Discussion Intel current and future Lakes & Rapids thread

[Zucker2k]

I see his 5600Х cpu-z mt score is a bit on a low side.

Spoiler: MT 5036 @76W default

View attachment 52120

Is this stock or tweaked? Not that it matters, reviews should settle this sometime in the near future.

 

[PJVol]

Is this stock or tweaked?

76W is stock PPT limit, so yes.
Here is tweaked (not OC-ed, just PBO & CO):

Spoiler: 86W PPT max

 

[Zucker2k]

76W is stock PPT limit, so yes.
Here is tweaked (not OC-ed, just PBO & CO):

Spoiler: 86W PPT max

View attachment 52122

Stock, yes, but are you using Curve Optimizer or applying some type of undervolting? Because if you are, your result is not exactly stock.

 

[PJVol]

Of course, PBO and CO were disabled.

 

[mikk]

So 12400 has the same efficiency or a bit better than 5600x at least If It's at 4GHz, right? How did you come to the conclusion that ADL is hugely more efficient than Tiger Lake? Zen3 is not that much more efficient at higher power draw than Tiger Lake.
Techspot

At 75W Zen3 is about 10-13%. I can't say that is a huge increase in efficiency.

Tigerlake is clearly below Zen 3 in efficiency, that's why. This is based on several reviews or other tests like this. The Tigerlake based NUC didn't look great either. The i5-12400 seems to have a 5-10% efficiency advantage over a comparable Zen 3, this is overall a huge improvement to me.

 

[TESKATLIPOKA]

Tigerlake is clearly below Zen 3 in efficiency, that's why. This is based on several reviews or other tests like this. The Tigerlake based NUC didn't look great either. The i5-12400 seems to have a 5-10% efficiency advantage over a comparable Zen 3, this is overall a huge improvement to me.

My graph also shows that Tiger Lake is under Zen 3, but with increased TDP the difference becomes smaller.
Considering how low Alder Lake needs to be clocked to be within 125W(P: ~3.6GHz and E: ~2.8GHz), based on the R20 leak, I seriously doubt It is more efficient than Zen 3.

 

[Accord99]

I see his 5600Х cpu-z mt score is a bit on a low side.

The CPUZ benchmark, especially version 17, takes less power than the AIDA64 FPU stress test. Since the power consumption of the 12400 was not reported for the CPUZ bench, the example had to first determine the clock speed of the 5600X needed to match the 12400 in CPUZ (4.5GHz) then run the AIDA64 stress test at that clock speed to come up with the closest equivalent power figure.

 

[PJVol]

the example had to first determine the clock speed of the 5600X needed to match the 12400 in CPUZ

I don't get what relation cpu clock has to power consumed when comparing different architectures. What was the goal then? Can it be approximated this way?

PS: If I get what was on screenshot right, during the AIDA FPU test it consumed 78W ?
What makes you think then, that in CPU-Z it would consume less, I mean what this is based on?

 

[mikk]

My graph also shows that Tiger Lake is under Zen 3, but with increased TDP the difference becomes smaller.
Considering how low Alder Lake needs to be clocked to be within 125W(P: ~3.6GHz and E: ~2.8GHz), based on the R20 leak, I seriously doubt It is more efficient than Zen 3.

I'm not talking about 125 watts nor E-cores, I'm talking about Golden Cove and mobile. Once again the 12400 (it's a 6+0 model) looks very promising and therefore it's promising for mobile as well. 78W in AIDA64 FPU could result in 70W Cinebench MT running at 4.0 Ghz. 5600x requires the same if not more power in Cinebench and runs slower. I haven't seen any reliable 125W ADL-S test by the way. The leak you are probably talking about isn't reliable, the 125W score is too low a tester hinted.

 

[TESKATLIPOKA]

I'm not talking about 125 watts nor E-cores, I'm talking about Golden Cove and mobile. Once again the 12400 (it's a 6+0 model) looks very promising and therefore it's promising for mobile as well. 78W in AIDA64 FPU could result in 70W Cinebench MT running at 4.0 Ghz. 5600x requires the same if not more power in Cinebench and runs slower. I haven't seen any reliable 125W ADL-S test by the way. The leak you are probably talking about isn't reliable, the 125W score is too low a tester hinted.

What I posted should be an official Lenovo presentation If I understood It correctly.
Golden cove should be less efficient than Gracemont and you don't really need higher clocks to achieve those 125W. It's enough to add 2 more Golden Cove cores and 8 Gracemont cores.
You say 78W for a 4GHz 6 Core Golden Cove is very promising, so I tried to apply It to 12900K.
Let's say from those 70W -> 10W is per core at 4GHz and 10W for uncore, If you add 2 more cores, then you end up with 90W, that leaves 35W for 8x Gracemont cores.

You are right, that 7492 points for 12900K looks pretty low, If 12400 really manages 4784 points. It would mean only 56% better performance, and that is a very low score for 8C+8c If an 8 core at 4GHz should manage 6377 points(4784 * 1.333). It would mean 8 Gracemont cores provide only ~18% better performance, that's worse than what SMT provides. I would expect ~50% higher performance in Cinebench.

 

[PJVol]

78W in AIDA64 FPU could result in 70W Cinebench MT running at 4.0 Ghz. 5600x requires the same if not more power in Cinebench and runs slower

What are you guys talking about? How can cpu consume less or more than what is deifined by the power limit preset in these tests?

 

[mikk]

You are right, that 7492 points for 12900K looks pretty low, If 12400 really manages 4784 points. It would mean only 56% better performance, and that is a very low score for 8C+8c If an 8 core at 4GHz should manage 6377 points(4784 * 1.333). It would mean 8 Gracemont cores provide only ~18% better performance, that's worse than what SMT provides. I would expect ~50% higher performance in Cinebench.

In Cinebench R23 there is only a ~15% difference between 125W and unlimited (260W) on a 12900k.

What are you guys talking about? How can cpu consume less or more than what is deifined by the power limit preset in these tests?

No power limit but limited by the multithread turbo clock speed in the first place. i5-12400 is limited to 4.0 Ghz when all cores are loaded. Stress tests with higher AVX load are usually more demanding on Intel CPUs, depending on the workload the power consumption can differ.

 

[Accord99]

I don't get what relation cpu clock has to power consumed when comparing different architectures. What was the goal then? Can it be approximated this way?

It's a rough approximation but it's something for the curious and enthusiasts to do while waiting for the review embargo to lift.

PS: If I get what was on screenshot right, during the AIDA FPU test it consumed 78W ?
What makes you think then, that in CPU-Z it would consume less, I mean what this is based on?

Based on the behavior on Tiger Lake-H that I see. On an Inspiron 16 Plus laptop, at 8x3.4GHz (undervolted) results in 56W maximum package power in AIDA64 FPU, 50W in Cinebench R23 and just slightly more than 36W in CPUZ v17 benchmark.

 

[PJVol]

Based on the behavior on Tiger Lake-H that I see

Oh really...? Is it mobile parts specific or it applies to all 11 series? Didn't know that.
I just have both mobile and desktop ryzen chips (5700g and 5600x) and their behavior is rather straightforward regarding power, i.e. consuming whatever is allowed by power or sustained current limit.

imited by the multithread turbo clock speed in the first place

Ah, yes, I see. Didn't know that.

 

[IntelUser2000]

It would mean 8 Gracemont cores provide only ~18% better performance, that's worse than what SMT provides. I would expect ~50% higher performance in Cinebench.

I swear, Gracemont goes from being a Bulldozer to a Conroe and back to a Bulldozer every leak.

 

[TESKATLIPOKA]

I swear, Gracemont goes from being a Bulldozer to a Conroe and back to a Bulldozer every leak.

I was only discussing the low score from supposedly official Lenovo marketing slides for 125W power limit.
I was not discussing If Gracemont is weak or strong.

 

[Abwx]

In Cinebench R23 there is only a ~15% difference between 125W and unlimited (260W) on a 12900k.

It means that at 125W the CPU will do the bench at 180W on average, doing half the bench at 240W before temperature limiting the CPU at 125W for the second half of the bench.

Intel Alder Lake: i9-12900K, i7-12700K & i5-12600K vorgestellt

Mit einem großen Schritt eröffnet Intel Alder Lake die Hybrid-Ära. Core i9-12900K, i7-12700K & i5-12600K machen den Anfang. Ein Überblick.

www.computerbase.de

 

[DrMrLordX]

It means that at 125W the CPU will do the bench at 180W on average, doing half the bench at 240W before temperature limiting the CPU at 125W for the second half of the bench.

But if PL1=PL2=125W then that's not the case.

 

[Abwx]

But if PL1=PL2=125W then that's not the case.

The difference would be bigger than 15%, from brick wall limit at 125W to 240W unlimited the perf difference should be around 30%.

 

[DrMrLordX]

The difference would be bigger than 15%, from brick wall limit at 125W to 240W unlimited the perf difference should be around 30%.

Guess we'll see in a few days if reviewers are smart enough to test at fixed power levels.

 

[coercitiv]

Guess we'll see in a few days if reviewers are smart enough to test at fixed power levels.

On this topic...

https://twitter.com/x/status/1455132419598790657

We all enjoy a new BIOS just days before NDA is lifted, now all we need is a Win11 performance update and we're all set for crazy town.

 

[AtenRa]

What I really like to see from reviews is taking the 12900K, disabling all 8 e-cores and compare against Ryzen 5800X at the same power consumption.
That way we could see the performance and efficiency of the new mArchitecture and Intel 7 process against last year ZEN3 and TSMC 7nm.

 

[moinmoin]

we're all set for crazy town.

Will be interesting to watch how transparent all the reviewers will be stating the exact version of all those moving targets. Add the inherent difficulty to do "correct" performance tests on hybrid designs and we are set for crazy town indeed.

 

[Mopetar]

Welcome to crazy town where every day is new BIOS day.

Although it sucks because results and work has to be throw out, it would be interesting to see how much the performance changes across multiple daily BIOS revisions.

 

[CakeMonster]

I had to wait months for my 5900X to arrive, and I thought the BIOS updates were a bit much even then. I'm kind of glad I don't have to deal with this now, adding W11 and DDR5 on top of it.

I have no doubt AL will be solid in about 6 months time. I'm very much looking forward to Raptor Lake, I figure tons of issues will have been sorted out by then and there should not be need for a whirlwind of BIOS and software updates given its more incremental design.