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.

 

[Carfax83]

There s no way that the 13900K is more efficient in CPU only tasks, the 7950X efficency is even better in INT based codes, FI the 7950X use only 189W in Handbrake and is 11% faster than the 13900K@253W.

And the 13900K is faster in Handbrake AV1 by 16% at 253w.

Besides Computerbase tests are GPU accelerated as well, such gaps cant be explained by CPU perfs.

Do you honestly think Computerbase.de are so stupid they can't even use software and hardware mode properly?

They benched both software and hardware mode and they explicitly state which is which.

 

[Abwx]

And the 13900K is faster in Handbrake AV1 by 16% at 253w.



Do you honestly think Computerbase.de are so stupid they can't even use software and hardware mode properly?

They benched both software and hardware mode and they explicitly state which is which.

13900K is faster in Handrake when QSV is used, otherwise the 7950X is 11% faster at stock.

As for computerbase always being on point just think that in their ST IPC comparisons they use a flawed version of POV Ray that use AVX2 only on Intel CPUs.

This software is also used in their MT IPC and perfs comparisons, albeit with much less inluence than in the ST tests where s there only two CB versions and POV Ray as metrics.

 

[lopri]

We shall see how the server adoption pans out this year. That will give a clearer answer of sort. Server people take perf/W very seriously unlike desktop folks.

 

[Carfax83]

And? How does it help you to use less power under that scenario?

If your PC is idling and barely being utilized, why shouldn't it use the least amount of power possible? Idle power usage is an important part of the overall energy efficiency equation.

I don't get your argument. You're implying that idle power isn't important, but who wants to deal with high power usage when doing simple tasks like web browsing and video playback?

GPUs idle as well. Are you going to tell me that my RTX 4090 dropping down to 210mhz isn't helpful when it has very low usage?

 

[Carfax83]

13900K is faster in Handrake when QSV is used, otherwise the 7950X is 11% faster at stock.

Would you stop repeating this nonsense? I already called you out on it before. The iGPU in ADL and RPL is the UHD 770 and it does not even support AV1 encoding. It does AV1 decoding, but not encoding.

Meteor Lake will support AV1 encoding with its iGPU.

As for computerbase always being on point just think that in their ST IPC comparisons they use a flawed version of POV Ray that use AVX2 only on Intel CPUs.

This was already debunked. PoV Ray hasn't updated their software in a long time and is no longer even relevant. Computerbase.de shouldn't use it for IPC performance at all.

 

[Timur Born]

AC/DC LLC settings are different in my Gigabyte BIOS compared to other boards, but essentially I am using lowest AC/DC LLC ("Power Saving", default is 2nd lowest) combined with 3rd lowest CPU LLC load-line ("Medium", default is lowest) and an adaptive offset of -0.84v. As a result my CPU's voltage stays between 1.2v (P95 SFFT AVX2 at 55x after droop) to 1.3v (BIOS Vcore reading) with spikes up to 1.32v (bin) for single/low-core load.

40k+ scores in Cinebench uses about 240 W (1.25v) below 80°C instead of 290 W edging on the 100°C limit at stock. This was my primary goal, hence the flattening of the voltage/droop curve via medium CPU LLC.

One important part of my undervoltage is that at -0.84v (effective -0.12v to -0.15v in combination with AC/DC LLC) I use a power limit of 253 W (PL1 and PL2) to keep P95 SFFT AVX2 load stable. Realworld load never (!) hits the power limit, only power-virus stress tests like P95 and Y-cruncher does.

Last but not least I use a per-core (non all-core) overclock within my undervoltage limits. This was just the icing on the cake to use the headroom, because voltages were already determined anyway. These settings currently seem stable on my 13900K (AVX load is hardlocked to 58x max per core even with 0 AVX offset):

Turbo Ratio Limits - AVX2, Resolved: 60x (1c), 59x (2-3c), 58x (4c), 57x (5c), 56x (6c), 55x (7-8c)
Maximum Per-core Ratio Limits (Current): 57, 56, 58, 59, 55, 55, 61, 60, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43, 43

I am disappointed at Intel for making me go through the troubles of finding my own stable undervoltage (and per core OC) after paying 750 EUR for a CPU (now down to 630 EUR). The differences between stock and undervoltage are huge and I should not have to put in the time and effort for the money paid, but gotten somewhat corresponding CPU stock settings right away (within reason).

PS: My AIO fans run at 400 rpm most of the time (pump is coupled to this) and max out at 1500-1600 rpm.

 

[coercitiv]

I am disappointed at Intel for making me go through the troubles of finding my own stable undervoltage (and per core OC) after paying 750 EUR for a CPU (now down to 630 EUR). The differences between stock and undervoltage are huge and I should not have to put in the time and effort for the money paid, but gotten somewhat corresponding CPU stock settings right away (within reason).

How exactly to do you expect Intel to dial in the optimal undervolt on your CPU when it drastically depends on:

• your motherboard
• your cooling (including the mechanical characteristics of your mount)
• your PSU
• your tolerance for system stability

 

[Timur Born]

I expect not to be able to dial in a 20% (twenty) difference, but get stock settings that are much closer to my own end-result. CB23 should not run at 290 W at stock if my end-result is 240 W, that's too much tolerance/headroom for the price. With CPUs already being sold long past the point of diminishing returns anyway I expect them to be better out of the factory.

 

[stebler]

The datasheet has been updated, including the section about DLVR.

 

[nicalandia]

Will any of you fools admit that MLD was right?

 

[Timur Born]

Gloom and doom keeps the audience invested. Be it classic 20th century housewives or 21th century tech nerds, they all love their idle gossip.

 

[IEC]

He still can't properly use their vs they're so I tend to discount his opinion heavily.

 

[Exist50]

Will any of you fools admit that MLD was right?

View attachment 73981

Right about what, pray tell? His predictions have consistently been a joke. Why even post this utterly vacuous tweet?

 

[Henry swagger]

Will any of you fools admit that MLD was right?

View attachment 73981

The basement dweller needs to be banned.. and stop posting him are you a fan of his ? Lol

 

[Kocicak]

Here is probably the first review of 13400 and 13500 CPUs, with the bonus of the funny polish language.

 

[DrMrLordX]

If your PC is idling and barely being utilized, why shouldn't it use the least amount of power possible?

Because it isn't on battery. My PC idles at over 100W in total, and it causes me no problems whatsoever. I can think of very few circumstances where partial load power draw is really all that important for a desktop system in use by a creator.

 

[Det0x]

How much power does an undervolted 7950X use for a 40-41k Cinebench 23 results? My undervolted 13900K uses about 240 W.

~170-180w

 

[Timur Born]

~170-180w

Could you publish a screenshot showing its 40k+ score at 180 W max?

 

[Det0x]

*double post, forum is really laggy atm*

 

[Det0x]

Could you publish a screenshot showing its 40k+ score at 180 W max?

Sure i could, but let me ask you first: how far do you want to move the goalposts after i have shown it ?

 

[Det0x]

Could you publish a screenshot showing its 40k+ score at 180 W max?

Here you go, 175w sustained = 40.7k score

https://imgur.com/a/tCBVe9B

Where are your screenshot btw ?
*forum insanely slow*

 

[Timur Born]

how far do you want to move the goalposts after i have shown it

I have no horse in this race and only care for information. My CPU journey was 4970K -> 1800X -> 9900K -> 5900X -> 13900K. I am very thankful that AMD introduced Zen.

Here you go, 175w sustained = 40.7k score

Very nice. I like that! One of the main drawbacks of going from 9900K to 5800X was that the same game would draw considerably more power with the 5800X using the same performance settings and fps. Idle power also was too high for my taste (the mainboard likely was also to be blamed for this).

My considerably undervolted 13900K uses a lot more power for this, albeit it could be undervolted further if I did not try to stay Prime95 SFFT AVX stable at the 253 W power limit. My Single score is 2360 and higher single/low core performance was one driving force to switch from the 5900X.

 

[Timur Born]

One thing I noticed is that your "average load" is only about 78%. So the CPU isn't fully utilized by CB23, which likely adds to its low power usage. Still a testament to its efficiency.

 

[Kocicak]

Here you go, 175w sustained = 40.7k score

This must be a world record or some reporting bug.

Note that your screenshot proves nothing, you show a result of a previous run and the current run without result.

If you got an excellent silicone and managed to squeeze the most out of it close to breaking point, it is unfortunatelly irrelevant for most people, because they cannot get such silicone, cannot tune it and will not be willing to run it without some safety margin.

 

[Det0x]

This must be a world record or some reporting bug.

Lower power runs are much more efficient.. Take at look in the Assessing Power and Performance Efficiency thread
Dont know where you get the idea this is anything special.. Zen4 is very power efficient when you know how to tune/undervolt (which Timur Born was talking about in the first place)

Note that your screenshot proves nothing, you show a result of a previous run and the current run without result.

What do you want me to do ? Record a video while running the benchmark ?
You do know benchmate score-screen is valid validation even for hwbot.. (tamper proof)

aaa why do i even bother