MorphCore paper - rumored new architecture in Skylake

[VirtualLarry]

http://hps.ece.utexas.edu/people/khubaib/pub/morphcore_micro2012.pdf

 

[ShintaiDK]

Already mentioned in both the Skylake threads.

Its a silly wccftech rumour.

 

[jpiniero]

Already mentioned in both the Skylake threads.

Its a silly wccftech rumour.

That is included in Skylake is a rumor, but I imagine we will see this eventually from Intel. It sounds more of a power saving technique than a performance boost though, so it wouldn't really help top end desktops although it would help perf/W.

 

[Arachnotronic]

WCCFTech literally made this "rumor" about Skylake up with absolutely no evidence and no sources.

It's Faraway Islands all over again.

 

[witeken]

WCCFTech literally made this "rumor" about Skylake up with absolutely no evidence and no sources.

It's Faraway Islands all over again.

How about the NDA thing they're talking about?

 

[Arachnotronic]

How about the NDA thing they're talking about?

If you're under NDA, you don't tell people you're under NDA.

 

[witeken]

If you're under NDA, you don't tell people you're under NDA.

That's false, obviously. Here's the quote..

One of the surprising things that our source mentioned was that Intel has always been fairly lenient about information after the NDA is signed. However, this time, for the first time in many years, they simply refuse to divulge the slightest information even under NDA. This ‘above top secret’ attitude seems out of place since the process was already introduced with Broadwell and is supposed to be just a ‘Haswell-equivalent’ for Broadwell. Something that definitely appears to not be the case.
The level of secrecy Intel is maintaining makes it very clear that they are bringing something brand new with the Skylake uarch.

 

[jpiniero]

That's false, obviously. Here's the quote..

Obviously Intel's changed their policy on releasing any kind of future info. This in no way reflects how they think of Skylake.

 

[Arachnotronic]

That's false, obviously. Here's the quote..

WCCFTech is most likely making stuff up. An example...

Now here is where it gets interesting, remember what I told you about AMD’s Zen? Our source stated that he is fairly certain AMD knows about this ‘new’ feature and will try to incorporate it into Zen high performance micro-architecture.

Right off the bat this kills any credibility because CPU IP definitions, especially for designs expected to ship in 2016, can't be modified this late in the game without a significant delay. Zen's definition is set in stone at this point, and anybody telling you that AMD (or any other company) is going to mid-flight make sweeping changes to a microarchitecture and then go through the costly and time consuming process to validate the core because they heard that Skylake might have it is making things up and doesn't have enough understanding of how things work to make things up that make sense.

 

[ShintaiDK]

That's false, obviously. Here's the quote..

Please dont quote wccftechs homemade rubbish. They have zero credibility.

Their entire concept is build up on gossip and newsstealing to make people click their site.

 

[jdubs03]

Yea I place no trust in that site, but I could see how Intel may want to utilize the MorphCore approach (or improvement on it) in a future uArch.

I read through some of it and from my limited knowledge it looks pretty promising in getting some extra performance and lower power usage.

 

[SAAA]

That is included in Skylake is a rumor, but I imagine we will see this eventually from Intel. It sounds more of a power saving technique than a performance boost though, so it wouldn't really help top end desktops although it would help perf/W.

Actually if you look at the paper is a serious deal: a full study of performance of various hypothetical architecture changes like 4 way SMT vs 2 way SMT, smaller but slower cores and this Morph thing too. Research funded by Intel themself probably.
Apparentely it's the best way to preserve single thread performance of an already extremely optimized architecture like the core line but also get a boost (20% average) when smaller multiple threads are loaded and you don't want the complexity of a 4-way SMT, plus it saves a lot of power.

WCCFTech literally made this "rumor" about Skylake up with absolutely no evidence and no sources.

It's Faraway Islands all over again.

The rumor may be dubious but the tech is definitely there and will be used one day, unless another more efficient option comes to mind.
It would also be in line with the rumored Skylake's great power savings (while still using 14nm process) and the better multi-thread scores I posted of a leaked Geekbench result.

 

[TechFan1]

I don't know about the morph core, but the rumor about the OEMs wanting all the different types of skylake to ship at the same time does suggest that there might be something special about skylake.

 

[III-V]

Actually if you look at the paper is a serious deal: a full study of performance of various hypothetical architecture changes like 4 way SMT vs 2 way SMT, smaller but slower cores and this Morph thing too. Research funded by Intel themself probably.
Apparentely it's the best way to preserve single thread performance of an already extremely optimized architecture like the core line but also get a boost (20% average) when smaller multiple threads are loaded and you don't want the complexity of a 4-way SMT, plus it saves a lot of power.

The rumor may be dubious but the tech is definitely there and will be used one day, unless another more efficient option comes to mind.
It would also be in line with the rumored Skylake's great power savings (while still using 14nm process) and the better multi-thread scores I posted of a leaked Geekbench result.

As I mentioned in another thread, I believe Skylake does not use MorphCore -- instead, I attribute the better MT performance to the (rumored, though likely) fact that Skylake shares L2 between groups of two cores, as well as having a better inter-SoC communication fabric (the 2D mesh).

I think MorphCore would result in a more substantial performance uplift -- it was 22% in the paper for MT by virtue of implementing MorphCore alone, with all else being constant. Geekbench's results do not support that at all. Also, I cannot imagine that MorphCore is ready -- the paper was published in 2012. I'd also think its first implementation would be a work of Hillsboro's team, e.g. Icelake, considering that was where part of the support came through.

Skylake looks interesting, in the sense that it may be a bigger deal than Ivy Bridge and Haswell were. Unlike Sandy Bridge however, it may not get a clock speed bump over its predecessors (and that's where its improvement really came from -- it was only about 10% faster per clock, just like Haswell).

I'm not totally convinced, though. I feel like the more I learn about Skylake, the less excited I get. There are still way too many unknowns to be making bets with any sort of certainty, though. In particular, I am wondering if Intel's revamped Skylake's clock trees...

 

[IntelUser2000]

Unlike Sandy Bridge however, it may not get a clock speed bump over its predecessors (and that's where its improvement really came from -- it was only about 10% faster per clock, just like Haswell).

What?... no.

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

You see a huge gap between i7 965 and i7 2700K and very small one between i7 3770k and i7 4770k. The thing disappointing about this chip is that it even fell significantly below 10% estimation Intel gave.

Clock-normalized:

SNB vs NHLM and HSW vs IVB
19.8% vs. 4.7%
14% vs. 6.0%
14.3% vs. 13.0%
7.4% vs. 1.2%
2.0% vs. 1.3%
22.1% vs. 2.2%
0.8% vs. 0.8%
13.5% vs. 5.7%
12.5% vs. 13.0%
9.9% vs. 14.9%

I think its pretty clear which has a greater gain, by far too. Sandy Bridge also added higher clocks, better Turbo sustainability, better overclocking, lower power, *and* cheaper price. Haswell had same clocks, same Turbo, worse overclocking, higher power, same price.

I hope Skylake reduces the clock from 4GHz for a halo version like the 4790K down to 2.5GHz, but improve the IPC by 80%. Wishful thinking, but that would much better fit their lower voltage optimized process, and give them headroom for clocks later. That would improve perf/watt vastly in the mobile SKUs I assume too.

 

[t0mt0m]

Seeing as Khubaib has been at Apple's CPU design group for a year now (with his doctorate thesis on Morphcore a year old, a previous paper on Morphocre from 2012) - Morphcore or some semblance to it could turn up in their A series chips as much as have an influence on Intel...

 

[mavere]

wccftech: "BREAKING: iPhone 7's new CPU architecture uncovered in detail"

 

[BD231]

It'll take another Athlon to get a new arch from Intel.

 

[Hulk]

It'll take another Athlon to get a new arch from Intel.

It's a bingo!

 

[Arachnotronic]

It'll take another Athlon to get a new arch from Intel.

Intel releases new architectures all the time.

 

[Dave2150]

Intel releases new architectures all the time.

He obviously means completely different architecture from the current 'Core' architecture that's been around for years.

You knew this too, I assume you just want to be pedantic as you have nothing better to do.

 

[ShintaiDK]

What would a new uarch solve?

 

[Nothingness]

What would a new uarch solve?

A radically new uarch might bring more performance or more power efficiency, who knows?

OTOH variations around out of order executions won't change things a lot, especially given how Intel (and IBM) have been refining this for decades.

 

[scannall]

What would a new uarch solve?

Good thing they didn't say that about Netburst...

Obviously you keep the bulk of your engineers on products that actually earn you money. But having a small team looking into new things is always good.

 

[ShintaiDK]

Good thing they didn't say that about Netburst...

Obviously you keep the bulk of your engineers on products that actually earn you money. But having a small team looking into new things is always good.

Yep. And I am sure they already do that. To get some kind of "magic solution" we need something extremely different like IA64 style.

And thats not going to happen due to the market.