Discussion Intel current and future Lakes & Rapids thread

[Zucker2k]

And at max clocks each core will be capable of consuming over 20W.

Define max clocks, please. At what clock does a WC core consume 20w?

 

[jpiniero]

Intel Sapphire Rapids for consumers:

Intel Sapphire Rapids HEDT appears in a roadmap with W790 chipset - VideoCardz.com

Intel Sapphire Rapids for consumers There has been a lot of talks recently on future HEDT processors from AMD and Intel. It seems that consumers waiting for a next-gen upgrade for 16+ core systems might need to wait a bit longer. We received a partial roadmap featuring Intel Sapphire Rapids...

videocardz.com

Slides look fake. Timing I think is optimistic given that SPR Server has pushed it's production start into Q1.

 

[uzzi38]

Define max clocks, please. At what clock does a WC core consume 20w?

Defining max clocks would be unfair given we're still in ES phases. There's still room for improvement from where they're currently at.

You'll see people talk about ADL clocks more once we get to the QS phase. We're all expecting higher clocks than now at least, because ES2 max clocks are lower much than TGL-H.

EDIT: Oh right, Igor's already talked about the clocks.

Here's the article: https://www.igorslab.de/en/exclusive-data-to-intel-alder-lake-s-hot-fight-with-amd/

 

[Dayman1225]

Slides look fake. Timing I think is optimistic given that SPR Server has pushed it's production start into Q1.

Reminder that left edge of bar only represents beginning of production, not release.

 

[tamz_msc]

Define max clocks, please. At what clock does a WC core consume 20w?

At 4.8 GHz, running single-threaded CPU-Z benchmark. I know because I have one.

 

[majord]

At 4.8 GHz, running single-threaded CPU-Z benchmark. I know because I have one.

What CPU?

 

[Zucker2k]

And at max clocks each core will be capable of consuming over 20W. The same as Tiger Lake does right now, mind you.

That's a really meaningless statement.

At 4.8 GHz, running single-threaded CPU-Z benchmark. I know because I have one.

Now, which Willow Cove based processor feeds each core over 20 watts at the same time??

 

[tamz_msc]

Now, which Willow Cove based processor feeds each core over 20 watts at the same time??

If you include H45-series processors then the i9-11980HK has a PL2 of 135 W in cTDP up mode, which amounts to ~16-17 W peak power consumption.

 

[tamz_msc]

What CPU?

It's in my sig. An i7-11370H.

 

[Zucker2k]

If you include H45-series processors then the i9-11980HK has a PL2 of 135 W in cTDP up mode, which amounts to ~16-17 W peak power consumption.

I appreciate your effort at providing these answers but I don't think you're even taken context into your equations. Since you readily provided the single core consumption of 20 watts, maybe you could simply give us a CPU-Z multithreaded bench power consumption number, too. Because the post I answered claims each core is capable of consuming over 20+ watts at max clocks. I'm merely asking in what practical scenario has anyone witnessed such an event? Care to share your ALL-CORE TURBO power numbers?

 

[tamz_msc]

I appreciate your effort at providing these answers but I don't think you're even taken context into your equations. Since you readily provided the single core consumption of 20 watts, maybe you could simply give us a CPU-Z multithreaded bench power consumption number, too. Because the post I answered claims each core is capable of consuming over 20+ watts at max clocks. I'm merely asking in what practical scenario has anyone witnessed such an event? Care to share your ALL-CORE TURBO power numbers?

The point is whether each core can draw 20 or more watts of power at max clocks, not whether all of them draw 20 W each at the same time.

 

[Zucker2k]

The point is whether each core can draw 20 or more watts of power at max clocks, not whether all of them draw 20 W each at the same time.

@uzzi38 What's the point of your statement then? @mikk was making an efficiency argument. The fact that a core is capable of consuming 20 watts means it's inefficient? By that logic, a core that's only capable of consuming 1 watt is the most efficient core, by default.

 

[uzzi38]

@uzzi38 What's the point of your statement then? @mikk was making an efficiency argument. The fact that a core is capable of consuming 20 watts means it's inefficient? By that logic, a core that's only capable of consuming 1 watt is the most efficient core, by default.

I didn't even reply to mikk, I replied to @eek2121 saying that being able to fit GLC cores in a 15W SKU is not an indication of how efficient they are as he was trying to point out, as evidenced by previous products.

 

[Zucker2k]

I didn't even reply to mikk, I replied to @eek2121 saying that being able to fit GLC cores in a 15W SKU is not an indication of how efficient they are as he was trying to point out, as evidenced by previous products.

I actually meant @eek2121 obviously. And why is it not indicative of efficiency? You're rather citing what a chip is capable of consuming, not what it actually consumes as an indicator of efficiency. In any case, the Willow Cove core actually scales well into the high clocks, as has continually been pointed out by @coercitiv.

 

[Asterox]

As someone on Twitter mentioned, Holy Lake is kind of a very nice name.

https://twitter.com/x/status/1410614835151196169

 

[Abwx]

I actually meant @eek2121 obviously. And why is it not indicative of efficiency? You're rather citing what a chip is capable of consuming, not what it actually consumes as an indicator of efficiency. In any case, the Willow Cove core actually scales well into the high clocks, as has continually been pointed out by @coercitiv.

Efficency is defined by the required power to get a given score in say Cinebench, so far TGL is not as efficent as the competition, either due to uarch, wich is doubtfull, or to the process, or to both, and probably that the process is the main culprit.

 

[Shivansps]

As someone on Twitter mentioned, Holy Lake is kind of a very nice name.

https://twitter.com/x/status/1410614835151196169

hahaha C2D miracle incoming.

 

[coercitiv]

New GB5 submission for SPR 20c/40t - score isn't relevant enough without proper clock information, but info on cache sizes is interesting:

Golden Cove: 2 MB L2$ per core
Sunny Cove: 1,25 MB L2$ per core

20C Sapphire Rapids: 75 MB L3$
40C Ice Lake: 60 MB L3$

PS: info via Andreas Schilling from Hardwareluxx on Twitter.

 

[lobz]

can we please have a 'request delete in case of user error' button?

 

[lobz]

New GB5 submission for SPR 20c/40t - score isn't relevant enough without proper clock information, but info on cache sizes is interesting:

Golden Cove: 2 MB L2$ per core
Sunny Cove: 1,25 MB L2$ per core

20C Sapphire Rapids: 75 MB L3$
40C Ice Lake: 60 MB L3$

PS: info via Andreas Schilling from Hardwareluxx on Twitter.

those are NOT going to be small chips

 

[lobz]

nevermind, I might be illiterate, as I so often click on 'reply' instead of 'edit'

 

[uzzi38]

New GB5 submission for SPR 20c/40t - score isn't relevant enough without proper clock information, but info on cache sizes is interesting:

Golden Cove: 2 MB L2$ per core
Sunny Cove: 1,25 MB L2$ per core

20C Sapphire Rapids: 75 MB L3$
40C Ice Lake: 60 MB L3$

PS: info via Andreas Schilling from Hardwareluxx on Twitter.

That L3 is combined between the two 40c dies. It should be 37.5MB L3 per die, with L3 per core at 1.875MB.

 

[lobz]

That L3 is combined between the two 40c dies. It should be 37.5MB L3 per die, with L3 per core at 1.875MB.

That L2 is still massive. I know it's in WC as well, but has that improved anything significantly enough compared to Sunny, that makes it worth the die space?

 

[semiman]

You don't know whether Intel plans a co-exist with MTL and RPT for a while like they did with CML and ICL. You don't have any timelines for 2023 and you don't factor in that the 1 year cadence isn't set in stone in every case, RKL-S and ADL-S is much less than 1 year. To summary, this very basic 2022 overview doesn't say anything about a specific MTL availability during 2023 or 2024. What can be said is that MTL is very likely a mobile first product. And for sure MTL can be a H2 2023 product, however some people like IntelUser2000 claiming Meteor Lake is a very late 2023, early 2024 product which is complete speculation at this point.

Based on a Mizuho Securities' analysis(ASML Stock Is In A Sweet Spot (NASDAQ:ASML) | Seeking Alpha ), Intel wasn't able to acquire enough EUVs to mass produce their own 7nm products.

Meanwhile, according to ASML(ASML to Ship 30 EUV Scanners in 2019: Faster EUV Tools Coming – WebSetNet), each 45k wafer per month capable fab(logic fab) requires 10~20 EUV scanners and intel needs 200kwpm to supply chips globally(historically).

Even if EMIB and foveros help reducing wafer usage by cutting off IOs and peripherals, there are not enough scanners for Intel.

Funny thing is that TSMC's expected 3nm capacity in 2022 is 50kwpm. There aren't many good wafers out there so I'm pretty sure that intel 10ESF will be going together for a very long time.

 

[Gideon]

New GB5 submission for SPR 20c/40t - score isn't relevant enough without proper clock information, but info on cache sizes is interesting:

Golden Cove: 2 MB L2$ per core
Sunny Cove: 1,25 MB L2$ per core

20C Sapphire Rapids: 75 MB L3$
40C Ice Lake: 60 MB L3$

PS: info via Andreas Schilling from Hardwareluxx on Twitter.

If you believe the .gb4 output it was running at ~4.7Ghz average for the single core load. Now obviously the result is horribly broken so that doesn't tell us much other than it possibly clocks quite well.