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.

 

[dullard]

It’s not V^4. Read my edited post.

Still isn't P∝V^2 either. Here is a graph of Ryzen power vs frequency. A quadratic fit is close, but it errs on both ends. Quadratic formulas over estimate power on the low frequency end, and underestimate power on the high frequency end. I wish I had more data to show the extremes. The industry assumption is to use P∝V^3 until data shows otherwise in a specific region of interest.


Cubic fits don't have the power errors on those two ends:

Dynamic Power Consumption - an overview | ScienceDirect Topics

www.sciencedirect.com

Thus dynamic power varies as the cube of the maximum frequency. Static power consumption is nominally independent of frequency but is dependent on voltage. The relation is more complex than for dynamic power, but, for sake of argument, assume it varies cubically with voltage. Since the necessary voltage is proportional to the maximum frequency, the static power consumption varies as the cube of the maximum frequency, too. Under this assumption we can use a simple overall model where the total power consumption varies by the cube of the frequency.

 

[dullard]

I am completely confused by your two statements bolded here:

8 cores are probably better able to get good PPW than 6 (can afford lower clocks, lower voltage, so quadratic reduction in power)

F1 and F2 doesn’t matter a whole lot here

So you first are talking about lowering clocks, but then argue that frequencies don't matter? Fact is, unless you are talking about extreme conditions, since frequencies DO matter, it is a cubic formula. I'm actually backing up your original quote. There is a reduction in power since frequencies do matter.

 

[nicalandia]

I am completely confused by your two statements bolded here:

I find myself equally bewildered by your statements as well.

 

[dullard]

I am equally confused by your statements as well

Could you please clarify what part of my statements is confusing to you? Or were you referring to pakotlar?

 

[Abwx]

It’s not V^4. Read my edited post.

edit: I see your edit. The relationship is (V1/V2)^2, roughly. F1 and F2 doesn’t matter a whole lot here, except in its effect on V1 and V2, you now see that. Call V1 V2^N, then you have ~ V2^2N, in that exponential regime.

In the most favourable part of the curve it will be almost quadratic, the exponent of the power vs frequency is about 2.2, but as frequency increase this morph progressively to exponent 3 (cubic) and then 4 at the upper range of the frequencies..

Edit : Dullard is closer to real numbers.

 

[pakotlar]

In the most favourable part of the curve it will be almost quadratic, the exponent of the power vs frequency is about 2.2, but as frequency increase this morph progressively to exponent 3 (cubic) and then 4 at the upper range of the frequencies..

Edit : Dullard is closer to real numbers.

V^2N gives this relationship.

 

[pakotlar]

I am completely confused by your two statements bolded here:


So you first are talking about lowering clocks, but then argue that frequencies don't matter? Fact is, unless you are talking about extreme conditions, since frequencies DO matter, it is a cubic formula. I'm actually backing up your original quote. There is a reduction in power since frequencies do matter.

I’m arguing that the effect of frequency Xf1 in (V2^2*X*f1/V1^2*f1) is linear.

edit: The post you’re quoting is about the voltage requirements for higher clocks, not the clocks themselves, which are a linear term that doesn’t change the order of the equation.

 

[nicalandia]

V^2N gives this relationship.

Would you stop talking Gibberish Please?

 

[dullard]

I’m arguing that the effect of frequency Xf1 in (V2^2*X*f1/V1^2*f1) is linear.

edit: The post you’re quoting is about the voltage requirements for higher clocks, not the clocks themselves, which are a linear term that doesn’t change the order of the equation.

I'll try one last time. I'm talking about Power since this is what started this branch of the conversation.

so quadratic reduction in power

1) Power is related to both (A) voltage and related to (B) frequency.
A) The frequency vs voltage curve has different shapes depending on where you are looking. But, you are correct that the effect of frequency alone in many regions is linear with respect to power.
B) But you also have the quadratic portion of the power function with respect to voltage.
2) Multiply both effect (A) and effect (B) since power depends on both frequency and voltage and you get a cubic formula for power with respect to voltage (or frequency however you look at it). As @Abwx said, if you go quite low in frequency/power then the relationship is close to a power of 2.2 and if you go quite high in frequency/power the relationship is close to a power of 4. But, over the majority of the curve where many CPUs exist, it is a cubic formula.

You are only talking about half the picture but you are referring to the result as the whole picture. You can't focus on just item (A) or just item (B). The end result, which is what the discussion is about is item #2 which combines two different effects. In other words, go back to your post #798 and update the graph for Power on the y-axis instead of voltage and you'll see what we mean.

 

[pakotlar]

I'll try one last time. I'm talking about Power since this is what started this branch of the conversation.


1) Power is related to both (A) voltage and related to (B) frequency.
A) The frequency vs voltage curve has different shapes depending on where you are looking. But, you are correct that the effect of frequency alone in many regions is linear with respect to power.
B) But you also have the quadratic portion of the power function with respect to voltage.
2) Multiply both effect (A) and effect (B) since power depends on both frequency and voltage and you get a cubic formula for power with respect to voltage (or frequency however you look at it). As @Abwx said, if you go quite low in frequency/power then the relationship is close to a power of 2.2 and if you go quite high in frequency/power the relationship is close to a power of 4. But, over the majority of the curve where many CPUs exist, it is a cubic formula.

You are only talking about half the picture but you are referring to the result as the whole picture. You can't focus on just item (A) or just item (B). The end result, which is what the discussion is about is item #2 which combines two different effects. In other words, go back to your post #798 and update the graph for Power on the y-axis instead of voltage and you'll see what we mean.

I appreciate the discussion. I was using a simplifying asumption because f in the power equation is a linear term, and delta v captures most of the effect, to argue that you can’t directly compare a 6 core’s PPW and and 8 core’s, without taking into account that the latter doesn’t need to clock as high. You’re arguing that I should have included the linear term f, and that taking f into account power scaling in that regime is closer to 2.2, and in the exponential voltage vs frequency curve regime more like 3 (power). I have nothing more to add, thanks for the interesting discussion.

 

[pakotlar]

Would you stop talking Gibberish Please?

Appreciate your thoughts.

 

[dullard]

I appreciate the discussion.

I think we are on the same page now. To bring it back to Raptor Lake, Intel is pushing the frequency quite high. That requires a higher voltage than otherwise would be necessary. Since power is more than quadratic with respect to performance, the end result is that while Raptor Lake will perform quite well it will also have a capability of high power usage at default settings. You can always lower the power usage without too much impact on performance since Raptor Lake will be in the highly non-linear part of the curve.

The end result will be:

• Reviews will show high initial Raptor Lake performance
• Reviews will show high initial Raptor Lake power consumption
• Most reviews will end there and we'll have endless debates about high performance and high power.
• If you actually run tests for a longer period of time, or if you don't use performance mode, or if you don't use a high-end motherboard, of if you don't use a high-end cooler then you will get lower performance but also much lower power consumption. In other words, reviews will only cover situations that enthusiasts use, not situations where the bulk of the users are at.
• You are correct that adding cores will impact performance per watt greatly in multithreading tasks.

 

[SmokSmog]

Stock 13900K AIDA FPU test holding 5,2ghz P 4-4,1ghz E
At that clocks it will easily score over 37K CB R23 ( I did the math)

https://twitter.com/x/status/1547940989234741249

 

[nicalandia]

Stock 13900K AIDA FPU test holding 5,2ghz P 4-4,1ghz E
At that clocks it will easily score over 37K CB R23 ( I did the math) >400 Watts

https://twitter.com/x/status/1547940989234741249

7950X will reach 38K+ on CB R23(Someone else did the math, pretty accurate by the way) While using less than power

 

[Schmide]

watts = volts² / ohms

In an 8 core processor, each core can be fed lower voltage to achieve iso performance in MT workloads (than in the 6 core version, assuming same binning), which will lead to a quadratic reduction in power usage.

Even if you could use that formula, a "quadratic reduction" would mean the change in power is a second order equation implying the power curve is a third order equation.

The only thing you can measure for a semiconductor is current and voltage thus P = V^2 / R is invalid as it is not an ideal resistor. You can only use P = I V.

You could probably find many orders of polynomials using a newtons divide difference method that satisfy the power curve.

Lets refer to it as a power curve and reference points on it. At small deltas the change is basically linear. Euler

 

[SmokSmog]

7950X will reach 38K+ on CB R23(Someone else did the math, pretty accurate by the way) While using less than power

CPU monkey calculated those scores from 5950X with PBO ON, not stock, so invalid, they multiplied 5950X PBO score by 1,35X

5950X is doing around 33500 points at 5ghz all core, I'm right ? Lets say that Zen4 has 5% IPC gain in cinebench R23 ( I know that they claimed 8% on average but last time when they inflated Zen3 IPC with games they claimed 19% IPC over Zen2 while cinebench gained 12%~) .

So you have 5ghz all core 7950X scoring 35200~ points. To reach 13900K from chinese review you would need around 5,3ghz on all cores on 16 core Zen4.

 

[nicalandia]

CPU monkey calculated those scores from 5950X with PBO ON, not stock, so invalid, they multiplied 5950X PBO score by 1,35X

5950X is doing around 33500 points at 5ghz all core, I'm right ? Lets say that Zen4 has 5% IPC gain in cinebench R23 ( I know that they claimed 8% on average but last time when they inflated Zen3 IPC with games they claimed 19% IPC over Zen2 while cinebench gained 12%~) .

So you have 5ghz all core 7950X scoring 35200~ points. To reach 13900K from chinese review you would need around 5,3ghz on all cores on 16 core Zen4.

You are day dreaming if you believe that it's only 5% IPC gains.

PBO is very easy to set up and will not use >400 Watts, how much more juice can the 13900K sustain?

 

[IntelUser2000]

You are day dreaming if you believe that it's only 5% IPC gains.

PBO is very easy to set up and will not use >400 Watts, how much more juice can the 13900K sustain?

You are missing something or you are deliberately ignoring it. At 5.5/4.3 All Core it's at 400W. Was anyone expecting that clocks at all core? He's saying 5.2/4.1 it's 253W which is Alderlake level.

 

[nicalandia]

You are missing something or you are deliberately ignoring it. At 5.5/4.3 All Core it's at 400W. Was anyone expecting that clocks at all core? He's saying 5.2/4.1 it's 253W which is Alderlake level.

Exactly, if a 5.2/4.1 is only using 253W but three hundred Mhz more it's consuming 180 Watts more. Then that means that at stock it's already at it's max...

 

[SmokSmog]

Exactly, if a 5.2/4.1 is only using 253W but three hundred Mhz more it's consuming 180 Watts more. Then that means that at stock it's already at it's max...

They removed PL2 limits and it started to boost to 5,5ghz with 400W power draw, I'm right or I'm missing something ? Because you can do way better job setting it manually.

 

[IntelUser2000]

Exactly, if a 5.2/4.1 is only using 253W but three hundred Mhz more it's consuming 180 Watts more. Then that means that at stock it's already at it's max...

Of course it is. Alderlake doesn't do 5.2 in MT either. TPU even with OC only got to 5.1GHz and they didn't measure power use. Alderlake reaches 400W just at 5.3GHz: https://wccftech.com/intel-core-i9-...400w-power-consumption-in-aida64-stress-test/

Core i9 12900K processor review

Meet Intel's new flagship CPU, the Core i9 12900K. It is based on the Alder Lake architecture and is reviewed here. This time around, Intel was back at the drawing board, creating a completely new ar... Overclocking

www.guru3d.com

Our Intel Core i9-12900K CPU used 469W in the stress tests at 5.3 GHz

Raptorlake is already quite a bit better if it can reach 5.5 at same power. Since it's 253W at 5.2 it does what Alderlake does at 400-460W. Seems they managed to up the clocks for the P cores without increasing power which I think it's very impressive at such clocks.

Of course you won't point that out because it's not in with what everyone is doing right? Which is to say Alder/Raptor is a furnace and AMD will crush it? If you believe it somehow benefits AMD then you aren't because misleading information backfires.

At 5.2/4.1 it should be over 35K which is going to be comparable with Z4 and the latter is increasing power use while former is capped so while Raptorlake is still using more power they'll be closer next gen than now. Doesn't seem bad at all for Intel to me. Also the Chinese bench is showing 40% gain for Raptorlake in Blender which means when AMD said "31% faster" which is 45% faster than Alderlake but maybe 3-5% faster than Raptorlake and considering Alderlake is behind Zen 3 in Blender quite a bit that also suggests Intel will close the gap.

The leakers such as Skyjuice are also part of the media that misleads and misinforms or at the best, exaggerates it to gain attention. And of course they are fully aware of what they are doing as people don't like hearing news contrary to information they want to hear.

 

[itsmydamnation]

5950X is doing around 33500 points at 5ghz all core, I'm right ? Lets say that Zen4 has 5% IPC gain in cinebench R23 ( I know that they claimed 8% on average but last time when they inflated Zen3 IPC with games they claimed 19% IPC over Zen2 while cinebench gained 12%~) .

SPEC 2017 is my favourite game of all time, that final boss battle , one for the ages.

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

IPC wise, looking at a histogram of all SPEC workloads, we’re seeing a median of 18.86%, which is very near AMD’s proclaimed 19% figure, and an average of 21.38% - although if we discount libquantum that average does go down to 19.12%.

people and their crappy agendas, why such dishonesty?

 

[SmokSmog]

At 5.2/4.1 it should be over 35K

Manually locked 5,2ghzP and 4,1ghzE would score 38K flat

5ghz 8P = 20 732~
5,2ghz 8P = 21 562~
4ghz 4E = 4050~
4,1ghz 4E = 4 151~
4,1ghz 16E = 16 605~

21562+16602=38 167

people and their crappy agendas, why such dishonesty?

Yet in Cinebench R20 it's only 11%IPC over Zen2, so I wouldn't expect much gains in cinebench for Zen4 over Zen3 if AMD claims 8% IPC gain on average.

 

[IntelUser2000]

If AMD said 8% in average I expect the same for Cinebench. Anything else is a guess. We don't have leaks like with RPL. Every microarch tweak affects applications differently every time. Just because Zen 3 got lower than average in Cinebench doesn't mean same will be for Zen 4. What happens if Zen 4 gets 15% in Cinebench?

Also I'd put anything over 36K for RPL to be on the optimistic side. Really, anything above without going over ADL power is a bonus. 5.5 doesn't do 38K on a QS chip so how can a 5.2?

Regardless I expect two competing companies to be the closest in a while.

 

[SmokSmog]

5.5 doesn't do 38K so how can a 5.2?

I don't know what clocks 13900K has in cinebench, it could have had 5,5ghz on P cores but way lower E cores.

5,2ghz on 8 P cores locked and 4,1ghz on 16 E cores locked will net you 38K. How I know this ? I have ADL and I tested it.
6P 5ghz = 15550
6P 5ghz 4E 4ghz = 19600
4E 4ghz = 4050

Rest is just a math.