Discussion AWS Graviton2 64 vCPU Arm CPU Heightens War of Intel Betrayal

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

ksec

Senior member
Mar 5, 2010
420
117
116
This one is easy: they tried and failed miserably because their design rules are so tied to their process that porting any existing design to their process is a royal pain. Add to that their first serious trial (with LG) was supposed to be on 10nm. Boom.

That was the *technical* excuses / explanation given. In reality this isn't really the case. The Custom Foundry has different node with different design rules. So it isn't really the 10nm used by Intel CPU. The truth is Custom Foundry just isn't in Intel's DNA. They will not share, open, or work with that many manufactures and prefer to keep everything to themselves. Compared to TSMC where everyone is a "partner" in the Ecosystem. For Intel, you are nothing more than a Custom Foundry "Customer".

I blame it on Andy Bryant.
 

ksec

Senior member
Mar 5, 2010
420
117
116
TSMC has a slightly bigger market cap than Intel right now. Imagine if Intel saw TSMC's business opportunity. Intel would be worth twice as much right now.

Well 1st. Market Cap is actually meaningless without looking at Profits and Revenue. Intel is ridiculously profitable. If Intel were rated with the same P/E as TSMC Intel would more than double its current market cap.

2nd Is the conflict of Interest. The reason why people trust TSMC is because TSMC do not compete with them. TSMC do not sell CPU / GPU/ I/O Controller/ Network Controller / FPGA / SSD Controller / Modem / Embedded SoC etc....
Intel does all of the above. i.e Why would Nvidia or AMD Fab with Intel to compete with their own CPU and GPU?

3rd, Fabs requires economy of scale, the key reason why TSMC managed to catch up to Intel is the sheer volume driven by the Smartphone revolution. Unless Intel decide to drop their product business they will never be able to achieve that scale.

4th is as mentioned in the post above, culture.

There are few other smaller points, but these four are big items that cant easily be changed.
 
  • Like
Reactions: Tlh97

Nothingness

Platinum Member
Jul 3, 2013
2,371
713
136
General ARM support is good. Server infrastructure tool chains are almost non existent.
Again what infra tool chains are missing? You are aware that RedHat and SUSE, two of the main suppliers of Linux distros for infra, have had official AArch64 support for years? I'd expect they made the proper effort to port the "tool chains".

Note the from an end-user point of view that's different. Some software will need some porting and tuning. But that's not what is called tool chain.

SeaMicro, it was still a good buy as AMD uses a lot of their Interconnect technology when working on IF.
SeaMicro was doing Atom servers, not ARM servers as far as I remember.
 

Nothingness

Platinum Member
Jul 3, 2013
2,371
713
136
Sure, but that just goes to show what poor strategic thinkers there are at Intel.
If something has worked for years, why change anything? That's how companies are missing opportunities and how some die.

I'll also add that I never understood why they didn't make a far larger effort to make their own discrete GPU's from 20+ years ago.

Yes I know they had a famous failure with the i740, but rather than learn from this, they just tucked their tails between their legs and ran away.
And there was Larrabee. Intel hubris made them think that x86 is so great that software rendering with limited hardware assistance was the way to go. Great vision.

For a company Intel's size, I expect a lot more than they deliver.
Definitely and that's a pain to see given the talents they have in-house. Another painful sight is how they destroy every company they buy.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
The need was to pick up extra billions of dollars there for the taking.

As I said in my previous post, Intel badly lacks quality strategic thinkers.
I don't disagree. Intel just tried to milk what they already had, and did the same with their Custom Foundry effort.

SeaMicro was doing Atom servers, not ARM servers as far as I remember.
It started with Atom and was originally announced to be extended to ARM back in 2010. After AMD bought them with SM15000 Opteron and Xeon became an option along Atom instead.
 

mikegg

Golden Member
Jan 30, 2010
1,740
406
136
Well 1st. Market Cap is actually meaningless without looking at Profits and Revenue. Intel is ridiculously profitable. If Intel were rated with the same P/E as TSMC Intel would more than double its current market cap.

2nd Is the conflict of Interest. The reason why people trust TSMC is because TSMC do not compete with them. TSMC do not sell CPU / GPU/ I/O Controller/ Network Controller / FPGA / SSD Controller / Modem / Embedded SoC etc....
Intel does all of the above. i.e Why would Nvidia or AMD Fab with Intel to compete with their own CPU and GPU?

3rd, Fabs requires economy of scale, the key reason why TSMC managed to catch up to Intel is the sheer volume driven by the Smartphone revolution. Unless Intel decide to drop their product business they will never be able to achieve that scale.

4th is as mentioned in the post above, culture.

There are few other smaller points, but these four are big items that cant easily be changed.
1. Market cap factors in profit, revenue, potential growth, opportunities, and basically everything about the company. It's a better measurement than just profit and revenue. It's the single most important way to value a company because... it's the value of a company. It's why wildly profitable companies are often worth way less than a loss-making high growth company. Uber is an example.

2. The same reason sellers sell on Amazin despite Amazon selling their own brand of items. When your technology is so much better than the competition, customers don't have a choice. This was the case with Intel's process lead before TSMC and Samsung took the lead. In addition, Intel could have spun off the manufacturing business as a separate entity to reduce conflict of interest.

3. Intel had more scale than TSMC before the smartphone revolution. They just didn't see the opportunity.

4. I'd argue that it's not culture. It's the lack of business leadership. I'm sure Intel had internal discussions of creating manufacturing as a service business but they didn't take it seriously enough.

The point is that Intel had a great opportunity to own TSMC's business. They didn't execute on it.
 

Dribble

Platinum Member
Aug 9, 2005
2,076
611
136
And there was Larrabee. Intel hubris made them think that x86 is so great that software rendering with limited hardware assistance was the way to go. Great vision.
Intel's strategy was always around making everything x86, and because only they can make x86 (with AMD to make it look like there was competition) they would control the whole market. If you control the market you control the pricing, you control supply so you can guarantee your foundries are always fully utilised, you make all the money. That's what happened with windows PC's. Hence everything new you do has to be x86 to pull everything into that x86 monopoly.
Obviously the rest of the industry wasn't particularly enthused with this as they wanted the money not Intel, hence when the phone market appeared it was the end device maker (apple/samsung) that make all the money not the chip designer/maker (ARM/TMSC effectively). That is due to ARM's different philosophy - they sell you a design cheap and you can do what you want with it (vs x86 where you get the end chip Intel/AMD give you at the price they decide). That is just continuing - if ARM has a chip competitive in an area then the end device maker (in this case Amazon) will be able to make their own solution using them, which moves a lot of the profits from Intel to Amazon just like happened with Apple/Samsung.
 
Last edited:
  • Like
Reactions: KompuKare

SarahKerrigan

Senior member
Oct 12, 2014
339
468
136
No these guys license core designs from ARM. They still actually have to develop and have the actual CPU built. Samsung and Apple just have a license that allows them to make changes to the core design to make derivative products.

That is not what an ARM architecture license is. Neither Samsung's, nor Apple's, nor Fujitsu's, cores are "derivative products" of ARM cores, and a five-minute look into their microarchitecture would make this evident. They are fully custom microarchitectures.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
That is not what an ARM architecture license is. Neither Samsung's, nor Apple's, nor Fujitsu's, cores are "derivative products" of ARM cores, and a five-minute look into their microarchitecture would make this evident. They are fully custom microarchitectures.
This is a semantics game that I am not sure you will win. Custom microartichetures that are fully compatible with ARM. That seems derivative to me.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
You're so wrong it's hilarious :D


It's not a semantic game as it makes a large difference in money to invest in designing a SoC, you know the point you made?
No it doesn't, they aren't building a design from scratch with a new ISA. Everything x86 is an 8086 derivative. Whether or not they are spending a fortune in its design, they are standing on the work that ARM put into their designs using their ISA, using their IP. Its a derivative work. To say its not because the amount of effort they put into it, is a semantics game, because at its heart its based on someone else's work. If this is about connantation like Apple is just tweaking ARM's. That's not what I was getting at. I get that it is a massive undertaking and that especially Apple's stuff barely looks like ARM. But that's assumption is on you guys, I took a word that meant for the most part what I was looking for. It fits its definition. I am not backing from that.
 

Nothingness

Platinum Member
Jul 3, 2013
2,371
713
136
No it doesn't, they aren't building a design from scratch with a new ISA. Everything x86 is an 8086 derivative. Whether or not they are spending a fortune in its design, they are standing on the work that ARM put into their designs using their ISA, using their IP. Its a derivative work. To say its not because the amount of effort they put into it, is a semantics game, because at its heart its based on someone else's work. If this is about connantation like Apple is just tweaking ARM's. That's not what I was getting at. I get that it is a massive undertaking and that especially Apple's stuff barely looks like ARM. But that's assumption is on you guys, I took a word that meant for the most part what I was looking for. It fits its definition. I am not backing from that.
Oh I see now you change subject. You wrote that Apple and Samsung license core designs from ARM for their own CPU's. You are just plain wrong. Turn it all around you want, change the subject, you were just writing something false. You know there's nothing bad in being wrong.

In fact it seems you are unable to see the difference between ISA definition and core design. Sorry you're clueless.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
Oh I see now you change subject. You wrote that Apple and Samsung license core designs from ARM for their own CPU's. You are just plain wrong. Turn it all around you want, change the subject, you were just writing something false. You know there's nothing bad in being wrong.

In fact it seems you are unable to see the difference between ISA definition and core design. Sorry you're clueless.
No don't put words in my mouth. They took ARM's work on their design, licensed it and changed it. Its derivative. You are the one that had an assumption on what I was getting at. I wasn't wrong, you and the other guy assumed....Crap I just went back and read that. Core design didn't mean "core as in a compute unit" core design being you know the "base" design, the central design. I am probably not explaining it properly. Foundation. I was talking about ISA. They licensed the ISA with a license that specifically allowed them to make ARM derivative products. I am not changing the subject, I just used a poor word considering that we were talking about CPU design. If I said that with any other market there wouldn't have been an issue and that is on me.
 

SarahKerrigan

Senior member
Oct 12, 2014
339
468
136
No don't put words in my mouth. They took ARM's work on their design, licensed it and changed it. Its derivative. You are the one that had an assumption on what I was getting at. I wasn't wrong, you and the other guy assumed....Crap I just went back and read that. Core design didn't mean "core as in a compute unit" core design being you know the "base" design, the central design. I am probably not explaining it properly. Foundation. I was talking about ISA. They licensed the ISA with a license that specifically allowed them to make ARM derivative products. I am not changing the subject, I just used a poor word considering that we were talking about CPU design. If I said that with any other market there wouldn't have been an issue and that is on me.

Fair enough. But in this case, "core" is a word with specific technical meaning, and the cores Apple and Samsung build are not derived from the cores ARM builds, any more than Zen2 is derived from Haswell. They're just an independent compatible implementation with an architecture license.

Let's drive on; no point beating a dead horse.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
Fair enough. But in this case, "core" is a word with specific technical meaning, and the cores Apple and Samsung build are not derived from the cores ARM builds, any more than Zen2 is derived from Haswell. They're just an independent compatible implementation with an architecture license.

Let's drive on; no point beating a dead horse.
I am aware and notated that above. I knew better then to do that in a CPU thread. But I use the phrase core design a lot when I am talking about the base design, or design philosophy. "the central or most important part of something."
 

Andrei.

Senior member
Jan 26, 2015
316
386
136
I am aware and notated that above. I knew better then to do that in a CPU thread. But I use the phrase core design a lot when I am talking about the base design, or design philosophy. "the central or most important part of something."
Yes but then you talk wrong. An ISA is not a "base design". It's literally the very least important aspect determining the performance of a CPU.
 
  • Like
Reactions: Lodix and Etain05

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
Yes but then you talk wrong. An ISA is not a "base design". It's literally the very least important aspect determining the performance of a CPU.
But it is when dealing with compatibility. I didn't mention performance of the processor in that sense at all. It is a licensed product with licensed IP and therefore derivative.
 

Richie Rich

Senior member
Jul 28, 2019
470
229
76
By the way for people interested, ARM has published latency and throughput for Neoverse N1: https://static.docs.arm.com/swog309707/a/Arm_Neoverse_N1_Software_Optimization_Guide.pdf

Sorry if that was already posted :)

(And A77: https://static.docs.arm.com/swog011050/c/Arm_Cortex-A77_Software_Optimization_Guide.pdf)
Very interesting. A77 was not just about added one simple ALU and doubling Branch unit. There are a lot of operations where ARM doubled throughput:
  • Branch ………..……………………….………………………………………..… 1 -> 2 …. (+100%) .... expected 2nd Branch unit
  • Arithmetic, basic ……………………………………………………….…... 3 -> 4 …. (+33%) ...... expected 4th simple ALU, but see the others
  • Arithmetic, extend and shift ……………………………………….... 1 -> 2 …. (+100%)
  • Arithmetic, LSL shift, shift <= 4 …………………………………..… 3 -> 4 …. (+33%)
  • Arithmetic, LSR/ASR/ROR shift or LSL shift > 4 …………. 1 -> 2 …. (+100%)
  • Conditional compare, select, logical, basic ………………… 3 -> no change
  • Logical, shift, no flagset ………………………………………………… 3 -> 4 …. (+33%)
  • Logical, shift, flagset ……………………………………………………… 1 -> 2 …. (+100%)

These are pretty huge changes in A77 in compare to N1/A76. Now it's clear where the +25% IPC come from.
Super impressive for every year core delivery/cadence.
 
Last edited:
  • Like
Reactions: Tlh97 and Etain05

Richie Rich

Senior member
Jul 28, 2019
470
229
76
X86 has control over servers like Italy has control over corona-virus.
Actual numbers looks good but they don't see what a disaster is coming.
Just imagine AMD wouldn't cancel K12 - Today they would have the best ARM core on market at almost perfect time. Shame on them.
 

Nothingness

Platinum Member
Jul 3, 2013
2,371
713
136
Just imagine AMD wouldn't cancel K12 - Today they would have the best ARM core on market at almost perfect time. Shame on them.
No matter how much I would have liked to get a K12, AMD did the right thing back then, they were too weak to track two targets, and x86 still is the preferred solution for servers and end user machines.

Edit: note AMD also failed terribly with their Cortex-A57 based chip which reinforces their decision to cancel K12.
 
Status
Not open for further replies.