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.

 

[xinthius]

Not enough power...

You're going to need to help me here... I haven't played around in a bios for many years now!

 

[TheELF]

I've been out of the PC game for a while.. Is this a good temperature? I'm at around the 60 deg C mark for the 9 minutes on multi core testing, but my Cores #0 - #5 are only running at 4.5GHz, not at the upper limit of 4.9GHz.

My two questions are:

1. Why are my P cores not running at maximum frequency here
2. Are my temperatures good for this test?

4.9Ghz is for one core maybe two at most, your score is the same that can be found in benchmarks so everything is fine.
60 degrees is super for full load.

Unlocking it, or outright overclocking will not improve performance by enough to make it worth dealing with the high power draw and cooling.

 

[Markfw]

You're going to need to help me here... I haven't played around in a bios for many years now!

I don't know exactly the setting, but this will make it take a lot more electricity, and run hotter. Right now you have a fair good and somewhat efficient machine. I would not mess with it.

 

[TheELF]

You're going to need to help me here... I haven't played around in a bios for many years now!

Download intel xtreme tuning utility it will allow you to make changes in real time while having some benchmark running to test how far you can push it.
There are plenty of guides out there if you can't figure it out by yourself.

 

[DrMrLordX]

So your theory is that it uses too much power, and your solution is less efficiency cores and more performance cores?

Of course! Golden Cove actually doesn't use "too much" power if you back the clocks off a bit, while Alder Lake-S suffers from feeding P cores and E cores off the same voltage rail, which causes the E cores to chew up more power than they should at a given performance level.

It's like comparing a 5800x to a 5950x. @142W the 5800X is hard to cool and suffers significantly in the perf/watt department compared to the 5950X in MT workloads where the 5950X sticks to lower clocks.

Intel allows the 12900k in particular to reach clocks of 4.7 GHz and above in MT workloads which makes it suffer like a 5800X, only worse. Intel sort-of has to do this to squeeze out enough performance to win MT benchmarks against the competition. They could add more Gracemont instead and back off clocks on the P cores, but it's not clear that the boost algo they're using has the right behavior to do that.

In any case, mimicking their competition and expanding the number of "big" cores while dropping clocks would produce the desired behavior, while also removing the need for scheduler hyjinx on their flagship CPU.

For equal die size with more P and less E cores, you would only get 10 P cores, and 0 E cores, which would just lose across the board.

10 P cores would be preferable in a lot of cases. Not everything MT scales with core/thread count well enough to utilize 24-32 threads. As for the die size, Alder Lake-S is actually surprisingly small given the package size, so I'm curious as to why they didn't tape out a larger die for the high-end desktop parts.

 

[uzzi38]

So your theory is that it uses too much power, and your solution is less efficiency cores and more performance cores?

IMO, the opposite would be better.

In about the same die size, you could do 6 P cores, and 16 E Cores, would boost MT performance higher, and lower power. For really high thread server loads, they would be better off with 0 P cores and large amount of E cores.

If you went to 16 P cores, it would require a MUCH larger die.

For equal die size with more P and less E cores, you would only get 10 P cores, and 0 E cores, which would just lose across the board.

Eh, it's not a 1:4 ratio, more like a 1:3 ratio. 4 Gracemont clusters are definitely bigger than 4 Golden Cove cores.

But that's just me nitpicking.

 

[guidryp]

As for the die size, Alder Lake-S is actually surprisingly small given the package size, so I'm curious as to why they didn't tape out a larger die for the high-end desktop parts.

As always, more die size costs more money, and in a era of massive chip shortage, bigger dies means fewer of them.

 

[guidryp]

Eh, it's not a 1:4 ratio, more like a 1:3 ratio. 4 Gracemont clusters are definitely bigger than 4 Golden Cove cores.

But that's just me nitpicking.

I didn't measure. I am just going by this:

He says the are between 20% to 25% of the size of P-Core.

 

[DrMrLordX]

As always, more die size costs more money, and in a era of massive chip shortage, bigger dies means fewer of them.

If the yields on 10ESF are as good as 14nm then I don't see that being a huge issue. If the yields aren't as good then it really depends on how bad they are and how many wafers are being chewed up to bring Sapphire Rapids to market.

 

[Accord99]

10 P cores would be preferable in a lot of cases. Not everything MT scales with core/thread count well enough to utilize 24-32 threads. As for the die size, Alder Lake-S is actually surprisingly small given the package size, so I'm curious as to why they didn't tape out a larger die for the high-end desktop parts.

Yeah, the only thing I can think of is the ring bus being used still has a limit of 10 cores (or core clusters) that won't be improved until Raptor Lake. Otherwise, a 10P+8E would have effectively Core 2ed Zen3, sweeping virtually all of the benchmarks with reduced power consumption.

 

[DrMrLordX]

Yeah, the only thing I can think of is the ring bus being used still has a limit of 10 cores (or core clusters) that won't be improved until Raptor Lake. Otherwise, a 10P+8E would have effectively Core 2ed Zen3, sweeping virtually all of the benchmarks with reduced power consumption.

So would a 16P Alder Lake.

Interconnect is a concern. Given the problems Intel has had with the memory controller on Alder Lake (see: posts by @JoeRambo on the subject), I am not confident they could have gone chiplet without suffering massive latency penalties. But I think they would have needed to do so to significantly up core count.

 

[guidryp]

If the yields on 10ESF are as good as 14nm then I don't see that being a huge issue. If the yields aren't as good then it really depends on how bad they are and how many wafers are being chewed up to bring Sapphire Rapids to market.

Intel constantly ran short of CPUs in the 14nm era when silicon shortages weren't as bad as they are now.

If you are looking at building a CPU for massively parallel workloads, (Which 16 P cores implies) then using a larger balance of E-cores makes more sense. They are both more Area efficient, and more Power efficient. This is exactly why they exist.

With more E core bias, you can build a chip that is smaller, uses less power, and delivers more multi-threaded performance than your suggested 16 P-core monster.

I would expect Big Server chips to be exclusively E-Cores.

 

[Hans Gruber]

Intel constantly ran short of CPUs in the 14nm era when silicon shortages weren't as bad as they are now.

If you are looking at building a CPU for massively parallel workloads, (Which 16 P cores implies) then using a larger balance of E-cores makes more sense. They are both more Area efficient, and more Power efficient. This is exactly why they exist.

With more E core bias, you can build a chip that is smaller, uses less power, and delivers more multi-threaded performance than your suggested 16 P-core monster.

I would expect Big Server chips to be exclusively E-Cores.

It's pretty difficult to have a silicon shortage when you are the end producer and supplier of CPU's. But I digress, scarcity like silicon is a manufactured process. It's like belief when faith alone has no value, bearing or reason.

 

[nicalandia]

It's pretty difficult to have a silicon shortage when you are the end producer and supplier of CPU's. But I digress, scarcity like silicon is a manufactured process. It's like belief when faith alone has no value, bearing or reason.

There is an actual Silicon Shortage(of the metal)

Silicon’s 300% surge throws another price shock at the world

The metal made from the second-most abundant element on Earth has become scarce.

www.mining.com

"A metal made from the second-most abundant element on Earth has become scarce, threatening everything from car parts to computer chips and throwing up another hurdle for the world economy. The shortage in silicon metal, sparked by a production cut in China, has sent prices up 300% in less than two months"

 

[Hans Gruber]

There is an actual Silicon Shortage(of the metal)

Silicon’s 300% surge throws another price shock at the world

The metal made from the second-most abundant element on Earth has become scarce.

www.mining.com

"A metal made from the second-most abundant element on Earth has become scarce, threatening everything from car parts to computer chips and throwing up another hurdle for the world economy. The shortage in silicon metal, sparked by a production cut in China, has sent prices up 300% in less than two months"

I just checked newegg. Intel has an adequate supply of alder lake CPU's. If there was a real shortage of Intel made goods. It would be the 10nm chips. You cannot manufacture scarcity when a semiconductor company makes their own products.

 

[UsandThem]

I just checked newegg. Intel has an adequate supply of alder lake CPU's. If there was a real shortage of Intel made goods. It would be the 10nm chips. You cannot manufacture scarcity when a semiconductor company makes their own products.

You're doubling down and sticking with that claim?

So you think Intel is completely immune to all the supply constraints over the last 2 years?

When every retailer was sold out of the 11600k about 45 days ago when I upgraded my son's computer, it was simply because Intel didn't want to sell any?

 

[nicalandia]

You cannot manufacture scarcity when a semiconductor company makes their own products.

They can't make Silicon by themselves, most likely they have a good supply of the metal for a few months

 

[UsandThem]

They can't make Silicon by themselves, most likely they have a good supply of the metal for a few months

You mean it's not made from magical unicorn farts (which Intel apparently also owns)?

 

[Hulk]

4.9Ghz is for one core maybe two at most, your score is the same that can be found in benchmarks so everything is fine.
60 degrees is super for full load.

Unlocking it, or outright overclocking will not improve performance by enough to make it worth dealing with the high power draw and cooling.

This is the correct answer. Intel guaranteed turbo boost frequencies for the 12600K are 4.5GHz for the P's and 3.6GHz for the E's. Depending on the quality of your silicon (and cooling/power supply) you may be able to get more but the increase in power is probably not going to be worth the added performance. Try some higher frequencies and check the package power and get back to us if you have the time.

 

[maddie]

I just checked newegg. Intel has an adequate supply of alder lake CPU's. If there was a real shortage of Intel made goods. It would be the 10nm chips. You cannot manufacture scarcity when a semiconductor company makes their own products.

Epic. From silicon and rare earth mines to finished products inclusive of all intermediate steps.

Intel, I never truly knew you.

 

[xinthius]

This is the correct answer. Intel guaranteed turbo boost frequencies for the 12600K are 4.5GHz for the P's and 3.6GHz for the E's. Depending on the quality of your silicon (and cooling/power supply) you may be able to get more but the increase in power is probably not going to be worth the added performance. Try some higher frequencies and check the package power and get back to us if you have the time.

It wasn't really worth it. I've just scored 17,700 for a marginal increase in core clock by 100MHz all core which took temps from 60 to 66 deg C. Total package power at 132 watts.

I guess I'll just stick at stock on the CPU, tweak my RAM timings and go from there. I game at 1440p anyway, so with my 6700XT I'm GPU limited. A fun exercise nonetheless.

FYI my PSU is a RM650x and my cooler is a Corsair hi100 Elite, so I think I'm fairly well covered there.

 

[nicalandia]

You mean it's not made from magical unicorn farts (which Intel apparently also owns)?

Last I checked Intel was not into mining their own wafer quality Silicon Metal, so China hoarding it will eventually catch up to their supply

 

[Dave3000]

I might upgrade from a Ryzen 5800x to a i7-12700k within the next 2 months. Does the BIOS in the Z690 motherboards have an option to disable the E-cores? I read that I need Windows 11 to have the optimized thread scheduler that distinguishes between the P-Cores and the E-Cores of the 12th gen CPU's that have E-cores but I want to stick with Windows 10 for now and that's why I want to know if I can disable the E-cores in the BIOS for the 12700k.

 

[nicalandia]

I might upgrade from a Ryzen 5800x to a i7-12700k within the next 2 months. Does the BIOS in the Z690 motherboards have an option to disable the E-cores? I read that I need Windows 11 to have the optimized thread scheduler that distinguishes between the P-Cores and the E-Cores of the 12th gen CPU's that have E-cores but I want to stick with Windows 10 for now and that's why I want to know if I can disable the E-cores in the BIOS for the 12700k.

8C/16T Zen3 vs 8C/16T Golden Cove is a side grade at best for gaming if you have a Mid Tier and Up GPU and play at 1440P or better

 

[clemsyn]

It wasn't really worth it. I've just scored 17,700 for a marginal increase in core clock by 100MHz all core which took temps from 60 to 66 deg C. Total package power at 132 watts.

I guess I'll just stick at stock on the CPU, tweak my RAM timings and go from there. I game at 1440p anyway, so with my 6700XT I'm GPU limited. A fun exercise nonetheless.

FYI my PSU is a RM650x and my cooler is a Corsair hi100 Elite, so I think I'm fairly well covered there.

This is what I basically decided. Left everything in stock speeds then undervolted it.