Will AMD support AVX-512 and Intel TSX ?

[tamz_msc]

Correct your math after this...
44 mm² => 4 cores + L2 + L3
L3 = 16 mm²
L2 = 1.5 mm² per core.
Zen -- ((44 - 16)/4) - 1.5 => 5.5 mm² per core without L2.
Jaguar => 3.1 mm² per core without L2.

~3.6x more dense for a 3x increase in perf?

I think the 44 sq mm figure is for the cores only.

 

[ThatBuzzkiller]

I think the 44 sq mm figure is for the cores only.

This is correct, the die space only accounts for the cores with the cache not included ...

 

[NostaSeronx]

I think the 44 sq mm figure is for the cores only.

This is correct, the die space only accounts for the cores with the cache not included ...

https://en.wikichip.org/wiki/intel/microarchitectures/skylake#Core_Group

~50.354 mm² (est) vs 49 mm² of:

Competitor A(*cough*Skylake*cough*) = Cores + L2 + L3 ((49 - 19.1)/4) - 0.9 => 6.575 mm² for 3 FP AVX256(FMAC+FMAC+FMISC)
Zen = Cores + L2 + L3. ((44 - 16)/4) - 1.5 => 5.5 mm² for 4 FP AVX128(2xFADD+2xFMUL+2xBridge(etc))

 

[Topweasel]

This is why we won't see AVX512 on consoles. From the Skylake-X review.

AVX-512 register file could probably fit a whole Atom core

 

[krumme]

This is why we won't see AVX512 on consoles. From the Skylake-X review.

Yeaa. Crazy stuff. Avx 512 is costly like nothing else.

 

[NTMBK]

Given what we know about the AVX-512 units in Knights Landing, we also know they are LARGE. Intel quoted to us that the AVX-512 register file could probably fit a whole Atom core inside, and from the chip diagrams we have seen, this equates to around 12-15% of a Skylake core minus the L2 cache (or 9-11% with the L2). As seen with Knights Landing, the AVX-512 silicon takes up most of the space.

http://www.anandtech.com/show/11550...-core-i9-7900x-i7-7820x-and-i7-7800x-tested/3

And Skylake wasn't a small core to begin with...

 

[ThatBuzzkiller]

Pretty much in line with what I had expected ...

It would relatively double the size of small cores but with some shrinks in the future that should mitigate the issue ...

 

[Topweasel]

Pretty much in line with what I had expected ...

It would relatively double the size of small cores but with some shrinks in the future that should mitigate the issue ...

How many shrinks do you think are coming?

 

[ThatBuzzkiller]

How many shrinks do you think are coming?

That depends on what you mean by 'shrinks' ...

If you mean iterative transitions half nodes then there will be lots of those and if you mean a generational transition to full nodes we could probably do maybe 4/5 more ... (at least 3 more times)

 

[JDG1980]

Intel is segmenting AVX-512 like crazy:

And it doesn't look like it's going to be in Coffee Lake at all. Given the small installed base and the degree of fragmentation indicated here, it's hard to imagine many developers investing in this. Besides, there are really only a small handful of applications that are parallel enough to benefit from AVX-512, yet also non-linear enough that they can't be offloaded to GPGPU.

I'm sure there are some use cases for this feature, but it will never become mainstream.

 

[krumme]

It looks to me avx512 is dragging Intel into hell on desktop as well as serverside. They comfortably win a small niche in the hpc segment and then gets dragged down for practically 99.9% of the rest due to insane overprovisioning insane mm2 cost and efficiency loss.

 

[krumme]

Its mesh at Intel right now.
Nobody is talking to each other.
No direction and calmness.
Canard was right.

Countdown for new ceo in 2018.

 

[formulav8]

Countdown for new ceo in 2018.

I honestly expected him to be gone already.

 

[jpiniero]

I was sort of thinking that Intel's plan was to use Skylake-SP to enter new markets and to fight POWER; while using Cannonlake Xeon D to deal with any theoretical ARM servers (and AMD if they got close).

Except (whoops) 10 nm yield sucks, so they had to kill off the Xeon D.

 

[krumme]

I think a problem with an organization like Intel is you have the ability to r&d nearly whatever you wish. The possibilities seems extreme.
Not many relevant choices and priorities is not open for you.
What do you take as strategic direction then? What is your core?
It seems counterintuitive but in a small company or eg very cash constrained situation like amd was in, only a few possibilities is actually open.
Then its far more easy to see your own core competences and what to make of them. And a lot of product development is also fairly simple.
Ryzen and IF is in hindsight actually pretty straightforward. Its about the only good way to go. They didnt have to worry eg about IOT, Power, avx512 and big dies. It was not relevant - at all.
It looks like threadding a needle. And it is an art comming up from the hole the dug for themselves. But it wasnt impossible to find the way - because most other ways was shut down.