Discussion Intel current and future Lakes & Rapids thread

[cortexa99]

Where are seeing 1.5x IGP perf increase? Are you referring to openCL from above link?

Sorry I make a terrible mistake, Xe seems do specially great in OpenCL, and I just directly translate the Xe OpenCL performance to graphic performance which is not correct.

 

[blckgrffn]

Even prices on the 10900k in the US are seeing a little downward pressure. The prevailing price is around $515, down from $540 where it had been for awhile.

$445 @ Best Buy today.

Intel Core i9-10900K 10th Generation 10-core 20-Thread 3.7 GHz (5.3 GHz Turbo) Socket LGA1200 Unlocked Desktop Processor BX8070110900K - Best Buy

 

[mikk]

Some very early geekbench leaks of Alder Lake. It seems that the top mobile version will have 6 big and 8 small cores:

Intel Alder Lake-P mobile CPU with 14 cores and 20 threads spotted - VideoCardz.com

Intel Alder Lake-P makes a debut with 14 cores A new Intel 12th Gen Core CPU has been spotted on Geekbench. Intel Alder Lake-P CPU with 14-core with 20 threads has made its first appearance on Geekbench. This unusual configuration can be traced back to Intel’s Hybrid Technology, which utilizes...

videocardz.com

Unfortunately no performance numbers yet.

Interesting to note this sample uses LPDDR4x with a fully enabled GT2 which implies it's a successor of TGL-U rather than CML-H/TGL-H. This is a big increase from 4 to 6+8 cores if this is the case. However they could sell it for both markets as they did with TGL-H35 which is a renamed TGL-UP3. The reported clock speed is also quite impressive for such an early sample.

 

[jpiniero]

which implies it's a successor of TGL-U rather than CML-H/TGL-H.

P is a successor to both really.

 

[LikeLinus]

Yes is some places, but with 10600KF there is no CPU cooler."Old R5 3600 is bit cheeper+CPU cooler in the box".This is Mindfactory/Germany vs Geizhals/Austria short comparison.

Sockel 1200

Hier finden Sie 50x Sockel 1200 diverser namhafter Hersteller in unserer Prozessoren (CPU)-Kategorie

www.mindfactory.de

Intel Core i5-10600KF, 6C/12T, 4.10-4.80GHz, boxed ohne Kühler ab € 192,69 (2022) | Preisvergleich Geizhals Österreich

✔ Preisvergleich für Intel Core i5-10600KF, 6C/12T, 4.10-4.80GHz, boxed ohne Kühler ✔ Bewertungen ✔ Produktinfo ⇒ Sockel: 1200 (LGA) • Codename: Comet Lake-S • Grafik: nein • TDP: 125W… ✔ Intel ✔ Testberichte ✔ Günstig kaufen

geizhals.at

AMD Ryzen 5 3600, 6C/12T, 3.60-4.20GHz, boxed ab € 236,67 (2022) | Preisvergleich Geizhals Österreich

✔ Preisvergleich für AMD Ryzen 5 3600, 6C/12T, 3.60-4.20GHz, boxed ✔ Bewertungen ✔ Produktinfo ⇒ Sockel: AM4 (PGA) • Codename: Matisse • Grafik: nein • TDP: 65W… ✔ AMD ✔ Testberichte ✔ Günstig kaufen

geizhals.at

If you look at sales numbers for various i5 10600 models, it looks absurdly vs R5 3600 small 114 000 CPU sold.

Suche nach ''Core i5-10600KF'': 5 Treffer | Mindfactory.de

Es wurden 5 Artikel gefunden

www.mindfactory.de

AMD Ryzen 5 3600 6x 3.60GHz So.AM4 BOX - Sockel AM4 | Mindfactory.de

AMD Desktop von AMD | AMD Ryzen 5 3600 6x 3.60GHz So.AM4 BOX :: Bestellt :: über 125.370 verkauft :: 24 Jahre Kompetenz | Hier bestellen

www.mindfactory.de

OK, I'm a bit confused. Why are peopled allowed to use random German retailers as some sort of actual metrics, lol, or a data point for real sales?

I've seen people condemn other sites based on their testing, but how is that any worse than using a single retailer as something important??

MindFactory is completely silly and doesn't actually represent a true overall sales point for any product. It's one country, one retailer, and nothing of importance. Seems great when you are trying to make a point, but the reality is that it's a drop in the bucket and doesn't mean anything.

This thread is about Intel, not AMD. I've seen post telling users to "stay on topic" in AMD threads, but I haven't seen anything in this thread or other Intel threads saying the same.

We don't care about "metrics" from some Germany company. They mean very little to the overall metrics.

 

[Hulk]

Alder Lake

I understand the primary reason for going with Big/Little cores is for power efficiency but I'm wondering if there is a possibility that the Windows scheduler could be optimized to assign threads to cores based on the workload of the core? If all of the cores have the same amount of compute but the threads don't it seems like such an approach could in effect "balance" thread performance as a whole?

I'm just wondering if there is any validity to this? It seems that a similar effect could be achieved by varying clocks of various cores but of course this could lead to wasted resources or power.

Also do we know anything at all architecturally about the Golden Cove core? I couldn't find anything except 10 to 20% IPC increase over Tiger Lake. Yes, yes, but how do we get there is the question, right? Addition of another decoder and execution port, better OoO scheduling/branch prediction, etc?

 

[Bouowmx]

OK, I'm a bit confused. Why are peopled allowed to use random German retailers as some sort of actual metrics, lol, or a data point for real sales?

Post redated

I understand the primary reason for going with Big/Little cores is for power efficiency but I'm wondering if there is a possibility that the Windows scheduler could be optimized to assign threads to cores based on the workload of the core? If all of the cores have the same amount of compute but the threads don't it seems like such an approach could in effect "balance" thread performance as a whole?

I don't fully understand your question, but see how Android/Linux does it:

Documentation/scheduler/sched-zone.txt - kernel/msm - Git at Google

init_load_pct




Trolling and inflammatory posts are not allowed.


esquared
Anandtech Forum Director

 

[lobz]

OK, I'm a bit confused. Why are peopled allowed to use random German retailers as some sort of actual metrics, lol, or a data point for real sales?

I've seen people condemn other sites based on their testing, but how is that any worse than using a single retailer as something important??

MindFactory is completely silly and doesn't actually represent a true overall sales point for any product. It's one country, one retailer, and nothing of importance. Seems great when you are trying to make a point, but the reality is that it's a drop in the bucket and doesn't mean anything.

This thread is about Intel, not AMD. I've seen post telling users to "stay on topic" in AMD threads, but I haven't seen anything in this thread or other Intel threads saying the same.

We don't care about "metrics" from some Germany company. They mean very little to the overall metrics.

This was an answer to another post that already compared Intel and AMD prices, so it couldn't be off-topic even if you really wanted it to be. Other than that, It's not the actual numbers that are important in the monthly mindfactory reports (which is one of the biggest e-tailers in the biggest economy of Europe), but the trend it shows over the past 5 years. Numbers can be sketchy, but trends are always something to look out for. So no need for you to cry out so loud this time.

 

[lobz]

... double

 

[Hulk]

I don't fully understand your question, but see how Android/Linux does it:
Documentation/scheduler/sched-zone.txt - kernel/msm - Git at Google init_load_pct

Yes. That is exactly what I was looking for. After reading this I'm still pondering a few questions.

If you have a high performance CPU, say a Sunny Cove core and it is working on a thread that is not have a lot of "parallelism" meaning instructions that can be executed simultaneously, in order to run the task quickly it would have to be run at a high clock even though many internal resources aren't being utilized. Yes, I realize this is why HT was implemented.

But to my specific question. Is it possible that a high efficiency core could execute this thread described above with the same IPC as the high performance core since much of the high performance core's resources aren't being utilized anyway? This would seem to me to be a big advantage of Alder Lake if this is the case. The HMP Scheduler of course has to be very smart but now we are moving into a realm of dynamically allocating resources based on the capability of various cores available to the scheduler. This is very interesting and exciting. I can't wait to see how Alder Lake performs on various workloads.

I'm sure Intel has run the simulations and has decided the path to the best performance and efficiency for a given number of transistors (die size) is to not only be able to vary frequency per core but also have the ability to move high/low workload threads to high performance or high efficiency cores.

 

[yuri69]

Intel surely posses the resources to make even the Windows scheduler work somewhat correctly with Alder Lake.

Due Intel $$$ it won't be the sad show as with AMD x86 stuff like - Bulldozer CMT scheduling, Ryzen CCX scheduling, Ryzen preferred core scheduling, Threadripper MCM-aware scheduling, Threadripper 128 thread issue, etc.

The *worst* case scenario for Alder Lake is to simply disable the "Atom Cores" in BIOS. So you end up with an uniform high-performance 8c processor.

 

[lobz]

Intel surely posses the resources to make even the Windows scheduler work somewhat correctly with Alder Lake.

Due Intel $$$ it won't be the sad show as with AMD x86 stuff like - Bulldozer CMT scheduling, Ryzen CCX scheduling, Ryzen preferred core scheduling, Threadripper MCM-aware scheduling, Threadripper 128 thread issue, etc.

The *worst* case scenario for Alder Lake is to simply disable the "Atom Cores" in BIOS. So you end up with an uniform high-performance 8c processor.

Regardless of any amount of money or even just making sense, you think FAR-FAR too highly of Microsoft. The conglomerate fails even at the simplest of tasks and makes too much effort to fix non-existent problems, making everything cringe. The only company out there that is even more tone deaf towards consumer demand is Apple, but that was always a very low bar to clear, so... I wouldn't assume anything to be 'fixed' or made ready just because it is suddenly in Intel's interest too. It's really not about that.

 

[dullard]

Regardless of any amount of money or even just making sense, you think FAR-FAR too highly of Microsoft. The conglomerate fails even at the simplest of tasks and makes too much effort to fix non-existent problems, making everything cringe. The only company out there that is even more tone deaf towards consumer demand is Apple, but that was always a very low bar to clear, so... I wouldn't assume anything to be 'fixed' or made ready just because it is suddenly in Intel's interest too. It's really not about that.

You are correct that Microsoft is slow and tone-deaf to many customer's needs.

However, Microsoft has a vested interest in keeping x86 alive. An ARM/Android dominated world could be the final straw for Microsoft. Having a good scheduler is one way for Microsoft to fight back against ARM/Android.

Plus, if I were Microsoft, I would drool over the possibility of dedicated cores to keep the operating system running while the big cores run the software. Think of all the ways they can run those small cores at 100% usage all the time without ever impacting the user!

 

[DrMrLordX]

OK, I'm a bit confused. Why are peopled allowed to use random German retailers as some sort of actual metrics, lol, or a data point for real sales?

mindfactory is one of the few retailers that will tell you what they've sold in exacting detail. Amazon has a "best sellers" list but provides no context. mindfactory is a microcosm of the DiY market. It's a case of "some data is better than no data".

 

[dullard]

mindfactory is one of the few retailers that will tell you what they've sold in exacting detail. Amazon has a "best sellers" list but provides no context. mindfactory is a microcosm of the DiY market. It's a case of "some data is better than no data".

It is also the equivalent of looking at Pizza Hut's sales numbers in exacting detail and concluding that hamburgers don't sell well in America. Non-scientific data (such as self-selected DIY buyers from one niche company) is a bad way of understanding the general CPU market. It might help explain the DIY market, but even then it isn't that great. For example Mindfactory can just purposely misprice item X or item Y and suddenly the numbers shift.

 

[DrMrLordX]

It is also the equivalent of looking at Pizza Hut's sales numbers in exacting detail and concluding that hamburgers don't sell well in America. Non-scientific data (such as self-selected DIY buyers from one niche company) is a bad way of understanding the general CPU market. It might help explain the DIY market, but even then it isn't that great. For example Mindfactory can just purposely misprice item X or item Y and suddenly the numbers shift.

First off, if Mindfactory "misprices" items, you'll know it from looking at their store listings. Feel free to gather that data if you like.

Secondly, duh? It only tells you where demand is piling up among experienced users, to the extent that mindfactory can fulfill demand (mindfactory's Zen3 sales are probably hampered by short supplies, for example). It's useful for trend analysis among those who actually care about the performance of a CPU. Nobody is trying to sell the idea that mindfactory is a good indicator of OEM sales when OEM buyers routinely pick from a list of CPUs that DiY users would almost never consider (or CPUs that are OEM-only).

 

[dullard]

Nobody is trying to sell the idea that mindfactory is a good indicator of OEM sales when OEM buyers routinely pick from a list of CPUs that DiY users would almost never consider (or CPUs that are OEM-only).

Just search this very thread and forum for "Mindfactory" and you get things like this (from the same person that started it a few posts above):

Look how is that going in real life, or Intel 6/12 vs AMD 6/12 CPU king of today Desktop market.

Intel CPU is cheeper, but hm 99% people buy only R5 3600.[/URL]

Really, 99% of people buy the AMD R5 3600?

The statistics are very accurate, or more precise for the for bigest Geman retailer Mindfactory.As far i now, only Mindfactory shows(über verkauft or sold over)the number of various sold hardwer parts.

- 80% sold CPU are from AMD

Oh, wait, that is 80% of all CPUs are AMD.

Here is another poster.

If it's about sales you can look at Mindfactory data showing AMD outselling Intel 2:1.

Yes, Mindfactory tells us useful information. But we shouldn't generalize it.

 

[Zucker2k]

It's a case of "some data is better than no data".

Except STEAM data.

 

[Abwx]

Just search this very thread and forum for "Mindfactory" and you get things like this (from the same person that started it a few posts above):

Really, 99% of people buy the AMD R5 3600?


Oh, wait, that is 80% of all CPUs are AMD.

Here is another poster.

Yes, Mindfactory tells us useful information. But we shouldn't generalize it.

During pre Zen era roughly 90% of the CPU sold by mindfactory were Intel, you can eventually find the numbers here or there, so their statistics are very accurate as a representative sample of the the european DIY market, as an exmple here are the sales of some previous Intel CPUs that are still in their offerings :

Sockel 1150

Hier finden Sie 2x Sockel 1150 diverser namhafter Hersteller in unserer Prozessoren (CPU)-Kategorie

www.mindfactory.de

Sockel 1151

Große Auswahl an Sockel 1151 CPU ? Namhafte Hersteller zu Top Preisen ? Hardware » Jetzt entdecken und Blitzangebote nicht verpassen!

www.mindfactory.de

For S1151 the best sellers was actually the 7700K.

 

[dullard]

so their statistics are very accurate as a representative sample of the the european DIY market, as an exmple here are the sales of some previous Intel CPUs that are still in their offerings

The dispute is when people talk about general CPU sales using DIY Mindfactory data. That is vastly different from DIY data.

 

[Hitman928]

Except STEAM data.

Steam data methodology is unknown. We know exactly how Mindfactory gets its numbers. I agree Mindfactory numbers should be used carefully and the source of the numbers understood and we shouldn't be trying to extrapolate beyond Mindfactory's market, but at least we know how the numbers are generated. Even steam numbers I think have some value but should not be used at face value or as an accurate representation of market share numbers at any given period of time.

 

[lobz]

It is also the equivalent of looking at Pizza Hut's sales numbers in exacting detail and concluding that hamburgers don't sell well in America. Non-scientific data (such as self-selected DIY buyers from one niche company) is a bad way of understanding the general CPU market. It might help explain the DIY market, but even then it isn't that great. For example Mindfactory can just purposely misprice item X or item Y and suddenly the numbers shift.

That's a very bad analogue. Mindfactory covers the same market as other retailers. Pizza hut really does not. Also, suggesting a purposeful mispricing over this time frame is just grasping at imaginary straws, because it's also possible that they're purposefully selling Intel products and re-labelling them as AMD products both are possible and extremely unlikely.

 

[Ajay]

Oh geez. Mindfactory 'data' is representative of Mindfactory customers only. Just because it's all we have doesn't make it statistically valuable.
Quote the numbers for 10 top retailers in 10 top markets and I might give a nod that that would represent useful data, otherwise, raspberries.

This is just such a ridiculous argument to have, I wish it would end.

 

[dullard]

Also, suggesting a purposeful mispricing over this time frame is just grasping at imaginary straws

Just try comparing Mindfactory.de prices for AMD products vs say Amazon's prices for AMD products. Then do that for Intel products. You'll find on average AMD products get about a 5% better conversion rate at Mindfactory.

 

[lobz]

Except STEAM data.

No. The same applies to Steam data. Look at the trends and you may get some useful info. I have another good one: JPR market share numbers. All of these are very misleading at first glance, so you have to put them in the proper perspective.