Question Alder Lake - Official Thread

[Hulk]

With the release of Alder Lake less than a week away and the "Lakes" thread having turned into a nightmare to navigate I thought it might be a good time to start a discussion thread solely for Alder Lake.

 

[Makaveli]

Right now the only chips out are the top end ones and in limited supply. The people buying these are the ones who will also spend a lot on an expensive board. Cheaper options will come with the release of the low-end and mid-range parts.

Call it the early adopter tax or whatever you will, but it's always like this. Let's just be happy that Intel has something worth the price for a change.

In Canada availabilty for all 3 chips are high my local store has lots of inventory.

 

[Zucker2k]

12400f listed......

https://www.reddit.com/r/intel/comments/qo6spr

 

[Arkaign]

12400f listed......

https://www.reddit.com/r/intel/comments/qo6spr

 

[Zucker2k]

Also, based on everything that we've seen so far, Golden Cove seems to be on par or better in power efficiency than Zen 3 Core for Core, at the same clock. The inefficiency only stems from the massive all-core frequencies Intel is pushing the the flagship 12900k, especially. Looking forward to some detailed power analysis tomorrow.

Not bad, huh? Intel7 is on par or better than TSMC 7nm, based on all the proper and detailed power consumption analyses trickling in. Called it. ADL-S is not a power hog. It may, in fact, be the most efficient x86 arch right now. Imagine that!
All of a sudden, all the die-hard Intel critics, including a few self-proclaimed chip engineers of ex-Intel stock, are looking like quack psychics right now.
All the 'expert analyses' you spewed on the pages of this forum for years, the condescending attitude, and mispredictions about the hybrid system and scheduling nightmare awaiting ADL-S on windows, you owe your readers an explanation how you could've gotten things so wrong when the writing has been on the wall for so long. Should be an interesting read.

 

[DAPUNISHER]

UHD770 with DDR5 is murking the UHD750 in Linux performance - Phoronix Linux tests

Strange Brigade is the only game it didn't crack the console like fully playable 30fps barrier on. And driver improvements may fix that soon. Can't wait to see what the Xe variant with DDR5 can do. Might take gaming on an ultra portable to the next level.

 

[DrMrLordX]

TSMC 7nm

Careful, TSMC doesn't have "7nm". It has N7, N7P, N7+, and (technically) N6.

N7+ is pretty niche, in that practically no one uses it anymore, but it is significantly more performant and dense than N7. But if you were trying to say that 10ESF is competitive with N7 then you may well be correct, at least as far as performance goes.

 

[TESKATLIPOKA]

Scores and power consumption taken from the Computerbase review.
i9-12900K @ 4,8 GHz P + 3,7 GHz E-cores -> 27232
i9-12900K @ 4,8 GHz P-Cores -> 19837
You end up with:
i9-12900K 3,7 GHz E-cores -> 7795

Now I will calculate a hypothetical ADL-P 2C(4.8GHz)+8c(3.7GHz).
19837/4 = 4959 points + 7795 points for a total of 12754 points in Cinebench R23. Power consumption was 69W in Corona Benchmark 1.3 without any limit set.

This doesn't look so great compared to ADL-H 6C(3GHz)+8c(2.4GHz) limited to 35W with a score of 14288 points in R23, but that's not the main point.
I wanted to know how It will score at the same clocks as the above ADL-H.

2x P-core(3GHz): 4959/48*30 = 3099 points
8x E-core(2.4GHz): 7795/37*24 = 5056 points
For a total of 8155 points with an unknown package power.

I am sure this can be under 20-25W, but I am not so sure about fitting within 15W, so I will lower the clocks even more to be sure.

2x P-core(2.4GHz): 4959/48*24 = 2480 points
8x E-core(2GHz): 7795/37*20 = 4213 points
For a total of 6693 points.
It's true I lowered the clocks quite a bit, but let's consider It as the worst scenario and In reality It will score anywhere between 6693-8155 points at 15W.

For comparison (Teschspot R23 MT):
R7 5800U manages 7480 points with 15W TDP and 9284 points with 25W TDP.
i7-1165G7 manages 3679 points with 15W TDP and 5189 points with 28W TDP.
i7-1185G7 manages 3500 points with 15W TDP and 5301 points with 28W TDP.

As you can see, against Ryzen It can be slower or faster depending on how high It can clock with a limited power budget, but even at worst It looks pretty competitive.
Against ULV Tiger Lake the MT performance is brutally increased, there is nothing more to say.

 

[Carfax83]

Not bad, huh? Intel7 is on par or better than TSMC 7nm, based on all the proper and detailed power consumption analyses trickling in. Called it. ADL-S is not a power hog. It may, in fact, be the most efficient x86 arch right now. Imagine that!
All of a sudden, all the die-hard Intel critics, including a few self-proclaimed chip engineers of ex-Intel stock, are looking like quack psychics right now.
All the 'expert analyses' you spewed on the pages of this forum for years, the condescending attitude, and mispredictions about the hybrid system and scheduling nightmare awaiting ADL-S on windows, you owe your readers an explanation how you could've gotten things so wrong when the writing has been on the wall for so long. Should be an interesting read.

*shots fired*

Seriously though, I think you're right. Alder Lake-S has been castigated on this forum throughout the months before launch based on leaks that provided snapshots with incomplete context, engineering samples with performance issues and a general bias against Intel, which I can't say I blame them for. Intel has been so far behind the curve the past few years, and making mistake after mistake whereas AMD and Apple have been executing like clockwork. Intel had to own the benchmarks to compete with the high end Zen 3 parts, even if it meant pushing the chips way out of their efficiency range otherwise Alder Lake may have been pinned as a weak comeback rather than a strong first strike. It's going to be funny to see the table turns over the next few months, especially when Alder Lake M drops early next year. I have a feeling the Apple fans in particular may have to do some soul searching

As a layman though, I was really skeptical of the big/little approach on desktop, but after finally seeing it in action, I think it does have some merit. I'm sure Intel will continue to increase the IPC and performance of their efficiency cores, and hopefully implement AVX-512 so that the technology does not fall to the wayside.

 

[arandomguy]

12400f listed......

https://www.reddit.com/r/intel/comments/qo6spr

The other consideration is how much B series boards will drive down platform costs. At this point we know the CPUs themselves will priced rather well for what they you get. But if the motherboards are approaching or exceeding CPU costs that becomes problematic.

The other issue also is down at this end even the cooling cost is more of factor. It'll be interesting to see what they bundle as the HSF. Not sure how much LGA 1700 support has trickled down to the budget coolers yet in terms of actual availability.

In Canada availabilty for all 3 chips are high my local store has lots of inventory.

But the issue is that motherboard selection and availability is more problematic. Particularly if you're looking at a value approach especially if pairing with a 12600k. At least personally I'm already not enamored with spending $200+ CAD on a motherboard already, much less $300+ especially if pairing with a 12600k. But then again I don't really put value on excess motherboard features, especially the VRM overselling that's become the norm (thanks to overhype reviewing).

The Rocket Lake launch in Canada had this issue has well. Initial motherboard selection was poor. By the time motherboard selection stabilized the value non K CPUs basically ran out of availability and/or became marked up and basically never replenished. Although I don't expect the latter issue this time around as I think part of that was likely due to not wanting excess stock in prep for ADL.

 

[Zebo]

HWUB - 12600K Review.
My favorite of the bunch (because I'm poor). I'd wait for B660 MB, and use DDR4 to keep costs down:

Matches 5800X performance and power usage.

12600K is a absolute beast for us 99%ers! Intel desperately needed a win to stay competitive and to keep AMD honest and has one here. Only thing that bothers me is socket 1700 motherboards are $200, double the Ryzen boards - effectively erasing the price/performance lead. That said it, still leads. 5600x will have to go down to $225-$250. So basically you'll be able to get into a 10% slower setup for $350 instead of $520. That might change your mind.

 

[TheELF]

The other issue also is down at this end even the cooling cost is more of factor. It'll be interesting to see what they bundle as the HSF. Not sure how much LGA 1700 support has trickled down to the budget coolers yet in terms of actual availability.

The pl2 for the 6+4 core one is 150W so even if you think that max turbo power is the default, that is going to be 150W sustained, the 12400f will be 6core only but I think it will have the same limits. Pl1 of 125 pl2 of 150W

Intel® Core™ Ultra and Intel® Core™ Processors Technical Resources

Intel® Core™ Ultra and Intel® Core™ Processors technical resources list includes applications notes, datasheets, packing information, product briefs, and more.

www.intel.com

Volume 1, pages 76 and 77

 

[Hulk]

Scores and power consumption taken from the Computerbase review.
i9-12900K @ 4,8 GHz P + 3,7 GHz E-cores -> 27232
i9-12900K @ 4,8 GHz P-Cores -> 19837
You end up with:
i9-12900K 3,7 GHz E-cores -> 7795

Now I will calculate a hypothetical ADL-P 2C(4.8GHz)+8c(3.7GHz).
19837/4 = 4959 points + 7795 points for a total of 12754 points in Cinebench R23. Power consumption was 69W in Corona Benchmark 1.3 without any limit set.

This doesn't look so great compared to ADL-H 6C(3GHz)+8c(2.4GHz) limited to 35W with a score of 14288 points in R23, but that's not the main point.
I wanted to know how It will score at the same clocks as the above ADL-H.

2x P-core(3GHz): 4959/48*30 = 3099 points
8x E-core(2.4GHz): 7795/37*24 = 5056 points
For a total of 8155 points with an unknown package power.

I am sure this can be under 20-25W, but I am not so sure about fitting within 15W, so I will lower the clocks even more to be sure.

2x P-core(2.4GHz): 4959/48*24 = 2480 points
8x E-core(2GHz): 7795/37*20 = 4213 points
For a total of 6693 points.
It's true I lowered the clocks quite a bit, but let's consider It as the worst scenario and In reality It will score anywhere between 6693-8155 points at 15W.

For comparison (Teschspot R23 MT):
R7 5800U manages 7480 points with 15W TDP and 9284 points with 25W TDP.
i7-1165G7 manages 3679 points with 15W TDP and 5189 points with 28W TDP.
i7-1185G7 manages 3500 points with 15W TDP and 5301 points with 28W TDP.

As you can see, against Ryzen It can be slower or faster depending on how high It can clock with a limited power budget, but even at worst It looks pretty competitive.
Against ULV Tiger Lake the MT performance is brutally increased, there is nothing more to say.

Nice analysis. Thanks.
It's going to be interesting to see how Gracemont reacts to lower frequencies, as yet we don't have much data of people "simulating" 2+8 at lower frequencies/power envelopes.

Also I would imagine the best mobile parts will be very low leakage/voltage that will perform a bit better in regards to power than the desktop parts we've seen tested so far.

 

[LightningZ71]

Cpu cooling costs wonbe excessive for anything but the 12900k and maybe the 12700k. The only thing that's overly pricey compared to AMD is the motherboards, and, that is partially justified by it being a more capable platform than the x570.

 

[clemsyn]

I turned on the AI Overclock in my bios and it says 32% overclock. Scores are nice. Finally able to hear my CPU fans running!

CPU-Z
Single thread - 856 (90W wall power draw)
Multi Thread - 7938.5 (260W power draw)

Cinbench R 23
Multi Score - 19031 (268W wall power draw)
Single Score - 2059 (92W wall power draw)

I downgraded my power supply to a 450W bronze (from 850w because of Microcenter associate's advice which is incorrect as they stated this eats a lot of power).

I'll attach photos of the results and HW monitor results of clocks, power draw, voltages and power. Please feel free to analyze it. Any advice is appreciated. Thanks.

 

[Heartbreaker]

12600K is a absolute beast for us 99%ers! Intel desperately needed a win to stay competitive and to keep AMD honest and has one here. Only thing that bothers me is socket 1700 motherboards are $200, double the Ryzen boards - effectively erasing the price/performance lead. That said it, still leads. 5600x will have to go down to $225-$250. So basically you'll be able to get into a 10% slower setup for $350 instead of $520. That might change your mind.

Soon there will be B-Series MBs that will lower the platform cost for lower end chips.

 

[uzzi38]

Also, based on everything that we've seen so far, Golden Cove seems to be on par or better in power efficiency than Zen 3 Core for Core, at the same clock. The inefficiency only stems from the massive all-core frequencies Intel is pushing the the flagship 12900k, especially. Looking forward to some detailed power analysis tomorrow.

Just wondering, what are you using as a basis? Because it seems to me to be the same situation as Willow Cove for the most part, higher performance scaling and power efficiency at high clocks, worse at lower clocks.

As proof, take a look at ComputerBased's 88W testing of the 12900K with the Gracemont's disabled and the 5800X:




The 5800X actually has a ~3% advantage at 88W, which is pretty significant given higher uncore power on AMD's platform thanks to the IOD.

 

[igor_kavinski]

I turned on the AI Overclock in my bios and it says 32% overclock. Scores are nice. Finally able to hear my CPU fans running!

CPU-Z
Single thread - 856 (90W wall power draw)
Multi Thread - 7938.5 (260W power draw)

Cinbench R 23
Multi Score - 19031 (268W wall power draw)
Single Score - 2059 (92W wall power draw)

I downgraded my power supply to a 450W bronze (from 850w because of Microcenter associate's advice which is incorrect as they stated this eats a lot of power).

I'll attach photos of the results and HW monitor results of clocks, power draw, voltages and power. Please feel free to analyze it. Any advice is appreciated. Thanks.

Only two cores seem to be going past 90 degrees. Much better than the 12900k's scary temperatures.

 

[TheELF]

The 5800X actually has a ~3% advantage at 88W, which is pretty significant given higher uncore power on AMD's platform thanks to the IOD.

If the higher uncore is significant then how about the fact that 88W is called the eco mode for a reason, that reason being that that is the most efficient point in the power/performance curve while for the intel part it's just a random spot.
If 3% is the difference between the most optimal setting on the ryzen and a random setting on the intel part then what else do you want?

 

[uzzi38]

If the higher uncore is significant then how about the fact that 88W is called the eco mode for a reason, that reason being that that is the most efficient point in the power/performance curve while for the intel part it's just a random spot.
If 3% is the difference between the most optimal setting on the ryzen and a random setting on the intel part then what else do you want?

I never said Eco mode was less efficient than the ootb settings for the 5800X?

I said the IOD contributes to a significant portion of that power consumption. Regardless of your power limits and how much the memory sub-system is being pushed the 5800X is seeing a pretty consistent ~15W that it can't dedicate to its CPU cores. In theory this should give an advantage to the 12900K in question.

EDIT: Here, this will better illustrate my point there:



Notice how the 3700X in question seems to have this static power draw that makes it seem so much less efficient than Renoir at that <40W range.

 

[remsplease]

(moved to correct forum)

 

[mikk]

The pl2 for the 6+4 core one is 150W so even if you think that max turbo power is the default, that is going to be 150W sustained, the 12400f will be 6core only but I think it will have the same limits. Pl1 of 125 pl2 of 150W

Intel® Core™ Ultra and Intel® Core™ Processors Technical Resources

Intel® Core™ Ultra and Intel® Core™ Processors technical resources list includes applications notes, datasheets, packing information, product briefs, and more.

www.intel.com

Volume 1, pages 76 and 77

i5-12400 is a PL1 65W SKU. With a max MT turbo of 4.0 Ghz and no E-cores there is no chance it can go as high as 150W. Even the i5-12600K doesn't need 150W in normal apps, in Cinebench it uses 115-125W:

https://twitter.com/x/status/1455905136958709760

 

[Heartbreaker]

The 5800X actually has a ~3% advantage at 88W, which is pretty significant given higher uncore power on AMD's platform thanks to the IOD.

I don't think that argument really holds much water.

TPU OC, the 5800x and 5700G both to 4.6GHz all core, and the 5700G used slightly more power.

Clearly the 5700G has no IOD, so if your argument held up, it should use less power...

 

[TheELF]

I never said Eco mode was less efficient than the ootb settings for the 5800X?

I said the IOD contributes to a significant portion of that power consumption. Regardless of your power limits and how much the memory sub-system is being pushed the 5800X is seeing a pretty consistent ~15W that it can't dedicate to its CPU cores. In theory this should give an advantage to the 12900K in question.

EDIT: Here, this will better illustrate my point there:

View attachment 52559

Notice how the 3700X in question seems to have this static power draw that makes it seem so much less efficient than Renoir at that <40W range.

So computerbase measures the power draw in a way that they include the uncore in the numbers?
Or are they just typing in a limit in the mobo? Is that subtracting or adding the uncore load?

i5-12400 is a PL1 65W SKU. With a max MT turbo of 4.0 Ghz and no E-cores there is no chance it can go as high as 150W. Even the i5-12600K doesn't need 150W in normal apps, in Cinebench it uses 115-125W:

https://twitter.com/x/status/1455905136958709760

So even less concern about cooling.

 

[dmens]

Not bad, huh? Intel7 is on par or better than TSMC 7nm, based on all the proper and detailed power consumption analyses trickling in. Called it. ADL-S is not a power hog. It may, in fact, be the most efficient x86 arch right now. Imagine that!
All of a sudden, all the die-hard Intel critics, including a few self-proclaimed chip engineers of ex-Intel stock, are looking like quack psychics right now.
All the 'expert analyses' you spewed on the pages of this forum for years, the condescending attitude, and mispredictions about the hybrid system and scheduling nightmare awaiting ADL-S on windows, you owe your readers an explanation how you could've gotten things so wrong when the writing has been on the wall for so long. Should be an interesting read.

The funny thing about Dunning Kruger is the people on Peak Ignorance are literally incapable of absorbing actual knowledge.

1: Apple M1 core cluster at 30 watts is the highest power it will ever draw, therefore the lowest efficiency it will operate at. 30W for a 12900k is about as low as it will go, and around the highest efficiency it will run at. You just attempted to make an efficiency claim on the most misleading single data point available.

2: Same claim compared an 8 + tiny 2 versus a 6 + 8 on a core scaling workload. By parity of argument, you would have to clock a single Firestorm to the moon to match that MT score and you can claim ADL is more efficient too. Go compare the die area on the two, and feel free to use a generous process scaling number for N5 for the latter.

3: Against AMD Zen 3, the data point here is to pick a single wattage (15W), eliminate the processing elements that are the worst at that power (the Golden Cove cores), to maximize the advantage against Zen3 where ~1.5W a core is not performant operating point. Bravo.

Golf clap, Ryan Shrout should give you a job. We chip engineers are just wasting our time looking at real full range performance/power/workload data curves. Just pick a single data point on a single workload that is absolutely favorable to your product and just run with it.

I suppose this could be an interesting read for you, if you had the intellectual honesty to absorb it, or the technical knowledge to understand a non-linear curve on a 2-axis plot.

Here, have some honesty: I don't like Intel and I think their products are trash. And from where I am sitting, it still is. AMD is winning either perf per area or perf per watt, usually both, with a less dense process and an older design. Every pro-ADL statement is made with isolated data points and without the full picture. Which is fairly typical, since they usually come from dabblers and dilettante fanboys. As I recall, you said I am just disgruntled and insulting my "betters" at Intel. Well... the stuff I work on came out last month. Looks like my "betters" got BTFO on actual product performance. Heh.

 

[uzzi38]

I don't think that argument really holds much water.

TPU OC, the 5800x and 5700G both to 4.6GHz all core, and the 5700G used slightly more power.

Clearly the 5700G has no IOD, so if your argument held up, it should use less power...

Link please? I can see where they overclocked in their reviews of each, I can't see any measurements of power draw under that OC.