Info [TPU] Intel Internal Memo Reveals that even Intel is Impressed by AMD's Progress

[Atari2600]

No it's not about that.
Steve Collins isn't even employed by intel as far as I can tell.

HAHAHAHAHAHAHA

Pathetic.

 

[lobz]

Interesting... Wonder what quantities that is going to be available in.

2H 2020 at the earliest, and in low quantities.

 

[lobz]

Of course vastly superior Conroe had nothing to do with Intel dominating the market. It was only due to cheating, (or sour grapes from the AMD camp).

duuuude, hold back your horses a little, I don't know if you were already around when AMD had _really vastly_ superior uarch and products overall, and intel in fact cheated A LOT, it was kinda a well known fact, some of it was proven even in court. Conroe came years after that - where are the sour grapes?

 

[DarthKyrie]

Woooowww, who leaked this gem? Somebody's gonna get smacked hard if they get caught.

Mentioning ICX seems interesting! I have a feeling they're only talking about 26c IceLake-SP though (bleh).

Practically no mention of 10nm failures. Heh.

You can summarize that entire blurb by saying, "Look guys: they have a better process and they can sell their best uarch to the public, right now. We are stuck on 14nm and can't sell Sunny Cove or Willow Cove except in low quantity at gimped clockspeeds. Any questions?"



Lisa Su
Lisa Su
Lisa Su

Okay, where's my 2P Rome system?

Sorry man, it only works on people that are afraid of Dr. Su, not people that want the product.

Not sure if you realize this, but more than 1 person can have the same name and work in the same industry.

Steve Collins,
Director of Data Center Competitive Marketing at Intel

Just like the leaked memo says. He's worked there for 22 years.

Something tells me that theELF just doesn't want to believe that Intel would put out something like this internally because he knows that Intel has something ready to release in a week that will blow AMD out of the water. Please ignore the fact that Intel has no clue about what this CPU is that theELF thinks they have.

 

[PotatoWithEarsOnSide]

Whilst you should always address the elephant in the room, I don't think the same applies to theELFiphant in the room. His Intel bias is astounding to say the least.
There are plenty of ways to say that AMD aren't awesome, but he invariably finds ridiculous ways of doing so.

 

[Thunder 57]

Whilst you should always address the elephant in the room, I don't think the same applies to theELFiphant in the room. His Intel bias is astounding to say the least.
There are plenty of ways to say that AMD aren't awesome, but he invariably finds ridiculous ways of doing so.

I will never understand fanboyism, it's like a disease. This isn't supposed to be like sports, where having a side is fun.

 

[maddie]

Sorry man, it only works on people that are afraid of Dr. Su, not people that want the product.




Something tells me that theELF just doesn't want to believe that Intel would put out something like this internally because he knows that Intel has something ready to release in a week that will blow AMD out of the water. Please ignore the fact that Intel has no clue about what this CPU is that theELF thinks they have.

Instant hire?

 

[DrMrLordX]

Sorry man, it only works on people that are afraid of Dr. Su, not people that want the product.

Aww crud. Why can't they hire the Candyman?

Instant hire?

Intel covets NostaSeronX. They know they can't handle him, though.

 

[moinmoin]

Zen 3 is 7nm+ so I don't think they will be able to increase core count on that one. If Zen 4 uses 5nm, then they might be move up the core count to 12-cores per chiplet.

TSMC's 7nm+ is not simply an update to their 7nm but their first process using EUV. Officially it increases transistor density by 20% and decreases power consumption by 10%. For TSMC the direct update to 7nm is 6nm, the direct update to 7nm+ is 5nm (first process fully on EUV). I can imagine AMD using the little space saving afforded by 7nm+ to add 50% more chiplets (Ryzen up to 3 chiplets, Epyc up to 12) if they want to do that.

I kinda expect Intel to use chiplet like design starting with Sapphire Rapids in 2021, with their own version of IF like solution.

Sapphire Rapids had been on Intel's roadmaps for ages, well predating Ryzen's launch afaik. It since having changed to a Zen like MCM/IF or even a Zen 2 like chiplet/IOC/IF design seems optimistic.

 

[jpiniero]

Sapphire Rapids had been on Intel's roadmaps for ages, well predating Ryzen's launch afaik. It since having changed to a Zen like MCM/IF or even a Zen 2 like chiplet/IOC/IF design seems optimistic.

SR has always been rumored to be the first Core Server product designed around EMIB. It may not end up being so because they hacked up something together for Cooper Lake, but that was the original intent. How many tiles SR might have remains to be seen though.

 

[moinmoin]

SR has always been rumored to be the first Core Server product designed around EMIB. It may not end up being so because they hacked up something together for Cooper Lake, but that was the original intent. How many tiles SR might have remains to be seen though.

The interesting thing is that regarding EMIB and Foveros, all Intel is talking about so far (since 2016!) is about being able to mix and match different IP blocks using different nodes, and for mobile oriented use cases of saving space and improving energy efficiency. Even the hiring of Jim Keller (in 2018) was primarily for SoC, creating a way to integrate "heterogeneous process and architectures" as Intel's news release put it.

 

[chrisjames61]

Of course vastly superior Conroe had nothing to do with Intel dominating the market. It was only due to cheating, (or sour grapes from the AMD camp).

Sour grapes? Your post reeks of it.

 

[chrisjames61]

No it's not about that.
Steve Collins isn't even employed by intel as far as I can tell,it's a co-founder of Havok,so even if this article actually was shared on intel Circuit News it has nothing to do with what intel thinks about this.
This is basically a listing of how the laymen see the competition right now,if intel posted this it was to show it's employees how "the media" can shape opinions.

Yup, there is only one Steve Collins in the world.

 

[jpiniero]

Even the hiring of Jim Keller (in 2018) was primarily for SoC, creating a way to integrate "heterogeneous process and architectures" as Intel's news release put it.

Right, but we are talking about just Core itself, not all of the other chips Intel's been designing. And that's been the rumor for Saphhire Rapids (Core EMIB). Now I wouldn't expect the chiplets to be the extent that AMD is doing until the one after it, Granite Rapids, which uses Foveros.

 

[JasonLD]

TSMC's 7nm+ is not simply an update to their 7nm but their first process using EUV. Officially it increases transistor density by 20% and decreases power consumption by 10%. For TSMC the direct update to 7nm is 6nm, the direct update to 7nm+ is 5nm (first process fully on EUV). I can imagine AMD using the little space saving afforded by 7nm+ to add 50% more chiplets (Ryzen up to 3 chiplets, Epyc up to 12) if they want to do that.

While they is a chance that they will increase chiplet count for Epyc, I don't think they will increase chiplet count for Ryzen. 7nm+ is likely give higher maximum frequency boost than 7nm, so they will go for that instead.

Sapphire Rapids had been on Intel's roadmaps for ages, well predating Ryzen's launch afaik. It since having changed to a Zen like MCM/IF or even a Zen 2 like chiplet/IOC/IF design seems optimistic.

Since Sapphire Rapids is 2021 product, even if Intel might have gotten inspiration from AMD to go MCM, 4 years should be enough to start the design from scratch to production. Intel also probably knows the limitation and workaround to offset the higher latency with MCM design, I wouldn't be surprised to see Intel doing something similar to Zen 2 design.

 

[Atari2600]

Yup, there is only one Steve Collins in the world.

and he is gonna box the head off theElf for being silly.

 

[lobz]

and he is gonna box the head off theElf for being silly.

that is actually hilarious

 

[positivedoppler]

There use to be amd groom and doom posts here all the time, I think the ring leader was Shinkaidaik(sp?) whatever happen to all those?

 

[lobz]

There use to be amd groom and doom posts here all the time, I think the ring leader was Shinkaidaik(sp?) whatever happen to all those?

2019 happened

 

[Mopetar]

There use to be amd groom and doom posts here all the time, I think the ring leader was Shinkaidaik(sp?) whatever happen to all those?

Perhaps they entered a Zen-like state and decided to tone it down.

 

[lobz]

Perhaps they entered a Zen-like state and decided to tone it down.

Hilarious! I should have written: yeah they seem to be pretty Zen about it now.

 

[DrMrLordX]

There use to be amd groom and doom posts here all the time, I think the ring leader was Shinkaidaik(sp?) whatever happen to all those?

That was Shintai. He jumped ship with maybe 1-2 others.

 

[Markfw]

That was Shintai. He jumped ship with maybe 1-2 others.

good riddance..... He was a pain to deal with.

 

[lobz]

good riddance..... He was a pain to deal with.

absolutely agree... I love arguing about tech even when opinions differ, but I really hate when all I get to do in a conversation is to try straightening twisted facts out

 

[positivedoppler]

absolutely agree... I love arguing about tech even when opinions differ, but I really hate when all I get to do in a conversation is to try straightening twisted facts out

He would throw out these 1 line short bombs designed to borderline troll and get a reaction. I think even his cat avatar was designed to irritate people. The trolling was so well done, I always suspected they had financial interest in seeing AMD fail.