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.

 

[tamz_msc]

Well the questionis, why would you run p95 at 230w? :O

At 230 W PPT it will more than likely throttle. With PBO enabled it'd be fun to see the limits of Zen 4.

 

[Just Benching]

At 230 W PPT it will more than likely throttle. With PBO enabled it'd be fun to see the limits of Zen 4.

Of course it will throttle but...so what? I'm pretty sure you can run the CPU at more reasonable wattage and barely lose any performance. Just because AMD decided to up the wattage by a ton to compete with Intel doesn't mean you have to run the CPUs at the stock limits. I'm sure the 7950x at somewthing around 150w will still be plenty fast and run cool.

 

[biostud]

Sure, I expect the 13600k to be at least 50% faster than the 7600x in MT workloads like CBR23.

The 13700k is basically a 12900k on steroids, and that should also be around 50% faster than the 7700x.

The 3d is only good for gaming (and even there, it depends on the game).I don't think it will bring any meaningful improvements in other workloadds

OK, so when you are talking about wiping the floor it is purely MT workloads?

Personally I only use my computer for gaming and light photo editing as the most demanding tasks, so anything above 8 cores is really wasted on me.

 

[Just Benching]

OK, so when you are talking about wiping the floor it is purely MT workloads?

Well what else is there, all CPUs from both companies will have very very similar ST performance. Even top to bottom models will have a 10% difference in ST (the 7600x to the 7950x or the 13600k to the 13900k).

For your use case a 13600k is a no brainer really, assuming it is priced below 350$, and will be a huge upgrade to what you have right now.

 

[uzzi38]

At 230 W PPT it will more than likely throttle. With PBO enabled it'd be fun to see the limits of Zen 4.

Define throttle.

 

[tamz_msc]

Define throttle.

Reach Tjmax faster, hence lower clocks sooner.

 

[nicalandia]

Numbers pretty much as expected, but it seems like even an elevated powerlimit wont be enough to save raptor lake against zen4 this time..

Gaming performance regressions are due to Beta BIOS, I fully expect Raptor Lake to beat ADL across the board by at least 5%

 

[uzzi38]

Reach Tjmax faster, hence lower clocks sooner.

That's not throttling by definition. Especially not on a chip that has no rated all-core boost.

Throttling refers to falling below base clock.

 

[PJVol]

Throttling refers to falling below base clock.

Not necessarily when talking about Zen. Afaik zen2/zen3 always throttling in fpu heavy workloads being unrestricted by any infrastructure limit, and this is achived by the internal throttler based on accumulated activity currents per part (IP).
(actually its real implementation could be based on other than EDC metric, i.e. Cac control within weighted Cac threshold)
Whenever a certain IP activity current exceeds it's predetermined weighted threshold, power and/or clock gating activated.
For example, in LinpackXtreme 10Gb stresstest my 5600X consume ~ 125W PPT (130W limit set) and no other limit is reached either.
Yet it's power gated by the EDC throttler, running ~ 4600-4650mhz all cores.

Throttling refers to falling below base clock

This should happen on PROCHOT followed by HTC mode (High Temperature Control) with a predefined low-power P-state.

 

[uzzi38]

Not necessarily when talking about Zen. Afaik zen2/zen3 always throttling in fpu heavy workloads being unrestricted by any infrastructure limit, and this is achived by the internal throttler based on accumulated activity currents per part (IP). Whenever a certain IP activity current exceeds it's predetermined weighted threshold, power and/or clock gating activated.
For example, in LinpackXtreme 10Gb stresstest my 5600X consume ~ 125W PPT (130W limit set) and no other limit is reached either.
Yet it's power gated by the EDC throttler, with a clocks at ~ 4600-4650 all cores.


This should happen on PROCHOT followed by HTC mode (High Temperature Control) with a predefined low-power P-state.

If you're going to talk about power and/or current throttling, then you should say so. By default people interpret "throttling" as falling below base clock.

By that measure, I would be entirely justified in saying that the 12900K throttles at it's 241W power limit. Or that the 13900K would still be throttling at 350W. Would you be perfectly fine if I made that statement? That also applies to Tamz as well. Mind you, I also think running Raptor Lake past 250-260W is plain stupid, but people will still do it anyway just to squeeze out every last drop of performance.

In any case, at the stock 230W power limit I'd bet on there being a lot less headroom than you might think. In fact, I'd say that you'd see a <10% increase to MT performance even when running AVX-512 code.

 

[PJVol]

I won't argue with regards to intel power management having little knowledge on this topic.
If we're speculating on zen4 throttling scenario, I believe it won't differ much from a previous gen, except that the thermal limit will manifest itself to a greater extent.
And of course people are free to interpret "throttling" however they want )

In any case, at the stock 230W power limit I'd bet on there being a lot less headroom than you might think

That might very well be true.

 

[tamz_msc]

That's not throttling by definition. Especially not on a chip that has no rated all-core boost.

Throttling refers to falling below base clock.

Nowhere is it defined that throttling means falling below base clock.

 

[coercitiv]

Nowhere is it defined that throttling means falling below base clock.

And yet, by definition, a CPU operating above base clocks is boosting, therefore not throttling.

PS: I do think though that we have started to use the term throttle much more loosely lately, so I have no formed opinion on the matter.

 

[uzzi38]

I should probably add that my expectation for extra MT headroom is a guess rather than something I know for sure.

 

[biostud]

And yet, by definition, a CPU operating above base clocks is boosting, therefore not throttling.

PS: I do think though that we have started to use the term throttle much more loosely lately, so I have no formed opinion on the matter.

The 5950X is sold as a 4.9/3.4Ghz processor, to me that means it can and should boost to 4.9Ghz under ST load, and be able to stay there under prolonged time given that you hav correct cooling, and that under full MT load it should boost all cores 3.4Ghz and also be able to keep those frequencies for prolonged time. So for me, boosting is just changing the frequency up when the core is under load, sometimes to, and sometimes above specified frequencies. But if it is not able to hold that frequency over time, then it is throttling.

 

[PJVol]

So for me, boosting is just changing the frequency up when the core is under load, sometimes to, and sometimes above specified frequencies. But if it is not able to hold that frequency over time, then it is throttling.

As i've said before anyone can mean anything by "throttling", as for the AMD themselves, they define boost as the maximum operating frequency achievable within predefined infrastructure limits.
The "throttling" implies forced power and/or clock gating over that boosting behavior.
I already gave an example of two cases when throttling come into effect with Zen2/3.
HTC (high temperature) throttling and the "high switching activity current" throttling.

 

[moinmoin]

The 5950X is sold as a 4.9/3.4Ghz processor, to me that means it can and should boost to 4.9Ghz under ST load, and be able to stay there under prolonged time given that you hav correct cooling, and that under full MT load it should boost all cores 3.4Ghz and also be able to keep those frequencies for prolonged time. So for me, boosting is just changing the frequency up when the core is under load, sometimes to, and sometimes above specified frequencies. But if it is not able to hold that frequency over time, then it is throttling.

3.4GHz is the base frequency, 4.9GHz the boost one. The former is promised to hold when all cores are active, the latter is the max a core is promised to achieve in ideal circumstances. The boost range is between 3.4 and 4.9GHz depending on workload and active cores. Throttling to me then is dropping below the base frequency. (Frequency is highly fluctuating anyway based on workload so calling that behavior "throttling" already would make the word pretty meaningless.)

 

[tamz_msc]

And yet, by definition, a CPU operating above base clocks is boosting, therefore not throttling.

PS: I do think though that we have started to use the term throttle much more loosely lately, so I have no formed opinion on the matter.

Intel, at least, defines throttling as the CPU reducing clocks when hitting Tjmax.

What Is Throttling and How Can It Be Resolved?

Explains throttle and tips on how to manage and avoid in Intel® Boxed processors and Laptop processors.

www.intel.com

In hindsight, I was wrong when I said that the 7950X would throttle running p95 small fft with AVX-512 enabled at 230 W PPT. It will be power-limited, but given the amount of heat that workload produces, I wouldn't be surprised if it throttles as well.

 

[Kosusko]

Geekbench5 a i9-13900K s 6000 DDR5 - Raichu user:

13900k+Z690+6400cl34-38 - Geekbench

Benchmark results for a Z690 Taichi Razer Edition with an Intel Core i9-13900K processor.

browser.geekbench.com

 

[Zucker2k]

Looking good but let's wait for reviews 😇

 

[Abwx]

Geekbench5 a i9-13900K s 6000 DDR5 - Raichu user:

13900k+Z690+6400cl34-38 - Geekbench

Benchmark results for a Z690 Taichi Razer Edition with an Intel Core i9-13900K processor.

browser.geekbench.com

DDR6400...

https://twitter.com/x/status/1567109256960417793

 

[Zucker2k]

DDR6400...

https://twitter.com/x/status/1567109256960417793

Intel loving that bandwidth. That's some nice performance scaling with RAM speed.

 

[tamz_msc]

That might be the first time a CPU breaks 2000 int score at stock. Mighty impressive.

 

[nicalandia]

That might be the first time a CPU breaks 2000 int score at stock. Mighty impressive.

It's Not at Stock..!

 

[tamz_msc]

It's Not at Stock..!

Of course it is.