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.

 

[Mopetar]

Is this It? Is this Intel's counter to the X3D? $700 CPU?

At this point it's pretty clear that Zen 4D will have the performance crown in most titles, but I'd argue that Intel's counter has always been the 13600 which has a much more affordable price, 95+ % of the performance as CPUs twice the price, and enough E-cores to stand up quite well against competing Ryzen chips in other workloads for people who aren't using their rig solely or primarily for gaming.

 

[PJVol]

I wish I knew what "Auto" actually was when it comes to LLC

Assuming VRM on intel boards is no other than on amd, "Auto" usually means one out of seven LL resistance value (most likely 0%), e.g. +40%, +20%, 0, -20%, -40%, -60%, -80% based on initially programmed RLL. (0.5-1.0 mΩ for example).
Anyway, you can see the exact value in PWM controller's datasheet.

 

[coercitiv]

Now I change CPU Current Capability from "Auto" to "120%." I haven't run a long-term test but CB seems to run fine.

From what I can tell "CPU Current Capability" is a term specific to Asus and represents the overcurrent protection. Triggering this protection probably results in a hard reset or system shutdown. If LLC on "Auto" and Current Capability at "120%" results in a stable system with everything else at stock value, then you likely have your culprit. It's very odd though, or maybe I have a different understanding of these limits based on my experience with MSI boards where I have current limit (which results in throttling) and current protection (which probably results in hard reset).

 

[exquisitechar]

Is this It? Is this Intel's counter to the X3D? $700 CPU?

View attachment 74520

Intel Core i9-13900KS Review: Hitting 6 GHz

The new Core i9-13900KS is very much like the 12900KS and 9900KS before it -- this is a mildly overclocked 'special edition' CPU, binned silicon meant to...

www.techspot.com

The KS is irrelevant, fast DDR5 is what will allow Raptor Lake to keep up.

 

[nicalandia]

....

The KS is irrelevant, fast DDR5 is what will allow Raptor Lake to keep up.

Well 7200 is fast(good luck having reviewers use 8000 RAM Kit) and it's 2.7% faster than the 7700X

 

[Carfax83]

....

Well 7200 is fast(good luck having reviewers use 8000 RAM Kit) and it's 2.7% faster than the 7700X

Um, GPU bottleneck?

HWUB once again shows that they can't really conduct proper CPU gaming benchmarks. Their benchmark selection clearly suffers from GPU bottlenecking. The fact that the CPUs are closely packed together like that says it all.

Compare that to more competent reviewers like Computerbase.de or PCGH.de that actually find CPU limited areas in these games and use settings that increase CPU bottlenecks, like ray tracing.

 

[nicalandia]

Um, GPU bottleneck?

Okay how about 1080P?

The 13900KS with 7200 RAM is only 5% faster than the 7700X.

 

[Carfax83]

The KS is irrelevant, fast DDR5 is what will allow Raptor Lake to keep up.

With the way that HWUB/Techspot does their testing, Raptor Lake won't even need fast DDR5 because they clearly have no idea how to do proper CPU benchmarks that actually isolate CPU performance.

 

[Carfax83]

The 13900KS with 7200 RAM is only 5% faster than the 7700X.

View attachment 74559

The tests should be done at 720p in the most CPU limited areas of the game, with RT turned on.

Then you will see large gaps between these CPUs. He has three games in the lineup that support RT, and he didn't even bother turning it on.

I used to think that HWUB were biased, but now I know they are just straight up incompetent.

 

[exquisitechar]

With the way that HWUB/Techspot does their testing, Raptor Lake won't even need fast DDR5 because they clearly have no idea how to do proper CPU benchmarks that actually isolate CPU performance.

Yeah, a lot of reviewers are like that. Well, what their benchmarks show is that the CPU often doesn't matter that much and you can usually get a good experience with far from the fastest ones, but they aren't representative of the differences in CPU demanding scenarios.

 

[Hitman928]

The tests should be done at 720p in the most CPU limited areas of the game, with RT turned on.

Then you will see large gaps between these CPUs. He has three games in the lineup that support RT, and he didn't even bother turning it on.

I used to think that HWUB were biased, but now I know they are just straight up incompetent.

Why not 384p like Anandtech uses?

 

[nicalandia]

In anyone needs 480P or 720P to demonstrate how good a 13900KS is over a 7700X instead of 1440P then it really does not matter what CPU but it's the GPU that you should be concentrating your efforts on.

 

[Carfax83]

Why not 384p like Anandtech uses?

View attachment 74561

Or they could just get a faster GPU like most reasonable people. The 6950XT just can't cut it with these newer CPUs.

 

[Hitman928]

Or they could just get a faster GPU like most reasonable people. The 6950XT just can't cut it with these newer CPUs.

But lower resolution is better for CPU tests, is it not?

 

[Carfax83]

In anyone needs 480P or 720P to demonstrate how good a 13900KS is over a 7700X instead of 1440P then it really does not matter what CPU but it's the GPU that you should be concentrating your efforts on.

Or you can just you know, turn on RT like I've been saying. I can tell you guys don't really game at all, because many modern games are actually more CPU heavy than they have ever been due to things like crowd densities, draw distances, ray tracing etcetera being incorporated to greater effects than ever before due to the current gen consoles much faster CPUs than last gen.

I posted a video on the previous page of a particular area in Cyberpunk 2077 that just crushes CPUs due to heavy load it puts on them from RT effects plus crowd and detail density.

None of you have bothered to even respond to it, which is quite telling. Every CPU is CPU bound in that area with those settings regardless of resolution. However, Raptor Lake is over 100% faster than Zen 3D in that particular area.

 

[Carfax83]

But lower resolution is better for CPU tests, is it not?

Yes, but only in terms of draw calls and frame set up which aren't really CPU intensive for modern CPUs typically speaking (especially with DX12 and Vulkan), unless a particular area in the game is exceptionally detailed. Workloads that are the most CPU intensive in today's world are RT BVH, A.I, physics simulation.

I posted a video on page 158 that showed an area in CBP 2077 that is extremely CPU bound due to the confluence of several factors, most notably RT BVH calculations, A.I from crowd densities, and the draw calls from object detail and additional RT effects.

Even the 13900K is CPU bound, but it handles the area far better than the 5800X3D comparison that I posted, to the tune of more than 100% faster.

 

[DrMrLordX]

It's funny how the 13900KS runs out of breath in the 10 minute test. This is something Intel needs to work hard on.

It comes down to cooling. Plus in the right (or wrong, depending on your PoV) motherboard, the 13900K is going to have similar default boost/power behavior to the KS anyway, with the KS maybe performing a tiny bit better (again, assuming the cooling is there).

If you really want a KS, you can probably take a decent 13900K sample and just tweak it yourself.

 

[Hitman928]

Yes, but only in terms of draw calls and frame set up which aren't really CPU intensive for modern CPUs typically speaking (especially with DX12 and Vulkan), unless a particular area in the game is exceptionally detailed. Workloads that are the most CPU intensive in today's world are RT BVH, A.I, physics simulation.

So why then would testing 384p or 480p not be the preferred resolution to make sure to eliminate all GPU bottlenecks?

I posted a video on page 158 that showed an area in CBP 2077 that is extremely CPU bound due to the confluence of several factors, most notably RT BVH calculations, A.I from crowd densities, and the draw calls from object detail and additional RT effects.

Even the 13900K is CPU bound, but it handles the area far better than the 5800X3D comparison that I posted, to the tune of more than 100% faster.

You mean the video you posted from 2 different random people running completely different systems (including the video card) in the same general area of the game but not in the same exact area or run? That's why no one responded, because it is dismissed as any kind of noteworthy comparison between CPUs on its face. No need to evaluate further.

 

[Carfax83]

It comes down to cooling.

As someone with an air cooled 13900KF at 5.3ghz 235w, the number one reason for Raptor Lake's heat and power draw issues is the factory overvolting.

It's not even the process node. It's the damn overvolting. Raptor Lake becomes another animal entirely when you optimize the voltage in terms of power draw and heat load. Because when the CPU is overvolted, it increases the power draw which also increases the heat.

Raptor Lake at loads can sip voltage like you wouldn't believe. I've had my CPU undervolted by as much as -165mV, and I never crashed not even once when the CPU was loaded, whether in games or CBR23 or Handbrake. The only time it crashed was during idling because the voltage was too low to sustain the CPU's operation.

 

[DrMrLordX]

As someone with an air cooled 13900KF at 5.3ghz 235w, the number one reason for Raptor Lake's heat and power draw issues is the factory overvolting.

Keep in mind we're discussing the KS here, which has ridiculously-high preprogrammed boost limits. It really does need that voltage/power headroom to achieve those clocks.

 

[Carfax83]

So why then would testing 384p or 480p not be the preferred resolution to make sure to eliminate all GPU bottlenecks?

I never said it shouldn't be, but because of low level APIs, lowering the resolution isn't as effective as it used to be for isolating CPU performance.

You mean the video you posted from 2 different people running completely different systems (including the video card) in the same general area of the game but not in the same exact area or run? That's why no one responded, because it is dismissed as any kind of noteworthy comparison between CPUs on its face. No need to evaluate further.

Um, you need to watch the video again. It was in the same area.....the exact same area. That area is infamous for punishing CPUs when appropriate settings are used, and both the uploaders knew exactly where to look.

As for your comment about different GPUs, I can tell you probably never even watched the video. Both GPUs were CPU bottlenecked in this particular area, and the guy with the RTX 4080 was CPU bottlenecked even at 4K DLSS.

 

[Hulk]

From what I can tell "CPU Current Capability" is a term specific to Asus and represents the overcurrent protection. Triggering this protection probably results in a hard reset or system shutdown. If LLC on "Auto" and Current Capability at "120%" results in a stable system with everything else at stock value, then you likely have your culprit. It's very odd though, or maybe I have a different understanding of these limits based on my experience with MSI boards where I have current limit (which results in throttling) and current protection (which probably results in hard reset).

Well thanks again because you're the one who figured this out for me. I think what was happening is that the CPU was requesting a certain amount of current, more than the board was set to deliver so this immediately trips the restart. Makes sense because the restart would generally happen as soon as I began the load. Why this suddenly became a problem is a bit of a mystery but apparently my CPU started needing more current. Although I will admit this did happen over the last few months but very infrequently. Now at 225W it's a guarantee.

So Intel got back to me and offered me a new CPU except for the fact that they are out of stock and it looks like 4 weeks to get one. Or they would refund me on my current CPU. I have to think about this because on one hand I seem stable now and don't want to abuse their return/refund policy, but on the other this has been an issue that has been getting worse.

 

[Carfax83]

Keep in mind we're discussing the KS here, which has ridiculously-high preprogrammed boost limits. It really does need that voltage/power headroom to achieve those clocks.

I'm not so sure. The HWUB testing showed the 13900KS averaging at 5.48ghz during the CBR23 10 minute run at 280w. My own CPU stays at 5.3ghz and draws 235w. If I had better cooling, I could get it down even lower.

 

[Hulk]

I'm not so sure. The HWUB testing showed the 13900KS averaging at 5.48ghz during the CBR23 10 minute run at 280w. My own CPU stays at 5.3ghz and draws 235w. If I had better cooling, I could get it down even lower.

I agree. From what I've read thus far the KS is a better binned K rather than an overclocked K. What's the difference you ask? Looks like the KS can achieve higher clocks than the K at the same voltage under load.

Intel Core i9-13900KS Review: Hitting 6 GHz

The new Core i9-13900KS is very much like the 12900KS and 9900KS before it -- this is a mildly overclocked 'special edition' CPU, binned silicon meant to...

www.techspot.com

 

[Carfax83]

I agree. From what I've read thus far the KS is a better binned K rather than an overclocked K. What's the difference you ask? Looks like the KS can achieve higher clocks than the K at the same voltage under load.

Yeah. Good binning is great and all, but it can't save you from stock overvolting. I figured that out real quick when I ran my CPU at default settings in CBR23. CPU hit 100c in just a few seconds and quickly down clocked to 5ghz and low 90s for the 10 minute test. I think it only hit 5.5ghz for just a second or so.

After undervolting, my CPU will do 5.3ghz (efficiency cores at 4.3ghz) over the course of the entire 10 minute CBR23 run, draw 235w and max out at 82c......on air cooling. With a really good AiO I would likely be near 200w power draw and in the mid to upper 60s at the same clock speed, with plenty of headroom to increase clock speeds.