Discussion Intel current and future Lakes & Rapids thread

[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.

 

[lobz]

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.

However it's probably the other way around.

 

[uzzi38]

However it's probably the other way around.

It scores worse on ST than Zen 3 at 800MHz.

Ignore the scoring.

 

[Ajay]

It scores worse on ST than Zen 3 at 800MHz.

Ignore the scoring.

Geez, do companies do this just to punk us? I get really annoyed with geek bench scores on a regular basis.

 

[coercitiv]

RIP Lakefield

So much for all the fancy graphs Intel showed us while talking about the advantages of hybrid configs in low power products. Hopefully Lakefield was a simple test vehicle, similar to Cannon Lake, and it was not market rejection that resulted in it's early demise.

 

[Hulk]

RIP Lakefield

So much for all the fancy graphs Intel showed us while talking about the advantages of hybrid configs in low power products. Hopefully Lakefield was a simple test vehicle, similar to Cannon Lake, and it was not market rejection that resulted in it's early demise.

By "market rejection" do you mean poor sales?

 

[coercitiv]

By "market rejection" do you mean poor sales?

No, I mean OEMs didn't bother with it. We had the Samsung Galaxy Book S at launch, a Levovo X1 Fold tablet, and a product from Microsoft that never made it to the market.

 

[Hulk]

No, I mean OEMs didn't bother with it. We had the Samsung Galaxy Book S at launch, a Levovo X1 Fold tablet, and a product from Microsoft that never made it to the market.

Why do you think those products never made it to market?

 

[Mopetar]

Who really wants to serve as a guinea pig for some new Intel product when it really doesn't offer suitable advantages over other Intel products and the software support clearly isn't there yet either?

There are plenty of great ideas that flop as products because the technology just isn't there yet. Apple really made their first iPad in the early 90's only it was called the Newton. It was kind of cool for the time, but a lot of the hardware and software just wasn't ready for what Apple was hoping to accomplish with that product and it eventually got killed off.

Hopefully Intel learned a lot as a part of bringing it to market and that knowledge is going to make future products and projects more successful. Even the first versions of what're regarded as successful products often don't age well. Consider the original iPhone compared to the models that released only a few years afterwards.

 

[semiman]

RIP Lakefield

So much for all the fancy graphs Intel showed us while talking about the advantages of hybrid configs in low power products. Hopefully Lakefield was a simple test vehicle, similar to Cannon Lake, and it was not market rejection that resulted in it's early demise.

Yup it's more like for a notebook vendors for thermal design and Microsoft schedulers. Legacy's remains to be seen, in mass adoption of Alder Lake(for scheduler) and Ponte Vecchio(foveros).

 

[Zucker2k]

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.

Imagine this chip on desktop with proper cooling. I think Alderlake is going to be a real beast.

https://www.reddit.com/r/intel/comments/nfmytu

 

[Abwx]

Performance is very good but not its perf/watt, they better improve their 10nm with ADL if they want to be competitive at the lower end of the power usage.

https://static.techspot.com/articles-info/2262/bench/27.png

Intel Core i7-11800H Review: Tiger Lake H45 Put to the Test

The Core i7-11800H is a substantial upgrade for Intel, bringing Tiger Lake designs up to 8 cores, suitable for gaming and mobile workstation systems which compete against...

www.techspot.com

Edit : At 70W it does a little over 11000pts, so the guy at Reddit is misleading people, to get 14500pts it should be set at a constant 140W at least...

 

[Accord99]

It's because Techspot didn't undervolt.

 

[Zucker2k]

It's because Techspot didn't undervolt.

13k+ even with the monitoring app and constant mouse movement

 

[Abwx]

It's because Techspot didn't undervolt.

Of course, because undervolting is to eat on the stability margin, it can pass a few CB runs but it will crash with Linpack and other Prime 95.

FTR once minimal stability is achieved 20 to 30% excess power (depending of the part of the V/F curve) is necessary to achieve robust stability, for servers who are indeed processing critical tasks the margin is even higher and can get up to 40%.

Back on TGL, this chip has a 160W PL2 but on a laptop it doesnt last long and it seems to hoover around 100W if the cooling apparatus is adequate for a 12700 score, wich validate Techspot measurements.

https://www.notebookcheck.net/Razer-Blade-15-Advanced-Model-11th-Gen-Intel-2021-Review-Small-Changes-Big-CPU-Gains.548802.0.html

 

[Mopetar]

Even at stock it's quite impressive and actually gives Intel something that can go toe to toe with AMD. The scaling looks good enough for desktop to be competitive as well. Going to be some great value if Intel releases a 12600 at the ~$200 range. The might not quite be able to catch AMD on core count, but right now most consumers really don't need more than 8 cores and even 6 will get most people by just fine.

 

[semiman]

Actually, problem of Tiger lake lies in architecture, not in 10SF. 10SF is good enough. But Sunny Cove-based cores are not performing well, even if they are using ~30% bigger CPU architecture. For example, # of reorder buffer entry of Sunny Cove is 352, but Zen 3 uses 224. These all apply to other core blocks. Intel Coves are bigger but suffer from underutilization.

 

[AMDK11]

Zen 2 - 224
Zen 3 - 256

 

[DrMrLordX]

Back on TGL, this chip has a 160W PL2 but on a laptop it doesnt last long and it seems to hoover around 100W if the cooling apparatus is adequate for a 12700 score, wich validate Techspot measurements.

That isn't very good. A 3700x with PPT of 88W can do around 12200.

 

[Abwx]

That isn't very good. A 3700x with PPT of 88W can do around 12200.

From 12200 to 12700 there someting like 15-20% higher power given that at those frequencies it s out of the optimal V/F curve, so the perf/watt is comparable despite TGL s SC being a huge core, the 5800X benefit not only of a higher IPC but use also an enhanced process in respect of Zen 2.

When you look at the basic math 15% lower power from the improved process and 5% lower frequency for same throughput (wich result in 15% lower power) and you have a grand total of about 30% lower power than TGL at same throughput, as displayed in NBC and Techspot tests.

 

[eek2121]

Of course, because undervolting is to eat on the stability margin, it can pass a few CB runs but it will crash with Linpack and other Prime 95.

FTR once minimal stability is achieved 20 to 30% excess power (depending of the part of the V/F curve) is necessary to achieve robust stability, for servers who are indeed processing critical tasks the margin is even higher and can get up to 40%.

Back on TGL, this chip has a 160W PL2 but on a laptop it doesnt last long and it seems to hoover around 100W if the cooling apparatus is adequate for a 12700 score, wich validate Techspot measurements.

https://www.notebookcheck.net/Razer-Blade-15-Advanced-Model-11th-Gen-Intel-2021-Review-Small-Changes-Big-CPU-Gains.548802.0.html

Where do you get the 160W PL2 figure from? maybe for the 11900kb, but the last I checked, PL2 was much lower. I will give you a link when I get home in a few days.

Laptops also use tau and pl1/pl2 in a much more proper manner, with the laptops I have seen following Intel guidance. This means that they should follow TDP values pretty closely.

Note that the only issue with Intel’s PL1/PL2/PLx system is that motherboard OEMs are allowed to override it. Bursty performance that briefly consumes more power is a good thing because it allows work to be completed quicker and allows the chip to get back to sleep.

Also your comment about TGL-H consuming a constant 100W is false. Once tau expires the chip falls back to PL1. Don’t take my word for it: https://www.anandtech.com/show/16680/tiger-lake-h-performance-review/2

 

[Abwx]

Where do you get the 160W PL2 figure from? maybe for the 11900kb, but the last I checked, PL2 was much lower. I will give you a link when I get home in a few days.

Laptops also use tau and pl1/pl2 in a much more proper manner, with the laptops I have seen following Intel guidance. This means that they should follow TDP values pretty closely.

Note that the only issue with Intel’s PL1/PL2/PLx system is that motherboard OEMs are allowed to override it. Bursty performance that briefly consumes more power is a good thing because it allows work to be completed quicker and allows the chip to get back to sleep.

Also your comment about TGL-H consuming a constant 100W is false. Once tau expires the chip falls back to PL1. Don’t take my word for it: https://www.anandtech.com/show/16680/tiger-lake-h-performance-review/2

I posted the link above, here it is again, the 160W PL2 is mentionned in top of the article :

https://www.notebookcheck.net/Razer-Blade-15-Advanced-Model-11th-Gen-Intel-2021-Review-Small-Changes-Big-CPU-Gains.548802.0.html

And it can be checked here in their pics :

CPU sustained power during CB runs can be deducted here :

Laptop total power for CPU loading is 172W for 30s and then it stabilise at 124W, now how much time to do a complete CB run.??.

Average CPU power during a run exceed 100W..

 

[eek2121]

I posted the link above, here it is again, the 160W PL2 is mentionned in top of the article :

https://www.notebookcheck.net/Razer-Blade-15-Advanced-Model-11th-Gen-Intel-2021-Review-Small-Changes-Big-CPU-Gains.548802.0.html

And it can be checked here in their pics :

CPU sustained power during CB runs can be deducted here :

Laptop total power for CPU loading is 172W for 30s and then it stabilise at 124W, now how much time to do a complete CB run.??.

Average CPU power during a run exceed 100W..

In all 3 of the images showing hwinfo, CPU package power was 65W or less with a peak of 103W including during the prime95 stress test. Given metrawatt directly contradicts hwinfo, I would take metrawatt numbers with a grain of salt.

EDIT: Metrawatt appears to measure full system power. That is where your confusion lies.

EDIT: Curious as to why total system power is higher?

1. Razer is pushing extremely high refresh rate panels. This pushes up power consumption considerably.
2. Current Razer models use RAM running at higher clock speeds. This also contributes
3. The reviewed Razer model includes a PCIE 4.0 SSD. This ALSO contributes significantly to the power numbers.

 

[Abwx]

In all 3 of the images showing hwinfo, CPU package power was 65W or less with a peak of 103W including during the prime95 stress test. Given metrawatt directly contradicts hwinfo, I would take metrawatt numbers with a grain of salt.

EDIT: Metrawatt appears to measure full system power. That is where your confusion lies.

Full system power at 172W for 30 seconds with idle power at 42W, this amount to 130W delta, that s 100W CPU power once you account for the losses in the AC adaptator and in the MB VRMs, assuming CPU power at idle is 0W...

Besides, duration of the 103W "peak" in HVinfo is 30 seconds according to the measurement at the main, how is that difficult to grasp.?.

 

[Timmah!]

Performance is very good but not its perf/watt, they better improve their 10nm with ADL if they want to be competitive at the lower end of the power usage.

https://static.techspot.com/articles-info/2262/bench/27.png

Intel Core i7-11800H Review: Tiger Lake H45 Put to the Test

The Core i7-11800H is a substantial upgrade for Intel, bringing Tiger Lake designs up to 8 cores, suitable for gaming and mobile workstation systems which compete against...

www.techspot.com

Edit : At 70W it does a little over 11000pts, so the guy at Reddit is misleading people, to get 14500pts it should be set at a constant 140W at least...

And this has 8 Golden Cove cores? How much gets 5950X and 32C Zen2 TR?