Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

If a particular application is going to scale well from 8 to 10 cores wouldn't it most likely also scale well from 8 to 16 cores, and then in that case might not 8+8 be more beneficial than 10 P's?

You would think so, but that's not always the case. There are, for example, some software encoders that won't scale up to 16 cores. They usually crap out at around 8-10 cores.

And then there are games, and those often don't even scale past 8 cores. Unless you're streaming . . . no idea how well adding Gracemont cores will fit that workload.

It's really down to which applications you intend to run, and how many applications you intend to run simultaneously.

 

[TESKATLIPOKA]

In the end, it's really down to which applications you intend to run.

Precisely. We can already see that some users would like to see only GC cores, others just 1-2GC with many GRMT cores. It's all about usage scenarios.

 

[Zucker2k]

Let's just call it the 5950x effect and move on. 16 cores is 16 cores, and if ADL-S ends up close or trumping the not-so-long ago HEDT chip that battled and sometimes bested Intel's 18-Core monster aka 10980XE, then Intel has really pulled out a fat rabbit out of a hat many naysayers here thought Intel never had. We wanted "moar cores" and now we have it. We should be happy. With these hybrid chips, Intel's shown it's ready to dish moar cores IF need be. In this regard, ADL is brilliant.

 

[DrMrLordX]

trumping the not-so-long ago HEDT chip

The 5950X isn't an HEDT CPU.

We wanted "moar cores" and now we have it. We should be happy. With these hybrid chips, Intel's shown it's ready to dish moar cores IF need be. In this regard, ADL is brilliant.

Don't fall for the hype. If Intel really gave us "moar coars" then they would serve up 16c Golden Cove.

 

[TESKATLIPOKA]

Don't fall for the hype. If Intel really gave us "moar coars" then they would serve up 16c Golden Cove.

The question is how much more performance would we gain in MT by using 16 GC cores.
Clocks would have to go down compared to 8+8.

 

[Abwx]

Let's just call it the 5950x effect and move on. 16 cores is 16 cores, and if ADL-S ends up close or trumping the not-so-long ago HEDT chip that battled and sometimes bested Intel's 18-Core monster aka 10980XE,

Sometime besting..?.

Actually 5950X is roughly 20% faster than the 10980XE, so that s not exactly battling, that fit rather the 5900X since it is on par with the 18C Intel.

CPU-Benchmark: Prozessor-Vergleich 2025

Welche CPU für Games und Apps kaufen? ComputerBase leistet Kaufberatung mit Intel Core Ultra, AMD Ryzen 9000 und 7000 auf Basis von Tests.

www.computerbase.de

 

[JasonLD]

Don't fall for the hype. If Intel really gave us "moar coars" then they would serve up 16c Golden Cove.

Though assuming there is no issues with Windows scheduler, I would say Intel's approach is better for the long run.

 

[Zucker2k]

Sometime besting..?.

Actually 5950X is roughly 20% faster than the 10980XE, so that s not exactly battling, that fit rather the 5900X since it is on par with the 18C Intel.

CPU-Benchmark: Prozessor-Vergleich 2025

Welche CPU für Games und Apps kaufen? ComputerBase leistet Kaufberatung mit Intel Core Ultra, AMD Ryzen 9000 und 7000 auf Basis von Tests.

www.computerbase.de

Errmmm, are you confusing the 10980XE with the 10900k? Because it's not in the link you posted.

 

[Abwx]

Errmmm, are you confusing the 10980XE with the 10900k? Because it's not in the link you posted.

It is, click on the right side of the graph on " + 64 eintrage" button, that s 64 more CPUs displayed (in MT and 67 in ST), and quite practical for comparisons since you can lock a CPU at 100% by clicking on its bar.
FI the 10980XE has 3-4% better ST IPC than usual SKLs, surely due to a huge cache.

 

[Zucker2k]

In most reviews, we saw the two chips trading blows. Unless you want to say the 10980XE doesn't have it's strengths in certain workloads where it can leverage superior mem bandwidth and simd units.

 

[IntelUser2000]

Hmm, I thought Tremont would have been ready by 2018 or 2019?

Goldmont Plus came out late 2017. And I can tell by it's great perf/watt improvement that it was designed for 14nm from the start, unlike Rocket Lake which was meant for 10nm but simply backported.

Note that Xeon Phi that came out in middle of 2016 had Silvermont cores, which came out in 2013. Part of the reason is because they had to modify quite a bit to fit HPC workloads, and not just adding AVX-512.

There's no way they'd have got Tremont in there in 2019. And I know that 10nm Phi was Goldmont-based.

Phi and OmniPath is an unfortunate victim of 10nm. The Silvermont-based Phi had problems reaching high results in the Linpack benchmark, likely because it's only 2-issue. When typical Xeon processors got 80-90% of it's theoretical peak, Phi got 60-70%. Goldmont is 3-issue which would have upped the throughput plus it would have performed 30% faster per clock.

 

[Abwx]

In most reviews, we saw the two chips trading blows. Unless you want to say the 10980XE doesn't have it's strengths in certain workloads where it can leverage superior mem bandwidth and simd units.

Dunno who are thoses reviews coming from, you can compare the scores in the softwares they use in their RKL review, including the 10980XE here :

Intel Core i9-11900K & i5-11600K „Rocket Lake-S“ im Test: Benchmarks in Anwendungen

Core i9-11900K und i5-11600K im Test: Benchmarks in Anwendungen / Multi-Core-Szenarien / Single-Core-Szenarien

www.computerbase.de

At some point 22% better IPC and perhaps also frequency (4050MHz in MT for a stock 5950X) make a big difference.

 

[IntelUser2000]

Beside a GC core clocked 40% lower than a GRMT would provide same ST perf and better throughput with SMT while consuming comparable power.

Only advantage is marketing since they can advertise a big core amount at lowish die area cost.

LOL. Try 25% lower clocks.

And in the ultra mobile space Gracemont is going to stretch it's lead.

Look at the U chips. They look great in single run Cinebench, but they throttle after that. The N-series chips based on mont cores don't throttle at all.

The Cove cores didn't throttle bad in the Skylake days, but they had to double the cores to compete with AMD and it was too much on 14nm.

You know if they could have put 8 Skylake cores in 15W they would have done so, and that's what Alderlake offers plus Golden Cove.

The question is how the schedulers work in reality. If it works decent then it's going to be a boon for mobile.

 

[Hulk]

Intel says at the same power consumption Gracemont is 40% faster than Skylake.
Assuming that "faster" means additional throughput, at what clockspeeds do you think this applies?

3GHz for Gracemont, 2GHz for Skylake?

 

[eek2121]

Intel to launch 12th Gen Core Alder Lake-S and Z690 motherboards on November 19th - VideoCardz.com

Intel Alder Lake has a release date According to Wccftech, who are quoting their own sources, Intel is expected to launch its Alder Lake-S CPUs on November 19th. Intel Alder Lake Engineering Sample, Source: HKEPC The site claims that the product launch will not take place at the same time as the...

videocardz.com

Looks like the actual release date of Alder Lake is possibly November 19th.

Yes, with an announcement a couple weeks prior.

And a pre-launch announcement in late October with performance numbers, specifications and prices. This is like a paper launch.

You don't claim AMD is doing a paper launch when they do their announcements, which ALWAYS have a few weeks delay between announcement and retail availability. Why would you do this for Intel...who is doing the exact same thing?

EDIT: Intel retail availability will probably be better than typical AMD launches as well...unless everyone buys all the chips. "Intel 7" isn't using EUV...

 

[DrMrLordX]

The question is how much more performance would we gain in MT by using 16 GC cores.
Clocks would have to go down compared to 8+8.

Given the massive size/power draw of Golden Cove, yes. Stop to consider for a moment why Intel has those problems.

Though assuming there is no issues with Windows scheduler, I would say Intel's approach is better for the long run.

Only because they must. If Intel could fit 16c Golden Cove performance into one package in 2021 without exceeding ~125-150W, then they wouldn't be in the state they are in today.

Goldmont Plus came out late 2017.

Guess what else came out in late 2017? Cannonlake. Goldmont Plus was stopgap solution for a failing 10nm process. Cannonlake was supposed to be out instead of Kaby etc. Tremont was heavily delayed by process.

 

[JasonLD]

Only because they must. If Intel could fit 16c Golden Cove performance into one package in 2021 without exceeding ~125-150W, then they wouldn't be in the state they are in today.

Then they would be able to do 8+8 on 90W package then. I still think focusing ST performance on 8 Big Cores (May increase to 10 if needed) and extending MT performance on small cores is better long term solution for none-HEDT desktops.

 

[DrMrLordX]

Then they would be able to do 8+8 on 90W package then.

For mobile, that might be a good idea. For desktop/workstation CPUs? Nah. You really want a homogeneous core arrangement where possible, especially if you are talking about your top-performing cores. Intel is going with heterogeneous core arrangements because they must.

 

[JasonLD]

For mobile, that might be a good idea. For desktop/workstation CPUs? Nah. You really want a homogeneous core arrangement where possible, especially if you are talking about your top-performing cores. Intel is going with heterogeneous core arrangements because they must.

If OS scheduler works properly, I don't see the downside of using many small cores for MT tasks/background task while maintaining ST performance on Big cores for most tasks. Even for desktop/workstation, I would think 8+32 will perform better than something like 24 big cores and the chip is going to be smaller and use less power.

 

[itsmydamnation]

If OS scheduler works properly, I don't see the downside of using many small cores for MT tasks/background task while maintaining ST performance on Big cores for most tasks. Even for desktop/workstation, I would think 8+32 will perform better than something like 24 big cores and the chip is going to be smaller and use less power.

then you need to look at AMD's patients, Their Big little if the pull it off will be way better for power, then MT performance is just a question of how much die area they want to throw at it


edit: linky

https://www.freepatentsonline.com/10698472.pdf

 

[IntelUser2000]

Guess what else came out in late 2017? Cannonlake. Goldmont Plus was stopgap solution for a failing 10nm process. Cannonlake was supposed to be out instead of Kaby etc. Tremont was heavily delayed by process.

You really believe that? The original Goldmont, while it was 30-35% faster per clock it increased power use somewhat.

Goldmont Plus, despite being on the same 14nm process(not even a variant like 14nm+) was an additional 30% faster with zero power increase. If you get a Goldmont Plus laptop the performance would be rock stable, and that was at 6W. What kind of a "stopgap" solution does that may I ask?

That's because it was planned to be on 14nm all along. Why? Due to artificial segmentation. In their original plans, by the time Goldmont and the Plus came out they'd have been far along with Cove cores. By keeping the mont cores on 14nm it saves production costs for dirt cheap laptops and tablets.

Atom was "saved" by that artificial segmentation. Had Goldmont Plus been on 10nm then it would have been a delayed part that was slower. Tremont on the original roadmaps were 2 years after Goldmont Plus. So 6-12 months delay for Tremont. I wouldn't call that heavily delayed.

Whatever delay Tremont had due to 10nm is completely made up with Gracemont, since it's another big jump just a year later.

I bet you if they had a proper 14nm successor to Skylake it would have done quite a bit better. The issue is that a proper design takes years. A backport is simple, but there's such a thing as a design made for a certain process.

 

[DrMrLordX]

You really believe that?

Yes. If I'm wrong I'm wrong, and I had thought Goldmont Plus was just a clockspeed bump over Goldmont. That's a significant IPC jump for something with the "Plus" monicker. If I recall correctly, the oldest roadmaps featuring Tremont didn't have Goldmont Plus on them at all.
edit: I might be mistaken, and it might be Goldmont Plus Refresh that was the stopgap, though it wasn't announced until 2019.

If OS scheduler works properly, I don't see the downside of using many small cores for MT tasks/background task while maintaining ST performance on Big cores for most tasks. Even for desktop/workstation, I would think 8+32 will perform better than something like 24 big cores and the chip is going to be smaller and use less power.

Highly doubtful. Intel's competition can sell 16c with an effective cap on power of 142W without making any compromises on SIMD extensions. Their server CPUs can scale up to 64c and stay under 300W. The only reason why Intel is even thinking about heterogeneous core situations is that they can't fit 16c Golden Cove in one consumer package without making huge clockspeed sacrifices in the process. We're seeing an inversion where Intel is throwing in the towel on Core and embracing Atom out of necessity. The transition will be awkward, but eventually, you may just see Atom.

 

[Asterox]

Hm, Alder Lake haiku details mix.

https://twitter.com/x/status/1433771027805270016

 

[RTX2080]

Hm, Alder Lake haiku details mix.

https://twitter.com/x/status/1433771027805270016

A Taiwanese MSI insider Toppc talked about ADL but due to NDA he decided to speak riddle lol

'5 become 4' ----------- DDR5 support Gear4?
'K's 100 no longer exclusive' ----------------- BCLK OC on non-K model?
'Head of contact be careful' ------------------ Looks like heatsink & bigger contact area on the lid needs to be taken care?

 

[JoeRambo]

A Taiwanese MSI insider Toppc talked about ADL but due to NDA he decided to speak riddle lol

'5 become 4' ----------- DDR5 support Gear4?
'K's 100 no longer exclusive' ----------------- FCLK OC on non-K model?
'Head of contact be careful' ------------------ Looks like heatsink & bigger contact area on the lid needs to be taken care?

"More R's to choose from" ------------- could reference the memory dividers 100 / 133, there might be more and they might actually play role in stability and performance?