Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[Exist50]

I know personally that the more the CPU is used, and for longer, Alder lake loses, but nobody here wants to hear my evidence. This is a raptor lake thread after all. I just suspect that Raptor lake will take after Alder lake in its general behavior. And I am personally not a fan of E-cores.

Then why not post this "evidence" instead of absurd claims contradicted by every available piece of data?

 

[Markfw]

Then why not post this "evidence" instead of absurd claims contradicted by every available piece of data?

I already did, but you ignored it all, and basically called the entire DC forum liars.

Back to Raptor Lake.....

 

[Exist50]

I already did, but you ignored it all, and basically called the entire DC forum liars.

Back to Raptor Lake.....

Dead links that, even if they worked, wouldn't include everything you claimed, such as a CPU consuming >150% it's PL2 supposedly without modification. Yeah, totally on me...

As for Raptor Lake, it seems some of the early optimism is starting to be tempered by the reality of benchmarks. It's an Alder Lake refresh and will perform like one. I wouldn't even be surprised if big core efficiency regresses at its peak. They didn't promise more L2 for mobile for a reason.

 

[nicalandia]

I know personally that the more the CPU is used, and for longer, Alder lake loses, but nobody here wants to hear my evidence.

Then why not post this "evidence" instead of absurd claims contradicted by every available piece of data?

I thought it was a known Fact that on many MT apps where Intel matches the 5950X(or exceed it by a few %) it ends up losing that lead the longer the work it takes? For example CineBench R23.

 

[dullard]

I thought it was a known Fact that on many MT apps where Intel matches the 5950X(or exceed it by a few %) it ends up losing that lead the longer the work it takes? For example CineBench R23.

This issue often comes down to silly tribal arguments. The Intel side wants to claim that Alder Lake is fast. The AMD side want to claim that Alder Lake consumes high power. The reality is neither tribal argument is true in long runs--they are only true in short benchmark tests. Alder Lake is fast for a short period of time and it is high power for a short period of time. The benchmark sites that want to get out a review at the start of launch day generally only test very short benchmarks.

What really matters to DC users and many business users is not short-term benchmarks, but how the chip will perform during a grueling work day or even 24/7 calculations. In that case, Alder Lake isn't as fast as benchmarks show but it is also not high power either. Both sides' main arguments are not only incorrect but also are generally based on assumptions and hopes. There is very little data out there for real-world usage. Benchmark sites have moved on after they got their launch article out and generally have no interest in doing long-term through reviews.

At least @Markfw was willing to buy and test an Alder Lake chip. Probably the worst Alder Lake chip for his purposes (4 E cores), but he did test it and I do trust his performance test results that he posted. However, as far as I can remember (and forgive me if I missed a post otherwise) Markfw only posted peak short-term power with long-term performance numbers. That is not a helpful combination nor a sincere attempt to determine the true value of importance for his use (long term performance) / (long-term average power). I really want to know how his chip performs in long runs with long-term average power (not peak power). Then we can make informed decisions.

 

[Markfw]

This issue often comes down to silly tribal arguments. The Intel side wants to claim that Alder Lake is fast. The AMD side want to claim that Alder Lake consumes high power. The reality is neither tribal argument is true in long runs--they are only true in short benchmark tests. Alder Lake is fast for a short period of time and it is high power for a short period of time. The benchmark sites that want to get out a review at the start of launch day generally only test very short benchmarks.

What really matters to DC users and many business users is not short-term benchmarks, but how the chip will perform during a grueling work day or even 24/7 calculations. In that case, Alder Lake isn't as fast as benchmarks show but it is also not high power either. Both sides' main arguments are not only incorrect but also are generally based on assumptions and hopes. There is very little data out there for real-world usage. Benchmark sites have moved on after they got their launch article out and generally have no interest in doing long-term through reviews.

At least @Markfw was willing to buy and test an Alder Lake chip. Probably the worst Alder Lake chip for his purposes (4 E cores), but he did test it and I do trust his performance test results that he posted. However, as far as I can remember (and forgive me if I missed a post otherwise) Markfw only posted peak short-term power with long-term performance numbers. That is not a helpful combination nor a sincere attempt to determine the true value of importance for his use (long term performance) / (long-term average power). I really want to know how his chip performs in long runs with long-term average power (not peak power). Then we can make informed decisions.

All number I quoted were for primegrid, and they are all loaded indefinitely to 100%. When I disabled the e-cores, a few times, AVX512 was used, and it was for at least 10 minutes and broke the PL2 limit of 180 watt for 8 cores. It was at 300 on the kill-a-watt with NO video card use. I suspect this is why it was disabled in later updates. I have a huge heatsink on mine(at least 250 watt), so it did not throttle. I don't remember all the other power numbers, but it certainly exceeded 142 watt for 8 cores, I think then it was 180. I don't remember the speed. Not doing primegrid right now, so I can't doublecheck. 5950x runs at 3.7 when 100% loaded indefinitely.

 

[Exist50]

I don't remember all the other power numbers, but it certainly exceeded 142 watt for 8 cores, I think then it was 180.

You've been consistently claiming your 12700f consumes 300W, and even directly comparing it to AMD's PPT or even TDP values. Now "whoops" maybe it was actually 180W...

Not that it makes your numbers line up. It's very widely accepted that peak to peak, Golden Cove is stronger than Zen 3. You're not only claiming that to be false, but your claiming that Zen 3 wins with the same dual channel memory (if not weaker), <1/2 the power per core, and a large ISA disadvantage from no AVX-512. It would be just as realistic to claim Bulldozer beats Skylake, or Cascade Lake beats Milan.

 

[TheELF]

This issue often comes down to silly tribal arguments. The Intel side wants to claim that Alder Lake is fast. The AMD side want to claim that Alder Lake consumes high power. The reality is neither tribal argument is true in long runs--they are only true in short benchmark tests. Alder Lake is fast for a short period of time and it is high power for a short period of time. The benchmark sites that want to get out a review at the start of launch day generally only test very short benchmarks.

This can happen if you use the default behavior that intel suggests, that no mobo that ever was reviewed has as default, usually all mobos come with PL1=PL2= infinity and beyond (the same 4000+ W that also some ryzen mobos come with as default)

This is why I keep uploading pics from this site, this is locked TDP at whatever they state, it does not go down after some time but stays at that point forever and will keep this performance for ever and as anyone can easily see it takes 160W to beat the 5950x in performance* which is more than the ~120W that the 5950x needs (maybe 142W if measured at the mobo) but for beating a CPU that has 30% more threads it's still very good.

*in this one bench but that's all we have. All their other benches are at 125W or at 241W

Messungen: Leistungsaufnahme, Temperatur und Energieeffizienz - Seite 9 - Hardwareluxx

Core i9-12900K und Core i5-12600K: Hybrid-Desktop-CPUs Alder Lake im Test.

www.hardwareluxx.de

 

[uzzi38]

This can happen if you use the default behavior that intel suggests, that no mobo that ever was reviewed has as default, usually all mobos come with PL1=PL2= infinity and beyond (the same 4000+ W that also some ryzen mobos come with as default)

This is why I keep uploading pics from this site, this is locked TDP at whatever they state, it does not go down after some time but stays at that point forever and will keep this performance for ever and as anyone can easily see it takes 160W to beat the 5950x in performance* which is more than the ~120W that the 5950x needs (maybe 142W if measured at the mobo) but for beating a CPU that has 30% more threads it's still very good.

*in this one bench but that's all we have. All their other benches are at 125W or at 241W

Messungen: Leistungsaufnahme, Temperatur und Energieeffizienz - Seite 9 - Hardwareluxx

Core i9-12900K und Core i5-12600K: Hybrid-Desktop-CPUs Alder Lake im Test.

www.hardwareluxx.de

ComputerBase have more testing that is also relevant as well: https://www.computerbase.de/2021-11/intel-core-i9-12900k-i7-12700k-i5-12600k-test/2/

 

[Markfw]

You've been consistently claiming your 12700f consumes 300W, and even directly comparing it to AMD's PPT or even TDP values. Now "whoops" maybe it was actually 180W...

Not that it makes your numbers line up. It's very widely accepted that peak to peak, Golden Cove is stronger than Zen 3. You're not only claiming that to be false, but your claiming that Zen 3 wins with the same dual channel memory (if not weaker), <1/2 the power per core, and a large ISA disadvantage from no AVX-512. It would be just as realistic to claim Bulldozer beats Skylake, or Cascade Lake beats Milan.

Boy you have a problem reading... The 300 only applies to avx-512. The 180 is just 8 cores, compared to 142 watt of 16 cores.

 

[dullard]

This can happen if you use the default behavior that intel suggests, that no mobo that ever was reviewed has as default

Or it happens if you use the Intel cooler or any other cooler designed around the chip TDP. To get it continuously at higher power levels, you need to purposely put in a cooling system for those higher levels. That is: you choose to run at that higher power.

I agree with you that there are few relevant reviews. Most put on the beefiest cooling system they have, find the peak performance, but also find peak power can be sustained. But many business users don't do that.

 

[igor_kavinski]

Where cost is concerned, 5950X may be the better performer in DC but if the ADL chip is paired with really fast low latency DDR5, it might be able to put up a really good fight. Whose to say that the ADL chip isn't simply starved for bandwidth with DDR4 and spending most of its time idling and waiting for data to arrive?

The face-off between Zen 4 and RPL will be a more fair fight as both will be benchmarked (hopefully) using identical RAM, thus levelling the playing field. It will then be more of a contest between the architectures and IMCs of the two contenders.

Power-wise, AMD is still expected to have the upper hand, at least until Intel gets their manufacturing act together.

 

[Exist50]

The 300 only applies to avx-512

Again, ISA does not change power limits, and furthermore, you're changing your story from comment to comment. First it was power from the CPU, then power from the wall, and now it's power from the CPU again? And that's without even considering your performance claims...

I think it's very clear you're just trying to mislead people.

 

[Markfw]

Again, ISA does not change power limits, and furthermore, you're changing your story from comment to comment. First it was power from the CPU, then power from the wall, and now it's power from the CPU again? And that's without even considering your performance claims...

I think it's very clear you're just trying to mislead people.

Every post is the same from me. You are the one with reading comprehension problems, as well as being a troll.

 

[Exist50]

Every post is the same from me. You are the one with reading comprehension problems, as well as being a troll.

You claimed the CPU was taking 300W. Then you claimed Kill A Watt was saying 300W. Then you went back to insisting the CPU takes 300W, and comparing to AMD's PPT. So either you have no idea what you're even measuring, or you're trying to mislead people.

 

[Markfw]

You claimed the CPU was taking 300W. Then you claimed Kill A Watt was saying 300W. Then you went back to insisting the CPU takes 300W, and comparing to AMD's PPT. So either you have no idea what you're even measuring, or you're trying to mislead people.

When you have 2 sticks of ram and an NVME drive, the difference is very small. Everybody knows what I mean. The system does NOT take 120 watts.

 

[Exist50]

When you have 2 sticks of ram and an NVME drive, the difference is very small. Everybody knows what I mean. The system does NOT take 120 watts.

RAM, motherboard, NVMe, power supply losses, etc. They add up. More importantly, you not only have insisted that it's just CPU power, but clearly don't use that same methodology for the Ryzen system, so why? And then there're your performance claims that somehow no one can reproduce...

 

[TheELF]

Or it happens if you use the Intel cooler or any other cooler designed around the chip TDP. To get it continuously at higher power levels, you need to purposely put in a cooling system for those higher levels. That is: you choose to run at that higher power.

I agree with you that there are few relevant reviews. Most put on the beefiest cooling system they have, find the peak performance, but also find peak power can be sustained. But many business users don't do that.

Well then the question is which one do you consider the CPUs TDP? PL1 or PL2? What you describe is more like a mobo maker pushing the max and you using a 125W TDP cooler, yes in that case throttling is going to be rampart when running heavy stuff, when using unlimited power you need the biggest cooler you can get.

If you use PL1=125 and Pl2=241 and proper TAU with a 125W cooler it will work at least at the 125W level and will boost whenever possible, as you can see on hardwareluxx or computerbase even at locked 125W the 12900k is at least on par with the 5900x ,so not bad.

If you use PL1=PL2=241W then you need a 240W cooler to keep max performance going when running heavy workloads.

And if you think that the TDP is power limits lifted (default of most mobos) then that's overclocking that can reach 300W and you would need an appropriate cooler.

 

[dullard]

Well then the question is which one do you consider the CPUs TDP? PL1 or PL2? What you describe is more like a mobo maker pushing the max and you using a 125W TDP cooler, yes in that case throttling is going to be rampart when running heavy stuff, when using unlimited power you need the biggest cooler you can get.

If you use PL1=125 and Pl2=241 and proper TAU with a 125W cooler it will work at least at the 125W level and will boost whenever possible, as you can see on hardwareluxx or computerbase even at locked 125W the 12900k is at least on par with the 5900x ,so not bad.

If you use PL1=PL2=241W then you need a 240W cooler to keep max performance going when running heavy workloads.

And if you think that the TDP is power limits lifted (default of most mobos) then that's overclocking that can reach 300W and you would need an appropriate cooler.

My point is that the customer has the ultimate control--regardless of what the motherboard makers do. The reviews are geared towards enthusiasts that push their computers to the limits. For a business user who needs to compile big software, model air flow around a fan blade, or do heavy database crunching the reviews we get of chips tend to be pretty lackluster. That is since those users generally aren't building their own computer and putting in the biggest cooling system they can find. Sure, some are, but the vast majority are not.

For example, if you buy a typical run of the mill computer (such as a Dell Inspiron) with a 12700 processor, and a maximum 300 W power supply, then I can guarantee that the CPU isn't drawing 300 W long term. Same goes with the Dell Optiplex with a 12700 CPU and a maximum 240 W or 260 W power supply--the CPU still isn't pulling 300 W. Go up to the XPS line and you get a 460 W power supply but a NVIDIA® GeForce RTX™ 3060 that can use 170 W of it. You still aren't going to draw 300 W continuously with your CPU. You'd have to go into the Precision workstations to really see CPU draws at that level (and that assumes Dell puts in a cooling system capable of handling it--I don't know personally what is being used).

I personally go with TDP of PL2 is the design power suggestion. For the 12700F being discussed above, that is 180 W. But, that is just a suggestion, you will see computers built with more or less than that. Any poster getting 300 W from that CPU with a PL2 of 180 W did that to themselves through motherboard settings and cooler selection.

 

[TheELF]

My point is that the customer has the ultimate control--regardless of what the motherboard makers do. The reviews are geared towards enthusiasts that push their computers to the limits. For a business user who needs to compile big software, model air flow around a fan blade, or do heavy database crunching the reviews we get of chips tend to be pretty lackluster. That is since those users generally aren't building their own computer and putting in the biggest cooling system they can find. Sure, some are, but the vast majority are not.

For example, if you buy a typical run of the mill computer (such as a Dell Inspiron) with a 12700 processor, and a maximum 300 W power supply, then I can guarantee that the CPU isn't drawing 300 W long term. Same goes with the Dell Optiplex with a 12700 CPU and a maximum 240 W or 260 W power supply--the CPU still isn't pulling 300 W. Go up to the XPS line and you get a 460 W power supply but a NVIDIA® GeForce RTX™ 3060 that can use 170 W of it. You still aren't going to draw 300 W continuously with your CPU. You'd have to go into the Precision workstations to really see CPU draws at that level (and that assumes Dell puts in a cooling system capable of handling it--I don't know personally what is being used).

I personally go with TDP of PL2 is the design power suggestion. For the 12700F being discussed above, that is 180 W. But, that is just a suggestion, you will see computers built with more or less than that. Any poster getting 300 W from that CPU with a PL2 of 180 W did that to themselves through motherboard settings and cooler selection.

Cooling is only an issue when the settings are "unlimited power" + "all auto turbos engaged" ,the 12900k uses ABT that will always try to reach 100° yes in this case selecting the cooling will be selecting your performance.

If running at 125W, which is between the TDP of the 5900x and the 5950x the 12900k is running at least 10 degrees cooler so it needs less cooling than ryzen.
At 241W it does hit 100° but as already shown so many times you don't need to use that much power, ever.
Sadly again we have no other measurements because everybody is only making click bait reviews.

 

[DrMrLordX]

(the same 4000+ W that also some ryzen mobos come with as default)

Even with power reporting deviation hacks, that's not true. At all.

 

[pakotlar]

Cooling is only an issue when the settings are "unlimited power" + "all auto turbos engaged" ,the 12900k uses ABT that will always try to reach 100° yes in this case selecting the cooling will be selecting your performance.

If running at 125W, which is between the TDP of the 5900x and the 5950x the 12900k is running at least 10 degrees cooler so it needs less cooling than ryzen.
At 241W it does hit 100° but as already shown so many times you don't need to use that much power, ever.
Sadly again we have no other measurements because everybody is only making click bait reviews.

Yeah Alder Lake power limits were juiced to make it faster than Zen 3, which was possible only because Alder Lake scales so well with power (much better than Zen 2 and probably 3, given my experience with PBO on my 5950x): https://chipsandcheese.com/2022/01/28/alder-lakes-power-efficiency-a-complicated-picture/

At intermidate power levels it trades blows with Zen 3 in efficiency, depending on workload. Zen 3 is probably more efficient, but not 2x as much, on average.

 

[Mopetar]

Alder Lake doesn't scale well with power. The efficiency goes in the dumpster really quickly. It's just that Alder Lake can handle that much power and still gain clock speed whereas Zen 3 (and all of the earlier incarnations) pretty much hit a brick wall when pushed beyond stock limits.

If you drop the voltage and power limits on Alder Lake it's as efficient as Zen 3 in most cases, but the problem is the core count. AMD looks to be increasing their power limits so I would imagine that Raptor Lake will compare much more favorably at ~120W as well.

Also, why wouldn't they release an 8+16 part? The process should be good enough to support those chips and it doesn't even need to be a top bin gaming CPU. Just sell it as a workhorse CPU that will benchmark more favorably on performance per power and overall throughput. People will pay for such a CPU.

 

[pakotlar]

Alder Lake doesn't scale well with power. The efficiency goes in the dumpster really quickly. It's just that Alder Lake can handle that much power and still gain clock speed whereas Zen 3 (and all of the earlier incarnations) pretty much hit a brick wall when pushed beyond stock limits.

If you drop the voltage and power limits on Alder Lake it's as efficient as Zen 3 in most cases, but the problem is the core count. AMD looks to be increasing their power limits so I would imagine that Raptor Lake will compare much more favorably at ~120W as well.

Also, why wouldn't they release an 8+16 part? The process should be good enough to support those chips and it doesn't even need to be a top bin gaming CPU. Just sell it as a workhorse CPU that will benchmark more favorably on performance per power and overall throughput. People will pay for such a CPU.

It does, compared to Zen 2, in terms of performance, which is clearly what I was talking about. Read the Chips and Cheese analysis I linked.


“””
Overall, Zen 2 is more power efficient at mid to low power levels. At higher power, Golden Cove is more efficient in vector workloads because Zen 2 doesn’t scale well as power increases. Compared to Gracemont, Zen 2 cores are better at all power levels, except when desktop Zen 2 cores hit an early voltage floor.
“””

edit: I’ll add that scaling is also workload dependent, as you can see above. But in general Alder Lake responds well to voltage, and probably somewhat better than Zen 2 and maybe Zen 3, which is what allows Intel to juice it and get a useful performance benefit vs Zen 3.

 

[igor_kavinski]

People will pay for such a CPU.

There are dumb gamer girls out there paying top dollar for the 12900K. Why bother when your power turd sells this well against the competition? They are not going to release a bigger chip and deal with the associated headaches (more power, less wafers per die, possibly more defects etc.) when the current die is working so well for them.