Discussion Intel current and future Lakes & Rapids thread

[adroc_thurston]

5 bucks it's because Intel tried using HD cells with TSMC 3nm and couldn't get clocks up lol.

That's not the interesting part.
PPW uplift nonexistent with two shrinks and two core updates, one of which is a tock.
itsover.png

 

[H433x0n]

Intel’s internal performance projection for Raptor Lake S Refresh and Arrow Lake S - How fast the CPU and iGP are expected to be | Exclusive | igor´sLAB

The Raptor Lake refresh is expected to arrive this year as the 14th CPU generation for the desktop. But the real successor, Arrow Lake-S for desktop, is also increasingly materializing along with the…

www.igorslab.de

Apparently clocks are shit. 5 bucks it's because Intel tried using HD cells with TSMC 3nm and couldn't get clocks up lol.
IPC uplift actually seems to be decent, bcuz apparently ARL-S isn't going to clock very well... around MTL levels.
Cache changes are supposed to bring greater than expect gaming perf uplifts though ¯\_(ツ)_/¯

ARL looks to be in a ... unique? situation lol

That.. would be dreadful if they can’t get TSMC N3 to achieve similar clocks to at least Alder Lake. I’d like to think that would be fixed before release by trying to working with TSMC to tweak the process.

Honestly, worst case just eat the costs that come from less than ideal defect density and run 20A or do a sprint and port to Intel 3.

 

[H433x0n]

That's not the interesting part.
PPW uplift nonexistent with two shrinks and two core updates, one of which is a tock.
itsover.png

… Granite Rapids is a thing and will increase PPW by an appreciable amount.

We’re also still 5 quarters out from release with ARL.

 

[adroc_thurston]

… Granite Rapids is a thing and will increase PPW by an appreciable amount.

It's a toy versus Turin and every vendor in the world knows it.
Please move your hopium towards CWF, maybe it's half-decent!

We’re also still 5 quarters out from release with ARL.

yea sorry I'm also ODing on copium due to how shit LNC is.
grim prospects for DMR all-around, and was the only hope now that GNR is barely a footnote.


No profanity in the tech forums.


esquared
Anandtech Forum Director

 

[Geddagod]

PPW uplift nonexistent with two shrinks and two core updates, one of which is a tock.
itsover.png

Mobile is apparently much better for some reason

That.. would be dreadful if they can’t get TSMC N3 to achieve similar clocks to at least Alder Lake.

I think it's ST freq is the worst part, MT score might be low because of SMT removal tbh.

I’d like to think that would be fixed before release by trying to working with TSMC to tweak the process.

Taped out. Not much they can do at this point IMO

 

[adroc_thurston]

Mobile is apparently much better for some reason

Not really, no.
Lunar is a lot better for core-unrelated reasons and might genuinely be a really decent/good product if stars align.

MT score might be low because of SMT removal tbh.

That means the new Atoms are garbage, brother!
Two core updates, two shrinks, 15% ppw bump ahoy.
This is what AMD did with one shrink, iso-CC.
Even the flaming wreck that is RDOA3 benches like 25% ppw bump on N5 going off N6, and the IP itself is a mess.

Taped out. Not much they can do at this point IMO

Yeap.

 

[H433x0n]

Mobile is apparently much better for some reason

I think it's ST freq is the worst part, MT score might be low because of SMT removal tbh.

Taped out. Not much they can do at this point IMO

Big oof. Guess we’ll have to wait and see. Perhaps we took forgranted the switch to TSMC would be a seamless transition.

Intel’s fabs for all of their flaws were practically tailor made to work with Intel design teams and their core arch.

Time will tell how it works out. I’ll be extra interested to see the performance differences between the 20A and N3 SKUs now.

 

[msj10]

yikes these ST performance numbers are horrible, even Raptor Lake with the same node and same architecture managed more than that.

 

[uzzi38]

@Geddagod Mobile ARL perf targets definitely don't tell a different story.

 

[tamz_msc]

I've always thought that Intel's P-core is a failure considering how (not) economical it is with its area, resources and power consumption. Looks like it is the same story with Arrow Lake.

 

[adroc_thurston]

Looks like it is the same story with Arrow Lake.

Atoms look like an even bigger failure given two shrinks, two core updates and the resulting 15% ppw in like the most generous to them workload that is specint mt.

 

[Exist50]

That means the new Atoms are garbage, brother!
Two core updates, two shrinks, 15% ppw bump ahoy.

The removal of SMT is clearly making a huge impact. That's like 30% throughput on all 8 big cores. A big gap that needs to be closed purely by IPC and iso-power clocks. Certainly doesn't say anything negative about Skymont. Not sure how you reached that conclusion.

 

[Geddagod]

I've always thought that Intel's P-core is a failure considering how (not) economical it is with its area, resources and power consumption. Looks like it is the same story with Arrow Lake.

area isnt the problem with LNC apparently.
Edit: you could kinda tell that with the LNL thermal die shot

 

[tamz_msc]

What is the P-core called in Arrow Lake? LNC stands for what exactly? Sorry if this has been asked before.

The removal of SMT is clearly making a huge impact. That's like 30% throughput on all 8 big cores.

I always wondered if there exists any recent benchmark(s) with a single core comparing SMT on vs off and whether in such cases the uplift is indeed the usual 25-30% you see with typical multithreaded workloads.

 

[H433x0n]

If this info is accurate, it'd make me believe 1 of 2 things is likely to happen:

1. They're searching for an alternative route to achieve Alder Lake level clock speeds.
2. We're getting a Meteor Lake Refresh next year (but for desktop). I'm fairly certain that RWC+ at 5.4-5.5ghz would produce better results than these numbers.

I don't see them knowingly performing a repeat of Pentium 4. This would be a much smaller generational jump than what Alder Lake produced.

I've always thought that Intel's P-core is a failure considering how (not) economical it is with its area, resources and power consumption. Looks like it is the same story with Arrow Lake.

If these numbers end up being what releases in the final product, Raptor Cove is a significantly better Arch than Lion Cove. That or TSMC N3 can’t achieve >5.0-5.1ghz.

 

[Exist50]

What is the P-core called in Arrow Lake? LNC stands for what exactly?

Lion Cove.

I always wondered if there exists any recent benchmark(s) with a single core comparing SMT on vs off and whether in such cases the uplift is indeed the usual 25-30% you see with typical multithreaded workloads.

Anandtech has some good data for both Intel and AMD here:

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

More recently they tested Zen 3:

Investigating Performance of Multi-Threading on Zen 3 and AMD Ryzen 5000

www.anandtech.com

Obviously, it varies a ton, but SPEC nT means running n independent instances, so it should be pretty ideal.

 

[tamz_msc]

More recently they tested Zen 3:

Thanks for the link.

So it is as I expected. ST performance is not affected by SMT at all. Assuming similar clocks, using the composite benchmarks like GB and SPEC, PPC is improved by only 7-10 % in Lion Cove compared to Raptor Cove.

That is extremely underwhelming.

 

[adroc_thurston]

chef's kiss

 

[tamz_msc]

30-40% higher ST than Raptor Lake. High Confidence.

Lulz.

 

[H433x0n]

Thanks for the link.

So it is as I expected. ST performance is not affected by SMT at all. Assuming similar clocks, using the composite benchmarks like GB and SPEC, PPC is improved by only 7-10 % in Lion Cove compared to Raptor Cove.

That is extremely underwhelming.

It’s not worthy of even being released. I don’t see a 4-6% ST improvement worthy of justifying the money spent on the packaging and marketing material.

A hypothetical 14900KS would be within spitting distance of the top ARL SKU at that point.

They’ll have to pivot and do something else.

 

[adroc_thurston]

30-40% higher ST than Raptor Lake. High Confidence.

40% MT at the same power iso CC lmaooooooooooo

It’s not worthy of even being released. I don’t see a 4-6% ST improvement worthy of justifying the money spent on the packaging and marketing material.

let's call it "a monument to madness".

 

[H433x0n]

let's call it "a monument to madness".

It’s a publicly traded company. They’re going to pivot to something else.

MTL for desktop with RWC+ could outperform what’s shown here. Hopefully some of the validation will have already been done on previous SKUs.

 

[adroc_thurston]

It’s a publicly traded company

Doesn't stop Pat from burning obscene amounts of cash on fabs!

They’re going to pivot to something else.

like what?
AI is no gusto since they've publically orphaned Habana roadmap in favour of mysterymeat GPU part that is Falcon Shores.
crawling to AMD's HQ and begging to tape their client/server stuff on i3/18A to get at least some incremental revenue while losing mss?

 

[H433x0n]

Doesn't stop Pat from burning obscene amounts of cash on fabs!

like what?
AI is no gusto since they've publically orphaned Habana roadmap in favour of mysterymeat GPU part that is Falcon Shores.
crawling to AMD's HQ and begging to tape their client/server stuff on i3/18A to get at least some incremental revenue while losing mss?

You’re extremely difficult to have any conversation with in good faith.

 

[adroc_thurston]

You’re extremely difficult to have any conversation with in good faith.

There is no conversation, Intel's best core in years is completely mid in terms of PPA, both in mobile and desktop.
All while Pat is burning obscene amounts of cash on fabs.