Discussion Intel current and future Lakes & Rapids thread

[jpiniero]

x86 instruction set breakage would be a bad idea.

This is a world where Linux is the dominant server OS, and a large portion of software being used for servers is either Open Source or device neutral like Java. Much different than the days of Itanium. Now I don't think it would be that serious of a cut, the rumor was specifically MMX but they could easily cut 32-bit and older for instance and not get much blowback. For clients it'd be a problem for sure without good BC.

 

[IntelUser2000]

I think he was referring to the leaner and faster version of x86 rumor

I thought that had to do with this: https://www.anandtech.com/show/12068/intel-to-remove-bios-support-from-uefi-by-2020

Eliminating support for older components will simply things for all sectors of the tech industry.

 

[jpiniero]

I am wondering if Intel would be using EMIB to connect the PCH with Icelake. If the PCH does include Thunderbolt 3, that would easily overwhelm the current OPI setup. Maybe they could put the ISP on the PCH too, to save some real estate.

 

[DrMrLordX]

This is a world where Linux is the dominant server OS, and a large portion of software being used for servers is either Open Source or device neutral like Java. Much different than the days of Itanium. Now I don't think it would be that serious of a cut, the rumor was specifically MMX but they could easily cut 32-bit and older for instance and not get much blowback. For clients it'd be a problem for sure without good BC.

Breaking x86 compatibility was only ever a secondary concern with Itanium. The primary concern with Itanium were that it was so devilishly difficult to write good compilers for the design, and that the performance/watt was often shockingly bad.

 

[IntelUser2000]

I am wondering if Intel would be using EMIB to connect the PCH with Icelake. If the PCH does include Thunderbolt 3, that would easily overwhelm the current OPI setup. Maybe they could put the ISP on the PCH too, to save some real estate.

Thunderbolt 3 needs only 40Gb/s. That's nothing. If they double that with Thunderbolt 4, its still only 80Gb/s.

The OPI used to connect the PCH with the main chip is a lower bandwidth version of the one they use to connect GT3e GPUs with eDRAM.

The OPI that connects the eDRAM can handle 50GB/s bi-directional bandwidth. That's 10x TB3. No EMIB needed.

This is a world where Linux is the dominant server OS, and a large portion of software being used for servers is either Open Source or device neutral like Java. Much different than the days of Itanium.

Compatibility matters today as much as it did back then. There's no reason to overhaul the ecosystem without a substantial advantage.

 

[jpiniero]

Compatibility matters today as much as it did back then. There's no reason to overhaul the ecosystem without a substantial advantage.

I do think that if Intel wanted to move to a brand new instruction set; they could. But that would be a lot more work. I'm specifically talking about killing 32 bit and older and stuff like MMX.

 

[jpiniero]

Saw on the Twitters that Intel confirmed they are going to support AdaptiveSync, perhaps with Gen11/Icelake.

 

[IntelUser2000]

Saw on the Twitters that Intel confirmed they are going to support AdaptiveSync, perhaps with Gen11/Icelake.

Isn't that based on a reddit reply by Intel back in late August?

 

[Dayman1225]

Isn't that based on a reddit reply by Intel back in late August?

Someone asked the question again, this time they put up a video

https://twitter.com/intelgraphics/status/1063117118705565696?s=21

 

[Dayman1225]

Alleged 5/5 Lakefield @2GHz has appeared on 3DMark with a Physics score of 2888

Score Tweet

 

[mikk]

How good is this score, do we have a Gemini Lake comparison?

 

[DrMrLordX]

https://www.3dmark.com/fs/15405845

That's Fire Strike only, though. I don't see any Time Spy numbers for a J5005.

 

[IntelUser2000]

How good is this score, do we have a Gemini Lake comparison?

Why compare it to Gemini Lake? The Lakefield config is for better performance at lower power using Icelake and Tremont so it'll be very disappointing if it only ends up like a previous gen small core chip.

Ideally, I hope the configuration allows the best of both worlds.
-In single thread, the few Icelake cores can be cranked up to the point so it performs like 45W Icelake H in low threaded applications.
-Then potentially(maybe needing recompiling?), the many Tremont cores perform like 15W Icelake U chips in multi-threaded applications.
-All at a TDP that allows it to replace Core M chips.

2800-2900 score in Timespy Physics is comparable to 4 core 15W U chips like the 8250U, 8550U.

 

[Dayman1225]

2800-2900 score in Timespy Physics is comparable to 4 core 15W U chips like the 8250U, 8550U.

Not bad if this is truly at 2GHz

 

[mikk]

If the Comet Lake-S 10C rumour is true it should mean Icelake-S is cancelled after all or at least the original Icelake, I mean they could port Tigerlake-S features into Icelake with so much additional development time and still call it Icelake-S. After all it looks like we won't get 10nm desktop CPUs till H2 2020 or 2021. I wonder if Comet Lake is a Skylake CPU+GPU clone, there is little hope Comet Lake is a ported Icelake into the old 14nm process.

 

[jpiniero]

If the Comet Lake-S 10C rumour is true it should mean Icelake-S is cancelled after all or at least the original Icelake, I mean they could port Tigerlake-S features into Icelake with so much additional development time and still call it Icelake-S..

The new Icelake (U/Y) is rumored have Tigerlake's CPU changes.

Edit: I'll add that the 2020 desktop is most likely Tigerlake, even if it would be slower in current gen games compred to Coffee/Comet because of lower max clock speeds. The new consoles will be out by then and there should be enough going on to keep high core count models reasonably busy.

 

[IntelUser2000]

You guys know Comet Lake-U exists right?

That means no 10nm -U chips next year. No 10nm server, -S, or -U. Icelake might end up being like Cannonlake and being extremely small scale launch. This may become a history-making disaster for Intel.

 

[IntelUser2000]

Not bad if this is truly at 2GHz

Not great either. I don't think the Coffeelake mobile chips do any better in sustained load clocks. If they can make a 5.2W 4-core Icelake-Y chip, then Lakefield should offer other advantages or its toast.

 

[mikk]

You guys know Comet Lake-U exists right?

As GT3 SKU only according to older roadmaps, in this case it doesn't matter.

 

[IntelUser2000]

As GT3 SKU only according to older roadmaps, in this case it doesn't matter.

Lenovo leak confirms we'll see i7 9550U. It's still 14nm.

While news sites mistakenly says it'll be on 10nm, Lenovo says "8th Gen 9550U or 8565U."

This also answers the discrepancy of having a 14nm Gen 9 GT3e SKU competing against a 10nm Gen 11 GT2. The latter will just come a year later.

 

[mikk]

Lenovo leak confirms we'll see i7 9550U. It's still 14nm.

And do you think this is Comet Lake-U 4+2? Whiskey Lake just launched and now we may get another refresh.

 

[LTC8K6]

What is "Skylake Z" though?

 

[IntelUser2000]

And do you think this is Comet Lake-U 4+2? Whiskey Lake just launched and now we may get another refresh.

It may not be Comet Lake, but every leak like this reduces the chance of Icelake in 2019.

Be realistic and let's get the Intel fan cap off. The ball is in THEIR court to prove they can ship 10nm this year. We fell for marketing tricks that 10nm will be available "soon" for 3 years. The odds are stacked against them. Intel was so arrogant about 10nm being ready that they were completely, and absolutely caught off guard. See in the Amberlake thread how "Amberlake" is really Kabylake-Y rebranded and factory overclocked.

They advanced on the platform level with the CNL PCH and things like the 1W panels but on the CPU side? Nothing.

You also did not address the 14nm Gen 9 GT3e vs 10nm Gen 11 GT2 issue. GT3e is maybe 30-50% faster. Gen 11 GT2 can probably better without the eDRAM.

How do we as individuals benefit by emotionally supporting companies? Not at all.

 

[scannall]

It may not be Comet Lake, but every leak like this reduces the chance of Icelake in 2019.

Be realistic and let's get the Intel fan cap off. The ball is in THEIR court to prove they can ship 10nm this year. We fell for marketing tricks that 10nm will be available "soon" for 3 years. The odds are stacked against them. Intel was so arrogant about 10nm being ready that they were completely, and absolutely caught off guard. See in the Amberlake thread how "Amberlake" is really Kabylake-Y rebranded and factory overclocked.

They advanced on the platform level with the CNL PCH and things like the 1W panels but on the CPU side? Nothing.

You also did not address the 14nm Gen 9 GT3e vs 10nm Gen 11 GT2 issue. GT3e is maybe 30-50% faster. Gen 11 GT2 can probably better without the eDRAM.

How do we as individuals benefit by emotionally supporting companies? Not at all.

There is no reason whatsoever to believe their timelines at this point. It's all just marketing. "Deadlines" come and go. Then new deadlines come and go. But wait! Hold on, it's coming really soon! Pinky swear!

If you need an upgrade, just get the best bang for your buck, in the range of the budget you establish for your build, and that meets your goals. What you think is coming 'tomorrow' might be next quarter, or next year or never.

 

[IntelUser2000]

If you need an upgrade, just get the best bang for your buck, in the range of the budget you establish for your build, and that meets your goals. What you think is coming 'tomorrow' might be next quarter, or next year or never.

You know what, Amber Lake and Whiskey Lake are not so bad. Yes, the CPUs are at a standstill but the notebooks based on the new platforms have nice improvements.

Amberlake-Y based Spectre Folio 13 has 10% sustained performance gain in multi-thread and the 1W Panel is likely the reason for the great battery life improvement over the predecessors, and beating nearly every single competitor by a substantial amount(10-40%), and even the Macbook Air which is rare.

https://www.notebookcheck.net/HP-Sp...le-Review.365941.0.html#toc-energy-management

Whiskeylake is doing pretty well looking at the UX433F. It offers 10% sustained performance gain with significant lower PL1/PL2 values and battery life seems to be better compared to the previous gen.

https://www.notebookcheck.net/Asus-ZenBook-14-UX433F-i7-8565U-Laptop-Review.361811.0.html


They'd have been so much better if it had a 10nm new arch CPU in them.