Discussion Intel current and future Lakes & Rapids thread

[Asterox]

Finally, an ADL-S score with power shown at the same time.

https://twitter.com/x/status/1446824964066185223

4.9GHz on the P cores, 3.7GHz on the E cores, MT perf about the same as the 5950X and a CPU package power of 257W. Mind you both of the above are supposed to clock higher on final silicon - both the P and E cores.

Well hey, at least the ST score is really quite impressive. But that's to be expected at this point.

As expected, in Cinebench R20 it will need 240-250W/or 100W+ more power to have same or similar score to R9 5950X.

Page 513 - Discussion - Intel current and future Lakes & Rapids thread

Page 513 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Alder Lake 8+8 Cinebench R20, MT performance per watt is garbage

 

[Hulk]

Nice find. Which app is running when power is 257W?

Thanks. Some other discrepancies there I'm not sure about...

1. CB20 ST score of 768 is same as 12700K score posted elsewhere? 12900K CB20 ST should be more like 817, right? Also CB20 MT of 10545 seems about right so the ST 768 seems off.
2. 4888 all-core clock is supposed to be 5000 for 12900K isn't it? Could be a cooling issue I guess?

Exhausted trying to put this jigsaw puzzle together with so many pieces missing. Can't wait for Ian's dissection of this thing.

 

[dmens]

Mind you both of the above are supposed to clock higher on final silicon - both the P and E cores.

Ummm, QS is release silicon. It is what it is. Don't hold your breath for 4ghz atom cores LOL.

 

[uzzi38]

Ummm, QS is release silicon. It is what it is. Don't hold your breath for 4ghz atom cores LOL.

Oh whoops, minor mistake there, that is the correct all core clock for the E cores isn't it? It's 3.9GHz for a single core - my bad.

P-cores are supposed to go to 5GHz all core iirc, so point still stands there.

 

[dmens]

Even having a high PL for the atoms is absurd and desperate. Hey, let's run the efficiency core at an absolutely inefficient point where the big core would yield better performance at less power! What a joke.

 

[mikk]

So Alderlake-M has a GT3 variant. Sites are saying GT3 is also 96EUs, but maybe it's larger?

ADL-M is GT2 only: https://videocardz.com/newz/intel-alder-lake-p-m-mobile-power-limits-detailed-in-coreboot-patch

The GT3 entry is old, it doesn't mean anything. There was a GT3 entry for ICL-U just to give you a relatively new example: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...s-rapids-thread.2509080/page-85#post-39694869

It never came out...

Finally, an ADL-S score with power shown at the same time.

Not at the same time unfortunately. There is no power consumption test from Cinebench or CPUz.

 

[Ajay]

So, for FPU heavy work loads, it's AIOs , custom LC or go home. Given the top clocks on the P cores, it's probably throttling a bit at 108C (or just a poor quality chip or HS).

Getting annoying waiting for Adlerlake to drop so we get 'real' benches.

 

[uzzi38]

Not at the same time unfortunately. There is no power consumption test from Cinebench or CPUz.

Those two will be lower for sure. AIDA64 FPU is a much heavier load by comparison, although still not as heavy as Prime95

 

[Abwx]

AIDA64 FPU, so it is indicative of a worst(-ish) case workload.

Aida test use less power than Cinebench, and Prime 95 use roughly 20% more than the latter.

 

[lobz]

Thanks. Some other discrepancies there I'm not sure about...

1. CB20 ST score of 768 is same as 12700K score posted elsewhere? 12900K CB20 ST should be more like 817, right? Also CB20 MT of 10545 seems about right so the ST 768 seems off.
2. 4888 all-core clock is supposed to be 5000 for 12900K isn't it? Could be a cooling issue I guess?

Exhausted trying to put this jigsaw puzzle together with so many pieces missing. Can't wait for Ian's dissection of this thing.

Why?

If it's the same core and it's running at a given frequency (4.9 at this particular case), the CPU could even be called the Nissan 300ZX TwinTurbo, the score will still be the same.

 

[Hulk]

Why?

If it's the same core and it's running at a given frequency (4.9 at this particular case), the CPU could even be called the Nissan 300ZX TwinTurbo, the score will still be the same.

Just saying the single thread score doesn't seem right that's all.

 

[Exist50]

Even having a high PL for the atoms is absurd and desperate. Hey, let's run the efficiency core at an absolutely inefficient point where the big core would yield better performance at less power! What a joke.

Lmao, you claimed repeatedly that it's impossible for Atom to hit anywhere near that high. And yet here we are.

 

[mikk]

Aida test use less power than Cinebench, and Prime 95 use roughly 20% more than the latter.

AIDA64 FPU uses much more power than Cinebench, it's an excessive AVX2 (or AVX512) instruction stress test which is a known power hog on Intel CPUs.

 

[deasd]

Thanks. Some other discrepancies there I'm not sure about...

1. CB20 ST score of 768 is same as 12700K score posted elsewhere? 12900K CB20 ST should be more like 817, right? Also CB20 MT of 10545 seems about right so the ST 768 seems off.
2. 4888 all-core clock is supposed to be 5000 for 12900K isn't it? Could be a cooling issue I guess?

Exhausted trying to put this jigsaw puzzle together with so many pieces missing. Can't wait for Ian's dissection of this thing.

Since 250w consumption @4.9 I assume this is all 8+8 ADL can go without 360mm AIO cooler... so no surprise or odd here. I already heard rumors before about cooler needed for ADL so it's all in line with what we expected.

 

[dmens]

Lmao, you claimed repeatedly that it's impossible for Atom to hit anywhere near that high. And yet here we are.

Hah, nice try. We have a sample point of atoms at 3.7ghz, not 4ghz as you claimed. Muh process scaling! Muh super-fin!

Oh, and note this is already a benchmark scenario where any semblance of power limitation has already been thrown out the door. So... even if some Intel benchmarketer decides to have some PL for the atoms where a single core can burst to 3.9ghz, which even if true (still just a baseless rumor at this point), would be utterly idiotic since the big core should be handling that kind of work, it still is not the 4ghz you claimed, and it is not the easy leap you claimed it would be.

I called this months ago. Power hungry trash as always. And as I said, your linear frequency extrapolation is also trash.

 

[LightningZ71]

Remember, four of the E cores at 3.5Ghz+ probably has more total MT throughput than a P core running at 5+Ghz with HT. The E cores are a play for more MT throughout to compete with AMD's higher core count products on the desktop.

 

[JoeRambo]

AIDA64 FPU uses much more power than Cinebench, it's an excessive AVX2 (or AVX512) instruction stress test which is a known power hog on Intel CPUs.

I think it might depend on what CB version and Aida version is. Maybe AIDA is burning more power on RocketLake with AVX512, but with my limited testing on Skylake 10C static 5.1Ghz HT off => AIDA FP is 185W, CB20 is 210W. THO an important thing to note, that this is with static volts, maybe stock CPU is requesting more volts when AIDA FP test is on (compared to what it requests with CB20 ) and that skyrockets watts?

Still, considering my SKL 10C is getting sub 5k for 210W, 250W for double CB20 score is epic efficiency gain.

 

[Abwx]

I think it might depend on what CB version and Aida version is. Maybe AIDA is burning more power on RocketLake with AVX512, but with my limited testing on Skylake 10C static 5.1Ghz HT off => AIDA FP is 185W, CB20 is 210W. THO an important thing to note, that this is with static volts, maybe stock CPU is requesting more volts when AIDA FP test is on (compared to what it requests with CB20 ) and that skyrockets watts?

Still, considering my SKL 10C is getting sub 5k for 210W, 250W for double CB20 score is epic efficiency gain.

Variable clock will change nothing, your exemple show that there s more instructions executed per cycle in CB, otherwise it wouldnt consume more power at same frequency.

Methink that Aida stress the FPU with a tiny routine that use a very little part of the L2 while CB require a bigger part of the pipeline since it makes large use of the ALUs, FTR in CB R15 only 70% of the instructions are SSE2.

Overall Aida is a realistic case that certainly do not exceed general purpose MT workloads power drain.

 

[Exist50]

Hah, nice try. We have a sample point of atoms at 3.7ghz, not 4ghz as you claimed. Muh process scaling! Muh super-fin!

It's 3.7GHz all core, 3.9GHz single core boost. Meanwhile, you claimed that Atom couldn't hit near 4.0GHz even when overclocked, never mind 3.9GHz at stock. And, might I point out, that's almost identical frequency scaling as Sunny Cove -> Willow Cove saw.

Page 420 - Discussion - Intel current and future Lakes & Rapids thread

Page 420 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

In fact, you explicitly claimed that the leak of these exact clocks was "nonsense", as well as that embarrassing diatribe about your EE degree.

Page 475 - Discussion - Intel current and future Lakes & Rapids thread

Page 475 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

So now that we've demonstrated that you haven't the slightest clue what you're talking about, maybe stop polluting the thread with even more nonsense.

 

[igor_kavinski]

i3-1115G4 is a great CPU for budget conscious consumers. Looks like Intel will have another winner on its hands for the masses with the Alder Lake i3. Negative points for AMD for shoving their "Zen 2" based Ryzen 3 5300U down the throats of unsuspecting consumers, making them think they are buying the same technology that powers the desktop 5000 series.

 

[dmens]

It's 3.7GHz all core, 3.9GHz single core boost. Meanwhile, you claimed that Atom couldn't hit near 4.0GHz even when overclocked, never mind 3.9GHz at stock. And, might I point out, that's almost identical frequency scaling as Sunny Cove -> Willow Cove saw.

Page 420 - Discussion - Intel current and future Lakes & Rapids thread

Page 420 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

In fact, you explicitly claimed that the leak of these exact clocks was "nonsense", as well as that embarrassing diatribe about your EE degree.

Page 475 - Discussion - Intel current and future Lakes & Rapids thread

Page 475 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

So now that we've demonstrated that you haven't the slightest clue what you're talking about, maybe stop polluting the thread with even more nonsense.

You do realize the high PL settings are overclocking, right? Or do you actually think maximum thermal velocity boost/PBO/whatever is "stock"? Don't answer, that is a rhetorical question.

Try actually look at the voltage limited Tremonts versus blowout voltage Gracemonts to get to this frequency before claiming any sort of "scaling". Pro-tip: shoving more volts to get more frequency is not scaling, it is desperation.

 

[Exist50]

You do realize the high PL settings are overclocking, right?

By definition, they're not.

Not that it even matters, when you claimed these exact frequencies, or anything near them, were completely impossible.

 

[igor_kavinski]

Pro-tip: shoving more volts to get more frequency is not scaling, it is desperation.

That should be on a T-shirt

 

[dmens]

By definition, they're not.

Not that it even matters, when you claimed these exact frequencies, or anything near them, were completely impossible.

So you actually think a manufacturer doing exactly what a consumer overclocker does (throw more volts at a chip to get frequency and ignore thermal constraints) is not overclocking, just because… reasons?

You might have a point if Intel/AMD just stopped specifying a stock frequency and TDP completely and simply state that the part will consume every last bit of thermal capacity presented and that is “stock operation”. Until then, you are wrong, as usual.

 

[Exist50]

So you actually think a manufacturer doing exactly what a consumer overclocker does

I can't believe I have to explain this, but overclocking, by definition, is going beyond what the manufacturer offers out of the box.

Now that's enough feeding the trolls from me. Anyone with eyes can look at our two predictions and see who was right.