AMD Expected to Occupy Over 20% of Server CPU Market and Arm 8% in 2023

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,785
“Everybody up” — Diamond Dave (Live Van Halen - Unchained)

“However, datacenter operators and server brands are still aggressive about Arm processors' development in the server market. Amazon and Alibaba have already started working on Arm-based products before 2022, Microsoft and Google also began projects with Arm products in 2022, and HPE is expanding its adoption of Arm-based servers. Nvidia is now pushing its GPUs to support Arm architecture and Ampere is developing Arm-based chips. In the upcoming years, the opportunity from ESG is expected to take off for Arm CPUs as demand from large-scale datacenter and edge computing servers will surge, Kung added.”

1677187745691.jpeg


Source: https://www.techpowerup.com/305141/...ver-20-of-server-cpu-market-and-arm-8-in-2023
 
“Everybody up” — Diamond Dave (Live Van Halen - Unchained)

“However, datacenter operators and server brands are still aggressive about Arm processors' development in the server market. Amazon and Alibaba have already started working on Arm-based products before 2022, Microsoft and Google also began projects with Arm products in 2022, and HPE is expanding its adoption of Arm-based servers. Nvidia is now pushing its GPUs to support Arm architecture and Ampere is developing Arm-based chips. In the upcoming years, the opportunity from ESG is expected to take off for Arm CPUs as demand from large-scale datacenter and edge computing servers will surge, Kung added.”

View attachment 551407

Source: https://www.techpowerup.com/305141/...ver-20-of-server-cpu-market-and-arm-8-in-2023
ARM is a huge deal for servers where you need lots and lots of very lightweight services hosted.
You don't need a lot of beefy cores for a web server, or a payment gateway, but you do need a lot of independent cores.
Altera offering some 8-way 128-core beasts allowing for 1024 independent CPU cores in a 2U chassis is an absolute powerhouse when it comes to processing a bazillion credit card payments a day, or a similar number of web browser sessions.
There is nothing that AMD nor Intel offers that can deliver on that many independent sessions at the price/power/thermal budgets that ARM offers there, not even close they are well over 3x more expensive over their lifetime which works out to millions of dollars per rack.
AMD though is offering insane virtualization density, they offer the best balance of core counts and core performance that lets you overprovision the crap out of a VM cluster. You are going to run out of ram and storage throughput before you reasonably max out the CPUs in a properly load-balanced VM rack with the AMD offerings.
Intel is in a rough place, sort of, they are a very known entity, and they do an absolute shit ton of CPU validations so in massive data center environments they offer a degree of stability and reliability that AMD does not currently match, but if you aren't running a massive datacenter or you aren't dealing with insane degrees of required precision then it's not something you need to generally worry about.
But AMD is getting better at their fault correction, so Intel can't count on that forever, and the jobs that do need that degree of reliability and precision are being taken over by ARM and GPUs so Intel needs to issue new marching orders, they have more competition than ever and they are gaining speed.
 
AMD has captured quite a bit of the server market, no wonder Intel is hurting, especially with their fab issues as well. Intel better figure it out soon or they might become the alternate player in the market instead of the main player, losing 7% of the market a year should scare their management.
 
AMD has captured quite a bit of the server market, no wonder Intel is hurting, especially with their fab issues as well. Intel better figure it out soon or they might become the alternate player in the market instead of the main player, losing 7% of the market a year should scare their management.
AMD is not the entity that Intel needs to worry about, their problem is TSMC, TSMC is giving AMD a clear node advantage that Intel won't catch up to until 2025 (by both Intel and TSMCs estimates) at the earliest. Intel 7 is really good when compared to TSMC 7, anything made there is a dead heat, but AMD isn't using TSMC 7nm, they have moved on to 5/4, ARM is moving to 4/3. Intel is still 6 months out from having their 4 nodes up and swinging (assuming they don't delay it again) and the Intel 3 and smaller won't be doing consumer runs until 2025 at the earliest. That is a long haul in the current markets, and Intel is going to get smacked around hard until then, and should they delay further it's only going to compound their troubles further.
Intel needs to keep its head down and stay on target, they are obviously going to get through it, how much of their current management team does is another story all together.
 
It's very simple. AMD with TSMC is an Intel nightmare. One without the other - it is nothing for Intel to worry about, but that's not the way it is.
Think Intel can shed the fat and go Fabless?
 
I think its just more than their fab. I think the AMD business model is working and Intel has way too many project's and CPU's that failed. They just wasted money in failed projects.
Agreed, Intel has sidetracked way too much outside it's core business at great expense. Some of it worked, but a lot of it didn't.
 
Intel's fabs are the US governments' hedge against China invading Taiwan 👌
TSMC is also building a multi-billion dollar plant in Arizona. They are also diversifying locations to both increase capacity as well as be in locations that are more "politically" stable.

Reuters previously reported that TSMC plans to build as many as six factories at the Arizona site over a 10- to 15-year span.

Fabs in general just coming to the US: https://www.tomshardware.com/news/new-us-fabs-everything-we-know
This is pretty serious stuff, in the sense that the US is securing chip manufacturing as a strategic resource in the event of conflict. Which is nearing daily.
 
TSMC is also building a multi-billion dollar plant in Arizona. They are also diversifying locations to both increase capacity as well as be in locations that are more "politically" stable.

Reuters previously reported that TSMC plans to build as many as six factories at the Arizona site over a 10- to 15-year span.
Fabs in general just coming to the US: https://www.tomshardware.com/news/new-us-fabs-everything-we-know
This is pretty serious stuff, in the sense that the US is securing chip manufacturing as a strategic resource in the event of conflict. Which is nearing daily.

Been discussed before, that doesn't save them if China invades, they can't just up and move everyone and operate the whole company from a fab in Arizona - TSMC is screwed if China invades
 
Think Intel can shed the fat and go Fabless?
No hard no. Who could take that on? Who would Intel contract their chips too?
TSMC may have more fabs than Intel, but those fabs already supply over 500 customers, TSMC doesn't have the manufacturing capacity to take Intel on as a client, and Samsung's nodes aren't viable for a processor of that size, so Intel would sell off its fabs to then buy time on them at a higher price like AMD did for GoFlo and unlike AMD at that time Intel is not hurting for cash.
Intel's only major strength is its design and manufacturing and they know it and they are leaning toward it. They have something like 5 new fabs being built in the US and Europe right now ($50B a pop) and I want to say 3 of those are going to be for 3'rd parties.
Intel is looking to get into making other people's chips like TSMC does because that is the future, Intel knows x86's days are numbered, maybe it's a decade out but if they don't prepare accordingly and work purely reactionary by the time it's time to react there won't be a place left for them.
So really it is quite the opposite, Intel is preparing for a future where its engineering and manufacturing are the core business and the x86 side of things is legacy.
 
Been discussed before, that doesn't save them if China invades, they can't just up and move everyone and operate the whole company from a fab in Arizona - TSMC is screwed if China invades
Everyone? No. But it's not as if when Taiwan is nuked, all of a sudden a fab in Arizona stops operating either. I'm not sure you know what a hedge is. At minimum it could continue to operate and service the customers in the US (which according to the article number in the hundreds), which is its intended purpose.
 
Last edited:
like AMD did for GoFlo and unlike AMD at that time Intel is not hurting for cash.

Also whatever misses Intel has had of late and whether that fault is due to design or fab people, Intel's fabs are still miles better than AMD's were comparatively when they were selling them
 
But it's not as if when Taiwan is nuked, all of a sudden a fab in Arizona stops operating either. I'm not sure you know what a hedge is. At minimum it could continue to operate and service the customers in the US (which according to the article number in the millions), which is its intended purpose.

More like it would be sold off in that case
 
TSMC is also building a multi-billion dollar plant in Arizona. They are also diversifying locations to both increase capacity as well as be in locations that are more "politically" stable.

Reuters previously reported that TSMC plans to build as many as six factories at the Arizona site over a 10- to 15-year span.

Fabs in general just coming to the US: https://www.tomshardware.com/news/new-us-fabs-everything-we-know
This is pretty serious stuff, in the sense that the US is securing chip manufacturing as a strategic resource in the event of conflict. Which is nearing daily.
until it's actually built and running means nothing. there are plenty of companies promised by plenty of states only to get fk'd and walk out on half finished projects. all you have to do is look at where i live where we have 8-10 completely empty massive warehouses that were abandoned by boeing, amazon, and a couple EV start up companies before they ever moved in because the deals were setup by one city government and dropped by the next batch of idiots that got voted in.
 
TSMC is also building a multi-billion dollar plant in Arizona. They are also diversifying locations to both increase capacity as well as be in locations that are more "politically" stable.

Reuters previously reported that TSMC plans to build as many as six factories at the Arizona site over a 10- to 15-year span.

Fabs in general just coming to the US: https://www.tomshardware.com/news/new-us-fabs-everything-we-know
This is pretty serious stuff, in the sense that the US is securing chip manufacturing as a strategic resource in the event of conflict. Which is nearing daily.
The US want's to have 30% of the global supply of silicon coming from its borders by 2030, it is currently less than 12% and of that 12% Intel is the "most advanced" and only makes up a little better than half of that, with Texas Instruments and the like way in the back.
 
Been discussed before, that doesn't save them if China invades, they can't just up and move everyone and operate the whole company from a fab in Arizona - TSMC is screwed if China invades
Yes TSMC is scewed but the US operations will live on either as TSMC USA or some other guise.
 
until it's actually built and running means nothing. there are plenty of companies promised by plenty of states only to get fk'd and walk out on half finished projects. all you have to do is look at where i live where we have 8-10 completely empty massive warehouses that were abandoned by boeing, amazon, and a couple EV start up companies because the deals were setup by one city government and dropped by the next batch of idiots that got voted in.
This is private investment and equity. Not a government. So if TSMC and these other foundries decide to not build, any of their initial investment is a sunk cost. That cost would be in the billions. Yes the US is making chip manufacturing attractive in the US by offering incentives. Again though that only offsets some of the costs that TSMC is having to spend themselves. Even if they choose to "not build" then they are also leaving huge amounts of money on the table. The demand for chips is only going up. If they aren't going to build fabs in the US they have to build them somewhere or more or less let other competition (like Intel/Samsung) gain marketshare. It is categorically not in TSMC's interest to not build fabs in order to make more money.

Also, though Democrats and Republicans are petty towards one another for political points, the one united front the US has is its position on China. I find it very unlikely that either will want to pull incentives or not have greater domestic chip manufacture when the reasons for having it is for defense, stability of the nation, and as a middle finger to China.
 
Yes TSMC is scewed but the US operations will live on either as TSMC USA or some other guise.

Maybe Nvidia buys/comes into possession of it by then cause they want to do literally everything in house eventually! 😆
 
Also whatever misses Intel has had of late and whether that fault is due to design or fab people, Intel's fabs are still miles better than AMD's were comparatively when they were selling them
Intel got fat and complacent, then they went full 180 and put a guy in charge who wanted to plow ahead regardless of the technological limitations of the time, what they tried doing were cool, revolutionary, and expensive as hell, but mostly all met the same end with "the manufacturing technology needed to do this doesn't exist, neither does the technology to make it".

ASML eventually got there but it took some 6 years, and even their latest and greatest stuff now Intel can't use for some of the tech they wanted to deploy in 10nm, they aren't getting that until they move down to 20A which is still some 2-3 years out.

Intel, Great engineering, and awesome manufacturing, but their leadership and management for the past 15 years really screwed the pooch and existed as the perfect living caricature of every old "fat cat", and are directly responsible for the mess Intel is in. They fell off the wagon and now they are having to not only scramble to catch up but fight for their seat.
a0MIWngZJ4kdnLTu98qlqMs2USI6yWRqLhvTy6QoY&usqp=CAU.jpg
 
Yes TSMC is scewed but the US operations will live on either as TSMC USA or some other guise.
You can always hire more workers, but infrastructure takes years to build and years longer to train staff to use.
 
Intel got fat and complacent, then they went full 180 and put a guy in charge who wanted to plow ahead regardless of the technological limitations of the time, what they tried doing were cool, revolutionary, and expensive as hell, but mostly all met the same end with "the manufacturing technology needed to do this doesn't exist, neither does the technology to make it".

ASML eventually got there but it took some 6 years, and even their latest and greatest stuff now Intel can't use for some of the tech they wanted to deploy in 10nm, they aren't getting that until they move down to 20A which is still some 2-3 years out.

Intel, Great engineering, and awesome manufacturing, but their leadership and management for the past 15 years really screwed the pooch and existed as the perfect living caricature of every old "fat cat", and are directly responsible for the mess Intel is in. They fell off the wagon and now they are having to not only scramble to catch up but fight for their seat.
a0MIWngZJ4kdnLTu98qlqMs2USI6yWRqLhvTy6QoY&usqp=CAU.jpg

Yeah the problem with being a multi-headed hydra that does everything is you then have to steer the ship, and fucking up is easy

Edit: resting on their laurels aside, which absolutely
 
You can always hire more workers, but infrastructure takes years to build and years longer to train staff to use.
Should china invade taiwan, the existing TMSC staff won't go anywhere and will have no problem continuing operation with US/Intel help.
 
No hard no. Who could take that on? Who would Intel contract their chips too?
TSMC may have more fabs than Intel, but those fabs already supply over 500 customers, TSMC doesn't have the manufacturing capacity to take Intel on as a client, and Samsung's nodes aren't viable for a processor of that size, so Intel would sell off its fabs to then buy time on them at a higher price like AMD did for GoFlo and unlike AMD at that time Intel is not hurting for cash.
Intel's only major strength is its design and manufacturing and they know it and they are leaning toward it. They have something like 5 new fabs being built in the US and Europe right now ($50B a pop) and I want to say 3 of those are going to be for 3'rd parties.
Intel is looking to get into making other people's chips like TSMC does because that is the future, Intel knows x86's days are numbered, maybe it's a decade out but if they don't prepare accordingly and work purely reactionary by the time it's time to react there won't be a place left for them.
So really it is quite the opposite, Intel is preparing for a future where its engineering and manufacturing are the core business and the x86 side of things is legacy.
TSMC has better nodes though, right?

NVIDIA chips are pretty sophisticated to be honest
 
You can have all the fabs in the world and it's worthless if you can't get the substrates and the rest of the products needed to produce CPU's. Until US industry/government puts the effort in to make sourcing everything within the US there are still national security implications for semiconductors.
 
TSMC has better nodes though, right?

NVIDIA chips are pretty sophisticated to be honest
TSMC 4 and Intel 4 are about on par, but TSMC 4 is actually producing products while Intel is still 6 months out, TSMC 3 and Intel 3 look to be about on par but TSMC 3 is now producing products but Intel won't have theirs up until 2025'ish.
Intel won't be able to claim the better node until their 18A nodes are online and only if they manage to stay on schedule, which they are looking good on. TSMC does have an answer to their 18A node, but ASML is still working on that hardware so much of Intel's schedule is based on their schedule and they have basically purchased all of it that ASML can manufacture for a solid 18 months, so they will retake node superiority then but only because TSMC physically can't purchase the machinery to compete due to the fact Intel bought it all.
 
  • Like
Reactions: erek
like this
TSMC 4 and Intel 4 are about on par, but TSMC 4 is actually producing products while Intel is still 6 months out, TSMC 3 and Intel 3 look to be about on par but TSMC 3 is now producing products but Intel won't have theirs up until 2025'ish.
Intel won't be able to claim the better node until their 18A nodes are online and only if they manage to stay on schedule, which they are looking good on. TSMC does have an answer to their 18A node, but ASML is still working on that hardware so much of Intel's schedule is based on their schedule and they have basically purchased all of it that ASML can manufacture for a solid 18 months, so they will retake node superiority then but only because TSMC physically can't purchase the machinery to compete due to the fact Intel bought it all.
Didn’t asml get breached tho

https://hardforum.com/threads/asmls...gedly-stole-confidential-information.2025768/
 
Back
Top