Discussion Intel current and future Lakes & Rapids thread

[JoeRambo]

Honestly I doubt it is worth to shed tears about Cannon Lake architecture at all. It was a "Lake" derivative, conceived in the days of Intel reigning supreme and probably included 2-3% of IPC improvement outside of AVX512 apps ( and even there i highly doubt "client" variant of CNL had full 512 width units). So from performance department we are not missing anything*

* Caveats are iGPU, that could be advanced and widened at same using die real estate on 10nm and of course unknown clocking abilities of 10nm process, but of course anything south of 5Ghz can't touch current CFL chips.

So once 10nm has failed as hard as it has with yields and probably transistor performance ( even Intel in benchmarketing slides is comparing 10nm+ vs 14nm+++ etc ), we might as well wait for fixed 10nm chips with architecture that was/is being developed in more competitive environment.

Is that Ice Lake? Only time will tell.

 

[csbin]

Cascade Lake-SP 8-socket

https://videocardz.com/newz/intel-to-offer-glueless-8-socket-scalability-with-cascade-lake

 

[dooon]

ya

 

[IntelUser2000]

Glueless 8-socket config for isn't new for Cascade Lake. Man, some press people prioritize getting news as fast as possible over anything else! Like accuracy!

It was already possible with Skylake-SP, the first chip for Purley platform. That's why the top end chip has 3 UPI links.

 

[The Stilt]

Cascade Lake-SP 8-socket

https://videocardz.com/newz/intel-to-offer-glueless-8-socket-scalability-with-cascade-lake

64-core EPYC processors

 

[jpiniero]

https://twitter.com/TMFChipFool/status/996478399794368514

Oh snap, Intel admitted that next year's server product (not Cascade Lake, the one after it) is on 14 nm.

So no Icelake Server.

 

[Dayman1225]

Go listen to the interview your self

https://jpmorgan.metameetings.net/events/tmc18/sessions/15020-intel-corporation/webcast

 

[witeken]

https://twitter.com/TMFChipFool/status/996478399794368514

Oh snap, Intel admitted that next year's server product (not Cascade Lake, the one after it) is on 14 nm.

So no Icelake Server.

Making it sound like it's big news, but does this surprise anyone? Server in the PC-first era has always been over a year later. They're much bigger dies.

 

[jpiniero]

Making it sound like it's big news, but does this surprise anyone? Server in the PC-first era has always been over a year later. They're much bigger dies.

I think there was some hope that 10++ would have real fixes for the yield problems (since they mentioned they reworked the entire metal stack), and that they would be able to release it by the end of next year. Sounds like the only real solution is EUV, and of course Intel is going to be last of the four to it.

In any case, AMD is going to really have to screw this up to not gain serious server share, especially now that they will be on a superior node for a good while.

 

[Dayman1225]

Ok i've basically typed up everything that is useful from that Interview.

https://pastebin.com/xKf2iaZ0

DC products are more than just CPUs. He never specifically mentioned more 14nm DC CPUs in 2019.

 

[wahdangun]

https://twitter.com/TMFChipFool/status/996478399794368514

Oh snap, Intel admitted that next year's server product (not Cascade Lake, the one after it) is on 14 nm.

So no Icelake Server.

Hahaha is this a joke ? 7nm vs 4nm++ ?

AMD need to go to TSMC and screw the WSA, just use it for medium low gpu like Polaris successor.

And since mining still strong they will sold out as many gpu as they want.

 

[IntelUser2000]

About the Cannonlake-Y, m3 8114Y,

the high base clocks are an indication that power management has improved noticeably. Which is due to 10nm. While + variants of 14nm may have performance benefits at the top end, no 14nm generation is going to be able to do what 10nm does in terms of power. Eventually they need to move to 10nm.

For the Cannonlake-U, 8121U, the all-core Turbo is at 2.7GHz, which is 100MHz lower than 8130U.

 

[jpiniero]

FCT on the 8130U is 3.4 actually.

 

[IntelUser2000]

Do you have official Intel documents?

Because 3 sources say 3 different things. 3.4GHz, 3.2GHz, 2.8GHz.

 

[jpiniero]

Do you have official Intel documents?

Because 3 sources say 3 different things. 3.4GHz, 3.2GHz, 2.8GHz.

No, just AT's article on the topic. Based upon the one review at NBC (3.1-3.2 sustained, CB15 score of 338) it might only allow 3.4 if the laptop was configured to be cTDP up or something.

 

[IntelUser2000]

In that case, it might be same with 8121U. The same source that said 8130U is 2.8GHz says 8121U is 2.7GHz. They are generally very accurate and get the information very early.

 

[LTC8K6]

Do you have official Intel documents?

Because 3 sources say 3 different things. 3.4GHz, 3.2GHz, 2.8GHz.

I read somewhere that Intel doesn't publish the specific turbo frequencies any more, just the single core turbo.

The NBC test was with single channel ram, if that affects the CB scores.

 

[LTC8K6]

In that case, it might be same with 8121U. The same source that said 8130U is 2.8GHz says 8121U is 2.7GHz. They are generally very accurate and get the information very early.

If we go with the same ratio as the 8130u, 2-300mhz below single core, it would be 2.9-3.0 for the 8121u.
If we grant the 10nm process a bit of improvement, maybe 3.0-3.1?

 

[jpiniero]

In that case, it might be same with 8121U. The same source that said 8130U is 2.8GHz says 8121U is 2.7GHz. They are generally very accurate and get the information very early.

By 3.1-3.2 I mean both cores loaded. So that's not really that accurate.

The FCT of the 8121U probably depends more on how the yields look and marketing than anything else.

 

[eek2121]

Yeah, I wouldn't be holding my breath for Zen 2 in 1H19, let alone 1Q19.

The data-points we have right now are, that 7nm vega is in the labs (wildly rumored to be the GF 7nm pipe-cleaner) and that AMD promises to sample 7nm CPUs in the very end of 2018.

So naturally we should compare that info to zen 1, which started sampling in 2Q16, and was expected to be released in very late 4Q16 and was subsequently delayed a few months.

Based on that, it seems like very-late Q2 at the earliest, with the usual quarterly delay quite possible.

That's just speculation of course. I would very much like to see Zen 2 earlier, just not getting my hopes up yet, without any further info. After all GloFo might screw up it's 7nm as well, and the architecture might need respins (which adds a couple months of delays). I mean we don't even have Engineering Samples yet ...

No, we will see Zen 3xxx series in April of 2019, just like we saw Zen 2xxx in April 2018, and Zen 1xxx in April of 2017. There will be no delays. AMD is playing rather aggressively at this point, and will continue to do so for the forseeable future. If Glofo can't keep up process wise, that's on them, and AMD will simply use TSMC. People were saying the same things about the 12nm release when it came out...on time and as expected.

Edit: And what about the contractual commitments with GloFo? That's where Vega 12 fits in. ANOTHER rumor that popped up recently. AMD has more than a few 'filler' products to keep GloFo happy. This includes 12nm Threadripper, the Rumored Vega 12 refresh, and a spindown of Ryzen 2xxx. It can then easily Ramp up Ryzen 3xxx on TSMC and move it over to GloFo and use both as suppliers as needed.

EDIT #2: Things are also rather different this time around, I nearly crapped my pants when I overheard a local best buy rep recommend a Ryzen system over an Intel one, and AMD advertisements were playing on all the monitors in the monitors section. AMD is executing on all fronts. If they can get best buy to sell a clueless dude a Ryzen system over an Intel one, well, I can't really express those feelings in words here without violating forum rules....but let's just say a certain hot place suddenly became a very cold one.

As far as Cannon Lake? Intel should be on track around Q2 of next year, but I expect clock speeds still won't be up to par with current flagship parts. So, expect mobile parts for a while, with a focus on power over frequency. This may help Intel in the long run more than it hurts them, as they are in dire need of a competitive mobile part. AMD has quad core parts that are 15 watts or less and have an integrated GPU. This is an area with Intel could use some improvement. Once Intel is able to ramp up clock speeds/IPC, they will roll out the desktop parts.

 

[LTC8K6]

In that case, it might be same with 8121U. The same source that said 8130U is 2.8GHz says 8121U is 2.7GHz. They are generally very accurate and get the information very early.

It apparently dropped to 2.7 in the Corona benchmark.
https://pic3.zhimg.com/80/v2-21d97466385347f1a4e297e55e101463_hd.jpg

 

[Eug]

Things are also rather different this time around, I nearly crapped my pants when I overheard a local best buy rep recommend a Ryzen system over an Intel one, and AMD advertisements were playing on all the monitors in the monitors section. AMD is executing on all fronts. If they can get best buy to sell a clueless dude a Ryzen system over an Intel one, well, I can't really express those feelings in words here without violating forum rules....but let's just say a certain hot place suddenly became a very cold one.

Best Buy sales agents will sell whatever is appropriate for the customer, in terms of usage and budget. Sometimes that's Intel, and sometimes that's AMD. That was true two years ago too. I don't understand why you'd think this is new.

Maybe at the high end Intel was often preferred, but that's because Intel was often simply better at the high end.

 

[mikk]

About the Cannonlake-Y, m3 8114Y,

the high base clocks are an indication that power management has improved noticeably. Which is due to 10nm. While + variants of 14nm may have performance benefits at the top end, no 14nm generation is going to be able to do what 10nm does in terms of power. Eventually they need to move to 10nm.

It means m3-8114Y with 1.5 Ghz base is a real improvement in the 5W range over the 14nm generation, m3-7Y32 only clocks with 1.1 Ghz. I wonder if there will be faster i5 and i7 models with an even higher base clock.

 

[Gideon]

No, we will see Zen 3xxx series in April of 2019, just like we saw Zen 2xxx in April 2018, and Zen 1xxx in April of 2017. There will be no delays. AMD is playing rather aggressively at this point, and will continue to do so for the forseeable future. If Glofo can't keep up process wise, that's on them, and AMD will simply use TSMC. People were saying the same things about the 12nm release when it came out...on time and as expected.

Edit: And what about the contractual commitments with GloFo? That's where Vega 12 fits in. ANOTHER rumor that popped up recently. AMD has more than a few 'filler' products to keep GloFo happy. This includes 12nm Threadripper, the Rumored Vega 12 refresh, and a spindown of Ryzen 2xxx. It can then easily Ramp up Ryzen 3xxx on TSMC and move it over to GloFo and use both as suppliers as needed.

EDIT: Sorry for the OT post, nothing about Intel here

I wouldn't mind at all to see it released sooner rather than later. Just a couple of points:

1. AMD released Zen 1 in February 2017 not April. And it promised it initially for the very end of Q4 2016.
2. 12nm Vega is almost certainly cancelled. There are plenty of mentions of 7nm, including that AMD has working silicon in the labs and that it's manufactured @ TSMC. AMD just reiterated Vega 20 and Na'vi in a new roadmap, saying nothing about 12nm Vega
3. AMD can't just switch Foundries at a whim, If they can produce it @ TSMC as well, they must have already spent significant engineering effort on that by now (both years of development time and $$$). It's possible that they did it already as would probably be a good move to hedge the bets, and very much needed if EPYC 3xxx indeed kicks off big time, but we haven't heard anything about the possibility yet.

Right now it seems that Zen 2 has taped out and the ball is in GlobalFoundries' court (AMD has yet to receive initial silicon). GF Screwed up their 20nm and 14nm processes, had big time issues with 32nm and 28 nm in the beginning. AMD has also postponed quite a few products a quarter or two from initial promises (Steamroller and Kaveri come to mind). 7nm looks to be doing better (particuarily because of IBMs engineers) but I wouldn't hold my breath until we hear of initial silicon, or better yet, Engineering Sample leaks.

Don't get me wrong, I would love AMD to deliver it sooner rather than later (as I will be upgrading to it skipping Zen+). After all considering how small the window between the sampling and release of Threadripper 2 is, they might even deliver in Q1. I'd just be cautiously pessimistic for now rather than be disappointed later. e.g. finding out that I have to wait 9 more months after waiting 6 months already. This has happend countless and countless times with AMD, the most recent one being VEGA.

 

[lefty2]

As far as Cannon Lake? Intel should be on track around Q2 of next year, but I expect clock speeds still won't be up to par with current flagship parts. So, expect mobile parts for a while, with a focus on power over frequency.

So, we will see more Cannon Lake SKUs next year, but not Ice Lake? I believe that Intel said Cannon Lake is 10nm and Ice Lake 10nm+, but from the earnings call it did not sound like 10nm+ would have any chance of making it in 2019.