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.

 

[pakotlar]

Based on the previous info with 5.7Ghz 8c/16 @ 300W we can be pretty sure that adding 16 E cores to the mix will put the system well past 300W even after accounting for the lower 5.4Ghz OC. Something like 360W+ makes more sense.

This is ridiculous. The PL2 is known, and the socket isn’t changing. They’re not adding 100W average sustained power draw. PL4 also cannot change significantly upwards without requiring board partners to recertify their 600 series motherboards.

 

[dullard]

It's only useful in mobile, very high diminishing returns past 50 Amps, zero benefit past 70A ( according to patent docs IRRC)

While there is no evidence yet that this will be used in desktop chips, listing just those numbers can be misleading.

The patent mentions two voltage regulators, VR1 and VR2. VR1 covers the bulk of the power needs and VR2 covers extra power needs such as transient excursions above normal usage (i.e. turbo power spikes). Then the patent gives an example if VR1 supplies 40 A of power then there would be diminishing returns above 50 A and nothing above 70 A. However, there is no reason at all that VR1 is limited to 40 A of power. Those numbers were just an example.

Start at paragraph [0041] for the example explained: https://patentimages.storage.googleapis.com/e3/f8/26/4646a66f041c8b/US20210208656A1.pdf

The only real requirement is that the motherboard provides a secondary voltage from the secondary voltage regulator. If that is possible in LGA1700 is beyond my knowledge.

 

[coercitiv]

This is ridiculous. The PL2 is known, and the socket isn’t changing. They’re not adding 100W average sustained power draw. PL4 also cannot change significantly upwards without requiring board partners to recertify their 600 series motherboards.

It's not ridiculous once you understand we're talking about OVERCLOCKING.

 

[Carfax83]

It's only useful in mobile, very high diminishing returns past 50 Amps, zero benefit past 70A ( according to patent docs IRRC)

From looking at the various articles and reddit threads, it seemed to me the DLVR feature is supposed to put a clamp on voltage droops during high CPU load, which results in voltage overcompensation from the motherboards loadline calibration to maintain stability. This is similar to what an experienced overclocker does when he tunes the voltage.

My 6900K at 4.3ghz on automatic voltage will use over 1.4v, but after tuning I have gotten it down to 1.28v under load with rock hard stability, so obviously the motherboard is definitely overcompensating the voltage to achieve stability at high clock speeds.

If anything, I would think that DLVR would make the most sense and have the largest impact on desktop CPUs due to the much higher frequencies and power loads.

 

[dullard]

If anything, I would think that DLVR would make the most sense and have the largest impact on desktop CPUs due to the much higher frequencies and power loads.

From the best that I can tell, the mobile-only argument started from an image in this link from VideoCardz (which is self-credited as being from VideoCardz). That image has three rows of improvements for mobile, three rows for desktop, and didn't put DLVR in the desktop rows. https://videocardz.com/newz/intel-raptor-lake-desktop-and-mobile-series-appear-on-a-leaked-roadmap

Of course that assumes the image is legit and assumes that there can only be three possible improvements with any CPU generation.

 

[Carfax83]

From the best that I can tell, the mobile-only argument started from an image in this link from VideoCardz (which is self-credited as being from VideoCardz). That image has three rows of improvements for mobile, three rows for desktop, and didn't put DLVR in the desktop rows.

Well if that is indeed how that argument (more like a rumor if you ask me) started, it seems to be based purely on an assumption. Raptor Lake as a platform has shared commonalities between mobile, desktop and even server variants because they all use the same basic core microarchitecture with a few alterations.

That's been the common trend in Intel CPUs for a long time now, and I doubt it's going to change anytime soon.

So if the mobile variant can use DLVR, there's no reason why the desktop variant can't. And as I mentioned above, that feature is more sensible on the desktop anyway due to much greater power loads.

 

[nicalandia]

and you believe those are worth 3x higher price? who cares about paying for more performance, when you can just pay for ability to plug more RAM or disks in, right?

Not just him. But the entire segment(Ultra High End Workstation) is based on More RAM, More PCI lines more Features. The CPU is just an entry point on that segment.

 

[mikk]

Well if that is indeed how that argument (more like a rumor if you ask me) started, it seems to be based purely on an assumption. Raptor Lake as a platform has shared commonalities between mobile, desktop and even server variants because they all use the same basic core microarchitecture with a few alterations.

That's been the common trend in Intel CPUs for a long time now, and I doubt it's going to change anytime soon.

So if the mobile variant can use DLVR, there's no reason why the desktop variant can't. And as I mentioned above, that feature is more sensible on the desktop anyway due to much greater power loads.

If it's somehow platform depending it can be different. Remember FIVR, motherboard VRs were on the board and with the introduction of FIVR it moved into the CPU. There is a downside, it makes the old platform incompatible. The current ADL-S platform would have been dead after just one generation. On mobile such platform change is not a problem.

 

[coercitiv]

From looking at the various articles and reddit threads, it seemed to me the DLVR feature is supposed to put a clamp on voltage droops during high CPU load

My understanding from looking over the patent is that this technology helps with boosting at high clocks while the CPU is not nearly close to TDP/thermal limits. When load is significant enough to push the TDP limit, clocks drop towards base clocks, voltage drops drastically from something like 1.35V to 1.15V or lower, and DLVR ceases to deliver the juicy gains. (voltage values are pure examples, although they do fit the graph bellow and also fit Alder Lake V/f curve for 4-5+Ghz)

Therefore DLVR seems perfectly suited in mobile environments where it can save energy for relatively low threaded burst loads, but it cannot help much when flagship CPUs with lots of cores engage in MT loads @ lower voltage. If DLVR were able to help in such a case, we wouldn't get this type of V/I graph:



If I got the wrong impression then I'd sure love an accurate explanation from someone with electrical engineering background.

 

[pakotlar]

It's not ridiculous once you understand we're talking about OVERCLOCKING.

I can make my 5950X draw 400W. What’s your point?

 

[coercitiv]

I can make my 5950X draw 400W. What’s your point?

I was talking about overclocking Raptor Lake, which is not subject to power limits. So your intervention was pointless.

 

[Carfax83]

If it's somehow platform depending it can be different. Remember FIVR, motherboard VRs were on the board and with the introduction of FIVR it moved into the CPU. There is a downside, it makes the old platform incompatible. The current ADL-S platform would have been dead after just one generation. On mobile such platform change is not a problem.

Yes but we already know Raptor Lake will be backwards compatible with Z600 motherboards and the power values between Raptor Lake and Alder Lake are going to be similar in the retail parts. If Raptor Lake S supports DLVR on desktop, it's likely that it will need the Z700 series chipsets to utilize DLVR.

So DLVR may just be a nice bonus for those buying Intel's latest and greatest chipsets. Or perhaps just a BIOS update will activate the feature on Z600 motherboards, who knows?

 

[Carfax83]

View attachment 65253

If I got the wrong impression then I'd sure love an accurate explanation from someone with electrical engineering background.

I agree, we need someone with an E.E background to properly explain the patent. This post on another forum from a guy that seemed to me like he knew what he was talking about and might possibly have an E.E background meshes well with other articles and comments I've seen on the web regarding DLVR.

Thread 'Intel Raptor Lake CPUs To Feature Digital Linear Voltage Regulator ‘DLVR’ – Could Help Reduce Power Consumption By 25%' https://forums.hardwarezone.com.sg/...-help-reduce-power-consumption-by-25.6641882/

 

[IntelUser2000]

I talked about it here: https://forums.anandtech.com/thread...ure-lakes-rapids-thread.2509080/post-40772218

DLVR is not the focus. I hate how the press doesn't even read the content and just post whatever they feel like.

(If there is a part of society that needs to go off the face of the earth, it's the press. I am not talking just about mainstream media. I mean same with tech channels and youtube channels, all of them. GONE)

The secondary VR is the point. As I said in my conclusion, the second regulator has to be off for the most efficient point, because the secondary regulator reduces efficiency. With higher loads it has to be engaged more and that's why the drop happens(among other reasons).

The secondary regulator is not much more than a guarantee that the CPU can work at lower voltages without causing stability issues. If say 50% is the cutoff point where the second regulator has to be active, then you' get less and less gains as you go above that.

And typically overclocking and adjusting BIOS messes with power management features. I can't imagine it working well in that case at all.

Also yes in heavy MT loads you lose gains. That's why it's a mobile feature and likely benefit low load and burst scenarios.

 

[IntelUser2000]

Imagine

Voltage = Time
Voltage regulators = You and the assistant.

In the first case you are dealing with a person that doesn't say exactly the time they want to meet and the time range can be anywhere from a min to 3 hours before the meeting. So you have to basically wait 3 hours every time to be in touch with him. And you can't miss the meeting.

In the second case you have an assistant that goes to the guy and nags him to tell him the exact time of the meet to report it back to you. So in the second case you know when to leave without waiting 3 hours.

Of course the assistant is dependent on you and requires more of your time the more you need him. At some point he requires so much of your time you might as just wait 3 hours for the guy instead and not having to pay for the assistant in the first place.

Edit: Changed Power = Time to Voltage = Time to be accurate

 

[Carfax83]

And typically overclocking and adjusting BIOS messes with power management features. I can't imagine it working well in that case at all

I think the reason why I thought that DLVR would benefit desktop CPUs the most is because undervolting works so well in Alder Lake and you can get huge savings in power and temp without sacrificing performance.

 

[IntelUser2000]

I think the reason why I thought that DLVR would benefit desktop CPUs the most is because undervolting works so well in Alder Lake and you can get huge savings in power and temp without sacrificing performance.

Alderlake has issues regarding power. The mobile chips idle at very high power and lost several years of advancements.

So if Raptorlake fixes this we'll see improvements but DLVR itself might be a minor improvement for desktops because the baseline might be better. I wouldn't hold my breath for it on the desktop. You also need a new motherboard since it's something that has to be added no matter how trivial. Not sure how many will buy into a feature for a CPU that'll have 250W continuous PL2.

All I see for easy perf/watt improvement is an 100-200MHz improvement for a stepping that'll be instead used to add 8 E cores at no power cost.

It doesn't seem like that's what they are doing according to leaks with rumors pushing 5.4/4.1. I guess it'll be faster than expected. Or have a lower power 13900KT chip that scales back clocks slightly to Alderlake levels and using E cores to boost MT performance greatly.

 

[eek2121]

Alderlake has issues regarding power. The mobile chips idle at very high power and lost several years of advancements.

So if Raptorlake fixes this we'll see improvements but DLVR itself might be a minor improvement for desktops because the baseline might be better. I wouldn't hold my breath for it on the desktop. You also need a new motherboard since it's something that has to be added no matter how trivial. Not sure how many will buy into a feature for a CPU that'll have 250W continuous PL2.

All I see for easy perf/watt improvement is an 100-200MHz improvement for a stepping that'll be instead used to add 8 E cores at no power cost.

It doesn't seem like that's what they are doing according to leaks with rumors pushing 5.4/4.1. I guess it'll be faster than expected. Or have a lower power 13900KT chip that scales back clocks slightly to Alderlake levels and using E cores to boost MT performance greatly.

Could always adjust the PL2 downward to 125W.

It will be interesting to see THOSE benchmarks. The 13900k @ 125W should be able to edge out the 5950X in theory.

 

[Saylick]

I talked about it here: https://forums.anandtech.com/thread...ure-lakes-rapids-thread.2509080/post-40772218

DLVR is not the focus. I hate how the press doesn't even read the content and just post whatever they feel like.

(If there is a part of society that needs to go off the face of the earth, it's the press. I am not talking just about mainstream media. I mean same with tech channels and youtube channels, all of them. GONE)

The secondary VR is the point. As I said in my conclusion, the second regulator has to be off for the most efficient point, because the secondary regulator reduces efficiency. With higher loads it has to be engaged more and that's why the drop happens(among other reasons).

The secondary regulator is not much more than a guarantee that the CPU can work at lower voltages without causing stability issues. If say 50% is the cutoff point where the second regulator has to be active, then you' get less and less gains as you go above that.

And typically overclocking and adjusting BIOS messes with power management features. I can't imagine it working well in that case at all.

Also yes in heavy MT loads you lose gains. That's why it's a mobile feature and likely benefit low load and burst scenarios.

Sooooo, if I'm understanding this correctly, this is like VTEC for CPUs? Default voltage regulator suited for high voltage operations (i.e. high RPM cams, which are bad for idle) with a secondary VR for low voltage operation (i.e. low RPM cam).

 

[dullard]

Sooooo, if I'm understanding this correctly, this is like VTEC for CPUs? Default voltage regulator suited for high voltage operations (i.e. high RPM cams, which are bad for idle) with a secondary VR for low voltage operation (i.e. low RPM cam).

Basically, but swap the "default" and "secondary" terms in your post. The default primary VR is for lower voltage for most tasks (low RPM cam). Then a secondary VR kicks in for higher voltage when needed (high RPM cams). Basically, most of the time that the CPU is running at idle or light to medium loads it will use less power. But when the power is needed, it will be there. If you run your CPU to the limit 24/7 this does nothing for you. If you look only at the maximum peak power, this does nothing for you. But, if you have a varied load, this will lower the total amount of power used during the day.

A problem for Intel is that most reviews only consider peak power and not total amount of power used to complete their test suite in their reviews. So, unless someone goes out of their way to review it, no one will know it helps.

 

[Saylick]

A problem for Intel is that most reviews only consider peak power and not total amount of power used to complete their test suite in their reviews. So, unless someone goes out of their way to review it, no one will know it helps.

It will help in mobile applications, right? And so if really does help, it would show up on battery life comparisons where they test more varied workloads.

 

[dullard]

It will help in mobile applications, right? And so if really does help, it would show up on battery life comparisons where they test more varied workloads.

Yes, it will show up in mobile battery life tests. But, it won't show up in reviews if Intel puts it into any desktop device.

 

[coercitiv]

https://twitter.com/x/status/1554756170568458240

i7 13700K OC : P 5.6, E 4.5
360 AIO, ~1.4V, ~280W

So the good news is the old OC leak for 5.7Ghz @ 300W was not ideal to say the least, now we're down to 280W for 8P + 8E with moderately high clocks on both types of cores. Temps aren't great though, but then again it seems ambient was 29C, definitely not ideal.

 

[Carfax83]

If I end up getting a Raptor Lake system, I'm going to limit the clock speed on the performance cores to 5ghz and the efficiency cores to 4ghz and undervolt the heck out of it. Unless you game at sub 1080p with the settings on low, you're always going to be GPU bottlenecked anyway.

I really wish Intel would be more sensible about targeting these astronomical clock speeds. It's impressive that the chips can hit those speeds at such low voltages, but that wattage!

 

[szrpx]

https://twitter.com/x/status/1554756170568458240

i7 13700K OC : P 5.6, E 4.5
360 AIO, ~1.4V, ~280W

So the good news is the old OC leak for 5.7Ghz @ 300W was not ideal to say the least, now we're down to 280W for 8P + 8E with moderately high clocks on both types of cores. Temps aren't great though, but then again it seems ambient was 29C, definitely not ideal.

That's with Hyperthreading turned off, so 16 threads total. Maybe it's more efficient MT perf/w? Not sure, we'll have to see tests when it comes out I guess.