Discussion Intel current and future Lakes & Rapids thread

[TheF34RChannel]

So this is delayed from the August release?

The following is not meant to you in particular but to anyone who has asked what I made bold in the quote. I chose to make the reply stand out so that's easy to find for anyone else curious about this so it's not an offense

There is NO delay. CFL-S will very likely get an October hard launch. CFL was brought forward from 2018, meaning that every date before that is still a quicker launch than originally planned. The August date came from an Intel slide which now looks to be the announcement/paper launch date, WE, the people of earth, suspected an August hard launch; nothing was said or leaked by Intel itself. I cannot make it any clearer than this, and one doing its own bit of research would have led to the exact same conclusion laid out above.

I doubt EMIB will debut with Ice Lake. It'll arrive with the 7nm gen, IMHO.

Wishing it was different, I have to admit you're probably right :/ It also makes sense from a developing standpoint since they'll need a few years designing it and although they have started (these things start quite a few years before we even hear about it), they are still adding more people to the research, as per Intel's job listings.

 

[Jan Olšan]

I'm really curious to see what will happen with Ice Lake. Some people here expected +15% IPC from Skylake, which never happened (maybe somewhere in gaming performance, but I don't really feel like that is a measure of orthodox CPU performance, more like memory controller/caches). It certainly wasn't as much of a tick as Sandy or Haswell were.

So 5% progress is probably in realm of possibility too, although that is barely more than some past "tick" IPC improvements (Penryn, Ivy). I would certainly not underestimate Intel though, it well could be those 15 % too.

I just have no idea which of those options to expect - or where between them is the realistic mean value.

 

[tential]

So is CFL still theorized to come in August or not?

 

[Arachnotronic]

So is CFL still theorized to come in August or not?

That's the latest rumor, yeah.

 

[jpiniero]

Wishing it was different, I have to admit you're probably right :/ It also makes sense from a developing standpoint since they'll need a few years designing it and although they have started (these things start quite a few years before we even hear about it), they are still adding more people to the research, as per Intel's job listings.

EMIB lets Intel crank up the segmentation machine up to 11. Especially with the way Intel is acting it's the kind of thing that Intel would 'reprioritize' for. Intel already has a product on the market using EMIB so they've had the tech for some time. Full exploit of it likely won't come until Saphhire Rapid, yes, but for now keeping the tiles small will do.

IIRC Intel said Tigerlake-SP is EMIB aware. The question to me is whether Icelake mainstream is also EMIB aware.

 

[TheF34RChannel]

So is CFL still theorized to come in August or not?

That's the latest rumor, yeah.

October is more likely, also, it looks like there won't be any boards before then so even if Intel does launch next month, where will you put it in? Still hoping this is dead wrong and I'll have both a CPU and board in my paws in a few weeks.

EMIB lets Intel crank up the segmentation machine up to 11. Especially with the way Intel is acting it's the kind of thing that Intel would 'reprioritize' for. Intel already has a product on the market using EMIB so they've had the tech for some time. Full exploit of it likely won't come until Saphhire Rapid, yes, but for now keeping the tiles small will do.

IIRC Intel said Tigerlake-SP is EMIB aware. The question to me is whether Icelake mainstream is also EMIB aware.

What current product is using EMIB? I'm clueless, I admit.

With TGL being a shrink of ICL I'd say that if TGL-SP is EMIB then so is ICL-SP. However you wondered about mainstream ICL, and although I'd like to say yes it might be a different design altogether. It's certainly something to keep eye out for, answers to such questions I mean.

 

[jpiniero]

What current product is using EMIB? I'm clueless, I admit.

The FPGAs.

 

[witeken]

I doubt EMIB will debut with Ice Lake. It'll arrive with the 7nm gen, IMHO.

Wishing it was different, I have to admit you're probably right :/ It also makes sense from a developing standpoint since they'll need a few years designing it and although they have started (these things start quite a few years before we even hear about it), they are still adding more people to the research, as per Intel's job listings.

That makes no sense. Intel is not the company that starts talking about technology in a big way 3+ years before release. EMIB will clearly debut with Icelake. EMIB is in production right now with Stratix 10, so the technology is developed and ready.

 

[witeken]

I'm really curious to see what will happen with Ice Lake. Some people here expected +15% IPC from Skylake, which never happened (maybe somewhere in gaming performance, but I don't really feel like that is a measure of orthodox CPU performance, more like memory controller/caches). It certainly wasn't as much of a tick as Sandy or Haswell were.

So 5% progress is probably in realm of possibility too, although that is barely more than some past "tick" IPC improvements (Penryn, Ivy). I would certainly not underestimate Intel though, it well could be those 15 % too.

I just have no idea which of those options to expect - or where between them is the realistic mean value.

Intel seems to target 10% with Tocks and 5% with Ticks.

 

[Arachnotronic]

That makes no sense. Intel is not the company that starts talking about technology in a big way 3+ years before release. EMIB will clearly debut with Icelake. EMIB is in production right now with Stratix 10, so the technology is developed and ready.

*cough* 3D Xpoint *cough*

 

[Ajay]

That makes no sense. Intel is not the company that starts talking about technology in a big way 3+ years before release. EMIB will clearly debut with Icelake. EMIB is in production right now with Stratix 10, so the technology is developed and ready.

Still, I don't really see EMIB as an advantage in client CPUs yet (die sizes are still small enough). Clearly, for Icelake-SP, it would significantly improve yields and possibly top clocks. Base 12 core could be used for 1x, 2x, 3x and 4x for an up to 48 core option. Core could be run on 10nm and uncore kept at 14nm - and a zillion other possibilities. [Just spit balling here].

 

[jpiniero]

Still, I don't really see EMIB as an advantage in client CPUs yet (die sizes are still small enough).

Given how bad yields must be at 10 nm I am not sure this is true. Either way the purpose of using EMIB with Icelake mainstream would be for Xeon-D (have 4 8 core tiles) and of course being able to segment the GPU harder than they do now. I also think they would want to get some practice in before they deal with the SP servers.

 

[raghu78]

That makes no sense. Intel is not the company that starts talking about technology in a big way 3+ years before release. EMIB will clearly debut with Icelake. EMIB is in production right now with Stratix 10, so the technology is developed and ready.

EMIB and data center first strategy jointly debut at 10++ with Tigerlake.

https://www.fool.com/investing/2017/02/12/intel-corporation-plans-seismic-change-to-data-cen.aspx

"According to Bryant, the first products to use Intel's 10-nanometer++ technology -- its third-generation 10-nanometer technology -- as well as the first products to use Intel's 7-nanometer technology will be data-center chips, rather than personal-computer chips."

"To get around this problem, Bryant explained, Intel is using a new silicon packaging technology called Embedded Multi-Die Interconnect Bridge, or EMIB for short.

EMIB, Bryant said, "allows [Intel] to cut up the Xeon die and debug just a piece of it at a time."

"So that combination is allowing us to go first on a next-generation process node," she said

 

[Eug]

In my experience, Intel Turbo Boost 2.0 on desktops runs indefinitely in real-world apps. The power limit (by default equal to TDP) is typically only reached using something like AIDA64 FPU test or Prime95 small FFT.
Another situation is simultaneously stressing the CPU and integrated GPU. After I raised power limit to 120 W, I ran Prime95 small FFT and FurMark on Core i7-7700, and CPU package power was 100 W.
Using stock cooler, CPU temperature at 100 W was also 100 C, although the chassis was ASRock DeskMini 110, not the most open chassis. Thermals won't be an issue at 95 W, if you consider not being an issue to be no frequency throttling.

IIRC, some people are now saying that under very heavy and extended loads the i7 7700K can occasionally throttle to 3.9 GHz multi-core in the iMac. Interestingly, that is the exact speed of the i5 7600 for multi-core Turbo, which never throttles in the same form factor.

BTW, is this leak accurate?

http://www.guru3d.com/news-story/intel-six-core-lineup-exposed-core-i7-8700k-and-core-i5-8600k.html

Curiously, the only 65 Watt Coffee Lake S part mentioned is the i5 8400 with a base clock speed of 2.8 GHz. Just by back-of-the-napkin calculation and estimates, it seems to me this part would perform similarly for multi-threaded applications to the 4-core i7-7700 (non-K), and would perform much worse in single-threaded applications obviously. Furthermore, my i5 7600 (non-K) might only be about 10-20% slower in multi-threaded applications. If so, that's a bit disappointing for the 65 W 6-core Coffee Lake, although even if those clock speeds are accurate, there is a conspicuous absence of the i5 8500 in that article. Perhaps there is a so far unleaked i5 8500 65 Watt part that is higher clocked. A binned 65 Watt i5 8500 non-K 6-core at say 3.2 GHz with multi-core at 3.5 GHz would be a great chip.

 

[TheF34RChannel]

@Eug yes that leak is accurate. If, however, those are the finalized clocks remains to be seen. @Sweepr knows more but he's teasing us ha ha ha! Drop the bomb Mr Sweepr!

 

[StinkyPinky]

3.7 base for the 6/12 is slightly underwhelming if it stays like that, although I suppose as long as it can overclock to 4.5 I'd take it. Would be a good strong upgrade for my 4790K.

 

[coercitiv]

3.7 base for the 6/12 is slightly underwhelming

Slightly underwhelming?! You get a guaranteed 20-30% total throughput increase over Kaby Lake, IMHO it's excellent news. ST and low threaded performance will also get a boost.

 

[TheF34RChannel]

3.7 base for the 6/12 is slightly underwhelming if it stays like that, although I suppose as long as it can overclock to 4.5 I'd take it. Would be a good strong upgrade for my 4790K.

Slightly underwhelming?! You get a guaranteed 20-30% total throughput increase over Kaby Lake, IMHO it's excellent news. ST and low threaded performance will also get a boost.

I agree with him saying it's underwhelming as I believe we should be past the sub 4.0GHz base clocks by now, especially on 14nm++ which can do better than that and should stay within its TDP nicely. It's the simultaneous additional throughput by more cores/threads and high frequency I'm after. Sure, I can (and will) overclock it but having somewhat better specs out of the box would've looked more interesting. For this reason any base clock (because it isn't a great indicator) is ultimately unimportant, and one doesn't buy a K CPU to run it at stock IMO.

So there're conflicting statements in this reply, and that's exactly what's going on in my mind ha ha! One thing is for sure though; I will get me some top SKU Coffee Lake-S madness!

 

[Arachnotronic]

I agree with him saying it's underwhelming as I believe we should be past the sub 4.0GHz base clocks by now, especially on 14nm++ which can do better than that and should stay within its TDP nicely. It's the simultaneous additional throughput by more cores/threads and high frequency I'm after. Sure, I can (and will) overclock it but having somewhat better specs out of the box would've looked more interesting. For this reason any base clock (because it isn't a great indicator) is ultimately unimportant, and one doesn't buy a K CPU to run it at stock IMO.

So there're conflicting statements in this reply, and that's exactly what's going on in my mind ha ha! One thing is for sure though; I will get me some top SKU Coffee Lake-S madness!

I care about all-core turbo, not base clocks.

 

[TheF34RChannel]

I care about all-core turbo, not base clocks.

I get that completely! It's the second part of my internal debate, that the all-core turbo is what matters. And why I am dying to know what the frequency is!

In addition to your two X299 systems, you're getting CFL-S too?

 

[Arachnotronic]

I get that completely! It's the second part of my internal debate, that the all-core turbo is what matters. And why I am dying to know what the frequency is!

In addition to your two X299 systems, you're getting CFL-S too?

Not getting CFL-S, nope. Just watching

 

[TheF34RChannel]

Not getting CFL-S, nope. Just watching

You lurker The same way I'm watching X299, I suppose; pure new tech interest.

 

[Arachnotronic]

You lurker The same way I'm watching X299, I suppose; pure new tech interest.

Yep. Also if CFL-S is a real killer, I'll swap out the 7800X for a CFL-X

 

[scannall]

I care about all-core turbo, not base clocks.

I wouldn't care if the base was 1 Ghz while doing nothing, and sipping just a few watts of power. Focusing on base clocks is a bit misleading.

 

[TheF34RChannel]

Yep. Also if CFL-S is a real killer, I'll swap out the 7800X for a CFL-X

He he nice move if it happens!

I wouldn't care if the base was 1 Ghz while doing nothing, and sipping just a few watts of power. Focusing on base clocks is a bit misleading.

You guys ended my internal debate and rammed some sense into me, thanks