP
Amazon CAN : geni.us/AmazonCAN-TTP
Amazon GER : geni.us/AmazonDE-TTP
Amazon Other : geni.us/TTPAmazonOther
Ending music: • An Jone - Night Run Away
-----------------------
Welcome to the TechTechPotato (c) Dr. Ian Cutress
Ramblings about things related to Technology from an analyst for More Than Moore
#techtechpotato
------------
More Than Moore, as with other research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, which may include advertising on TTP. The companies that fall under this banner include AMD, Armari, Facebook, IBM, Infineon, Intel, Lattice Semi, Linode, MediaTek, NordPass, ProteanTecs, Qualcomm, SiFive, Tenstorrent. - วิทยาศาสตร์และเทคโนโลยี
The Polish distribution facility is about avoiding EU duties and taking advantage of cheap Polish logistics.
Poland is an integral part of the EU, so there is no way avoiding EU duties. The EU countries all apply the same tariffs to goods imported into EU territory from the rest of the world, and apply no tariffs between EU countries. The only reason to go to Poland is cheap labor and cheap ground.
@@bertnijhof5413of it goes from Ireland to Poland to EU markets they might avoid tariffs they would get if it was packaged outside the EU.
@@bertnijhof5413 There are always internal loopholes with something like the EU. The central body of the EU isn't a strong federal government like the US is. The EU is more akin to the early United Confederate States of America that existed between the end of the Revolutionary War to the adoption of the US Constitution. This is just how business has always worked.
@@bertnijhof5413 Just how daft are you? Thats the point! They bring the chips into the EU and package and distribute them in Poland claiming a massive value add. That way they avoid taxes on EU shipped chips
@@mycosys Ok, but Poland doesn't have low taxes. Why wouldn't they do it in Ireland?
Hopefully this new type of transistor comes out so we can stretch out the limits of what is possible on silicon.
The absolute limit is pretty far out however the practical limit is currently holding us back a bit even today.
What a lot of people miss is that software often lags behind hardware. If CPUs, GPU nodes completely stagnate today.. we would still be seeing plenty of uplifts due to software optimisation for years to come as well as plenty of efficiencies pulled out of the same node.
CPUs and GPUs today would make me a happy gamer for the next 40 years so long as they don't intentionally make games harder to run in order to sell new hardware. (Yes, absolutely this is the market today)
@@christophermullins7163 maybe in some cases but for the most part game devs don't have any interest in forcing people to buy GPUs. studios do, however, have an interest in having the best graphics (=most marketable) while investing the least possible dev time (=less cost & shoddy code). it's the same across all of software so long as faster hardware keeps coming out.
@@christophermullins7163they make games better because the hardware exists to run them, not just to sell hardware. 40 years is a long time to be stuck on current hardware. As amazing as games are today, they'll be better in the future because better hardware is available. Less than 40 Years ago the orginal Nintendo Entertainment System came out. Another 40 years we'll look at todays games like we consider NES games now, because new and improved hardware will make it possible
@@christophermullins7163
> oops, your garden variety toolchain now requires optimizations made upon the latest generation or two of silicon! Do not go past go and do not collect $200.
Cooler chips, I doubt it. Somehow chips have manged to be pretty hot for the last decade. They will just up the core count/clock to bring the temperature up to previous chips.
You might be right when it comes to desktops. But I can definitely see Microsoft taking advantage of intells 2d transistor tech for their 2 in one Laptops like the surface pro. Apple has been mauling Microsoft in the light laptop/ tablet market.
So Microsoft would gladly buy up large numbers of an intel mobile skew that could trade blows with apple silicon. Since Apple silicon has a monstrous edge in power consumption, and thermals due to the inherent advantage of ARM chips.
But that 2d transistor technology could help intel compete with apple silicon. Which would, in turn, give Microsoft a chance at taking back the mobile computing crown.
It's interesting that 127 and 433 are prime numbers while 27 and 12 are not. Just an observation.
I have an elaborate question you might know how to answer, out of most media people: when manufacturers have identical cores in consumer chips and HEDT (Xeon or Threadripper), why don't they allow individual cores to reach consumer chip clocks when occupancy is low? What prevents a Threadripper to run a few cores just as Ryzen would if more is not needed? Why not run 6 cores at the same clocks 6 consumer cores would run as long as the rest of the 58 cores are idle?
Interesting developments in optical computing, much lower power and much less heat dissipated.
1:52 - 6/31 resistance in the TIM and lid? this would imply that delidding without improving the attached cooling solution would only decrease the temp delta by 20% (50c in a 20c room -> 44c in a 20c room...)
NO, I have NO CPU that's using more than 125W. My gaming system uses a 5800X3D and avg. power consumption for that part in games is around 50W.
I'm just not into buying CPUs that eat up power for home use. I have to pay the electric bill and electricity is getting more expensive.
Amen to this. My gear is hilariously old by comparison, but it's new enough that either it idles low or stays off most of the time, and I'm sure not paying for ipc I don't need that costs me even more.
These chips probably aren't aimed at the gaming market, but the server market.
AMD has a problem in idle or middle tasks, uses to much power
@@nivea878 Then the motherboard bios isn't configured correctly.
Great video, one minor nit though: The city is spelled Magdeburg, not Magdeberg. ;)
I was wondering - is there any way to convert this heat difference (between data center chips and environment) to energy.
Very interesting on the PFAS..
Id like to see 6nm be maxed oht in potential, so we can have cheaper but great parts, we saw how intl stretched 14nm for so many generations and performance from first gen 14nm to last gen 14nm is pretty big.
Intel is never down, just that it was not able to market well like TSMC. Big blast of release gonna come in future.
Ah, yes, TSMC's marketing was so good, it convinced Intel to buy some capacity from them.
I want fins grown on the die from aluminum oxynitride for watercooling.
The DoD is usually exempt from restrictions on materials. They still use lead and beryllium.
Love these missives from a storage cupboard x
That's my office! It is also a storage cupboard. And a test area. And an editing suite.
Your spelling of Wrocław is allright. Cheers from Poland 😆
N transistors on top of P transistor - that's a peltier. Which be interesting if done right as would aid cooling.
Peltier junctions generate heat. Only if they were at the edge of the chip could they move the heat out.
Thx for the latest "Intel news" Ian!👍
Believe it or not that Intel is in county Kildare as quoted "Intel decided in 1989 to build its European manufacturing operations in Leixlip, County Kildare, and formed Intel Ireland to be the holding company on September 29, 1989. The manufacturing plant manufactured its first chip in 1993"
Quantum computing is so interesting, hopefully through the research program Intel succeeds in creating breakthroughs in quantum computing.
Interesting news, thx for them. On paper Intel is doing great. Finally Intel found the solution how to live in the future. No worry for the present, where they missed almost all the deadlines.
Vertical integration.
As long as there are computer chips, there will be intel as a leading provider.
"There is a big BUT", Ian can't lie.
Thank God, I'm not alone in going there. 👍
How important are the news about intel 3nm "Intel's 3nm-class process technology has met its defect density and performance targets"?
7:10 Intel will build a final assembly plant not in Wroclaw but in Miękinia near Wroclaw near the Odra river, which is the second largest in Poland.
The resistance to change induces a lattice energy, Max Yotta ohmagery, those boys got us off the moon, out of the clutch's of Roswell.
You don't need that complicated things, just boron arsenide semiconductor.
Hey, doc, at LTX looking for ya! You frequent any booths?
Been hanging out in the creator green room for lunch 🤣 look at Asus booth around 2pm
? blue spot (in the video) on the head...intention or
Intel - we aren't dead yet
I was under the impression they've been breaking or challenging historical revenue records consistently
*The black knight has entered the chat with African swallows.*
@@orthodoxNPCIntel is at around 2016 level revenue now so there has been decline for a while. It is still one of the largest and most diversed companies in tech hardware space though.
@@innocentiuslacrim2290 And soon they will have a lot of foundry capacity.
... yet...
I love how some technology is moving to three dimensions and other technology is moving to two dimensions.😂
Intel and TSMC need to build more foundries outside of Taiwan, especially if the 2027 invasion rumors become true.
TBF TSMC is about to build a new fab in Germany Dresden & also Japan Kyushu & on top of having moved one of their fabs to US Arizona already.
All we need is super conductors. Simple!
WHAT DO YOU MEAN 2000W PROCESSORS XD
I don’t think we must change cooling to submerging hardware in liquid. I think CPU’s and GPU’s must be made by other materials. Like grafeen instaid of silicone
"GPU: 150 to 350 watts"
Brother, let me tell you about the Radeon 6950 xt and its transient spikes.
And the 3090 as well. Ampere has some massive transients.
No cat tax??? 😱😱😱
How could you possibly build a 2 angstrom transistor? That's the exact width of a single Silicon atom. If a single atom could act as a transistor we would just do that.
Transistor sizes are nonsense now and will continue to be so
Computing on a single atom has been demonstrated a lot of times in the labs. So it is feasible both on theoretical and practical level. The question that remains , is when we will have the engineering capacity to produce such chips en masse.
What is still speculative, however, is if we can compute on a subatomic level. We'll, as crazy as it sounds some scientists at the extreme of theoretical
computation, even think of this!
👍
Y'all may need to chill a bit. 12+ years blazing with i5-2500k.
It was about time Intel got good use of their Pun department.
4090 is over 450 watts
Please provide timestamps, otherwise great video
6KW processors!! :O
13:55 "to 🅱ain Capital"
Hahaha, yes
Do you know why a square inch of chip real-estate has not gone down in price over the years?
Probably because the density is getting higher. 7nm -3nm etc.
Because density is going up and smaller nodes are just harder to make. You want a good node, you pay for it. Older nodes can be really cheap, like the node used for MTL interposers and the more mature N6 node AMD uses for N33 rather than the N5 node for N32 and N31.
it's Simple: ask Apple how they made M2 chip. I bought M2 Macbook and no fans, no heat. no jumbo jet fan noise.
I wonder if we will get some of those transistors to no longer be 100% accurate and maybe have some probability to them. Given that some workloads like machine learning do rely on statistics and if you had specific blocks that have a known probability... You could really use that for sampling. Only matters if the workload becomes the vast majority of computing soon.
Intel is sampling PVC-56 internally and to 'partners'. But they said to not sell individual cards for workstation... Because "it's a data center product". So our hope is some integrator will do it anyway. Right now only Supermicro is selling a 10x GPU server and that's not a workstation. Intel is building out their DevCloud capacity instead and will likely tell developers to use that. And they do use Gaudi2 instead of PVC since it's performing around 3x as good as PVC for LLM training and inference.
Intel is now selling SPR CPU with AMX for workstation, small deployment.
I have to rely on openVINO inference and their 2023.1 prerelease is supposed to be "end of July" and bring improvements for GPU language model inference. Not having an Nvidia GPU anymore is delaying my research by what feels like a year.
Flourine is bad stuff
We don't need better cooling methods, we need cooler CPUs!
Truer words have never been spoken
Its hard to feel good about Intel. Shredding profitable non-core businesses can end quite badly.
It's a simple burning of the fat to survive lean times. Their quarterly earnings report shows they're not falling any further behind on their product/node timelines, so by the end of the year they should have stopped diving.
This is starting to look like what GE did.
I think it's bad news that they want to invest more in cooling than working on performance per watt.
Fortunately, it’s only in data center… for now.
No need to choose only one of these.
@@whyjay9959 Indeed. It’s likely Intel is busy working on both power and cooling efficiency, what with Arrow Lake being purported to have only E-Cores, with no Hyperthreading.
Uh, you do realize that performance per watt improvements is largely achieved through transistor scaling, right? Is "5 nodes in 4 years" not fast enough, and 10s of billions of dollars not significant enough an investment?
@@Moonspec Is it fast enough or not will be decided through competition. Besides transistor density matters architecture and they are slow in both cases
I have no issue with temperatures, running a Ryzen at 65W TDP.
NOVEC Does NOT seem to have any environmental or human safety hazards, UNLIKE Fluroinert. NOVEC is a Flourinated Ketone, can't find anything other than it is safe...
Novec and Fluorinert are families of compounds from 3M, and the Fluorinert product line falls under the Novec branding.
Ironic that Intel is getting paid for researching how to cool 2kW TDP processors since they are partially responsible for raising TDP in order to goose the performance of their own consumer CPUs when they were stuck on 14nm for several generations! I'd like to see a return to 65W as being the norm for desktop rather than the 'low end', as it's way easier to manage the heat at that power - and quieter, to boot!
less mechanic parts then thinner devices
This is research hardware for HPC. Not your gaming crap. Keep that at home.
If you want 65W, set your motherboard to run at 65W PL1. That's an option on Intel, AMD AND Nvidia GPUs and CPUs.
Bain capital is a bad news company. They will squeeze customers.
imagine, water was programmable, cpu on water base
7Ghz
Why intel is going to Germany. Why not stay in USA?
In my political view :D it's just a political desicion. why does US let German talents build their own CPU ?
Search Intel Magdeburg deal.
@@shooyooyoon9007Don't worry it's just a front end production site. Development may mainly somewhere else.
rather than make Intel CPU more efficient (performance/watt), they just want to keep pumping up the power and improve cooling. sounds like Pat is more of the same.
Yeah, AMD has proven that better efficiency is possible with existing performance. I don't need any more "free heat" in my office!
What a bad take, as thats not related at all. The DoE wanted this and is partially paying for the Cooler project, and total power usage has increased across the board, IBM, Intel, AMD, Nvidia, etc are all gulping more power than ever in data centers. Also the architectural teams for all these companies are not the same teams working on cooling the chips.
@@UncleKennysPlace Intel's CPU's are more efficient in real use than AMD lol.
great persecution of Wrocław .. i live 30 km from new factory in Poland :)
1st ~
aaand spelling is incorrect once more! Wroclav, not Wro-claw, Wroc-law, and "c" here reads like metal knok (do not know how to explain to EN-only speaker), and "law" reads as "l-a-w", not "l-o-w"
as ukrainian who knows polish (two very close languages btw)
Or, if I haven't explained well - just use ye olden name - Breslau (:
Spelling or pronunciation? There are comments here from Polish people who said I did a good job on both :)
Intel better make sure the chinese don't steal the technology
Intel is way behind AMD in server chips...like by a LOT and no joking. Mobile is basically non existing and Laptop again, AMD just killed them and Apple M3 will kill both of them. GPU? Some say it's too little, too late. What about AI? (They're JUST NOW working on it) Even Supercomputers are moving away from Intel chips What DA F is happening with Intel??????
Intel is way ahead of AMD in laptop shipments. Sorry to burst your bubble. Especially during 2020-2022, AMD had a lot of trouble getting the supply out for their Zen 3 TSMC N7 chips while Intel waited until they had the yields and fab capacity for 10nm to launch Tiger Lake and Alder Lake which had the perfect supply. Anything that was Zen 3 was vaporware until Intel's 10nm was out for like a whole year, then when there was enough Zen 3 supply, Alder Lake blew Zen 3 out of the water with ST and MT (because there was no X3D on mobile).
AMD only recently had a chance to beat Intel squarely at mobile with Zen 4 and X3D. That's AMD's latest architecture on the latest TSMC node while Intel's Raptor Lake is a refreshed architecture on a 3 year old 10nm node (but with major improvements). Even now, there are only a select few laptops with Zen 4 7840HS and 7940HS, majority of AMD laptops being sold on the lower end are still Zen 3.
With Intel 4 and Meteor Lake, Intel is in a good position against AMD and Apple on laptop. On Desktop and Server - they're still a bit weak.
We need intel to keep losing marketshare until AMD gets to 50% across ALL x86 markets.....that would bring about a balqnce of power and the best possible situation for consumers in a defacto duopoly....so NO, I dont want Intel to do anythint "good" for a few more years
So basically you want to lose 50% of progress for at least a few years. Very smart of you.
@@MrVladko0 I don't need that level of progress, 99% of people don't need it, and AMD can do the progress part for now until they balance eachother out,
then they can compete even harder and progress together farrr more intensively
X86 is an obsolete architecture. Time to move to arm
Lol never gonna happen
arm isn't great for desktop usage. I think by the time we *need* an architecture restructure, the industry will want a more drastic improvement than arm
😂😂😂 ok
ARM lol, never ever, AMD + Intel and in the future others gonna show new tech and its not arm for certain
If the socket AM5 LGA ... can't be made tighter what's the point of all this....There's room enough when I watch this NOW