Question Raptor Lake - Official Thread

[mikk]

It's an incremental revision. There's no real reason the chip would clock low. In any case, the figure wasn't from the actual demo, and refers to the actual product expectations.

It's from an 0000 ES, it's way too early for a QS.

 

[Exist50]

It's from an 0000 ES, it's way too early for a QS.

The "double digit" claim was from a slide about the product. They never talk about ES performance. And again, both the process and uarch are mature. There'd be no reason for an underperforming ES in the first place.

 

[IntelUser2000]

Intel is caught in a vicious circle and it's going to take an outstanding CPU generation to bring power usage under control (from the power limit PoV). Or new executive leadership, whichever comes first.

That's the thing. I don't believe the chip vendors are going to reduce TDPs, because if you can make a chip with radically improved perf/watt, then it means you can make it dramatically faster at the same power.

But you still can get the chips with lower TDPs. Like we argue you can just set the TDP lower on Alderlake or get something like 12600 and 12700 right?

The dynamic range has always been increasing. While they can get a chip that's 20% faster at 125W, if you can get to 40% faster at 250W, why not?

 

[igor_kavinski]

The dynamic range has always been increasing. While they can get a chip that's 20% faster at 125W, if you can get to 40% faster at 250W, why not?

Agree. Let the user decide. If I have a workload on which I'm willing to spend more power to save time, give me that choice.

 

[coercitiv]

But you still can get the chips with lower TDPs. Like we argue you can just set the TDP lower on Alderlake or get something like 12600 and 12700 right?

Agree. Let the user decide. If I have a workload on which I'm willing to spend more power to save time, give me that choice.

Welcome to the Intel Pop Quiz! It's a quicky, so let's get started. You have an i7 12700 CPU and a Z690 board. Assuming you use stock board settings and start CB23, the sustained CPU package power for the next hour will be:

1. 65W
2. 180W
3. 240W
4. I have no idea!

Feel free to consult the Intel spec sheet in order to deduct the expected power usage of your CPU!

Also, here's a preview of Cooler Tunning options on a Z690 MSI board:


And remember, YOU decide!

 

[IntelUser2000]

Welcome to the Intel Pop Quiz! It's a quicky, so let's get started. You have an i7 12700 CPU and a Z690 board. Assuming you use stock board settings and start CB23, the sustained CPU package power for the next hour will be:

Oh heck. Anyone who's aware of PL settings can buy a board that has custom settings lol. Half of the reviews out there have benchmarks using custom PL settings to see whether Alderlake was running beyond sane settings.

 

[igor_kavinski]

Intel Core i9-12900K Alder Lake Tested at Power Limits between 50 W and 241 W

We test Intel's Core i9-12900K at various TDP levels all the way down to 50 W to determine how much efficiency is really in the new Alder Lake core, and how these power limits affect performance. Competing with the efficiency of AMD's Zen 3 Ryzen lineup is just two settings changes away.

www.techpowerup.com

Something around 350W max. It could possibly go higher with a cryocooler.

 

[coercitiv]

Oh heck. Anyone who's aware of PL settings can buy a board that has custom settings lol.

I find it encouraging how you think so highly of the average consumers as to assume they will know what PL1 and PL2 is. But the fact still remains, you loled but were unable to expect a predictable (stock) behavior for the 12700: 65W sustained or 180W+ sustained.

Imagine buying graphics card and not knowing what the stock gaming configuration is: 100W or 300W? Don't worry though, you'll be able to customize it after reading some forums. That's what matters.

 

[IntelUser2000]

I find it encouraging how you think so highly of the average consumers as to assume they will know what PL1 and PL2 is.

I'd have chosen 240W for the poll.

No, I don't have *any* confidence on average users to adjust PL1. Knowing such information is something that'll get you bullied in school basically. As adults, that's something that's met with a dumbfounded look.

I'm not sure that matters, because they can get something like the -T.

Same logic, there will be people that don't care about power use and go for the extra bit of performance.

AMD is pretty much following the same road. They are increasing TDPs in Zen 4 since Intel is using TDPs to catch up. Why hamper Zen 4 to current TDPs?

 

[mikk]

This is what MLID claims about Raptor Lake performance gains:

 

[eek2121]

That's the thing. I don't believe the chip vendors are going to reduce TDPs, because if you can make a chip with radically improved perf/watt, then it means you can make it dramatically faster at the same power.

But you still can get the chips with lower TDPs. Like we argue you can just set the TDP lower on Alderlake or get something like 12600 and 12700 right?

The dynamic range has always been increasing. While they can get a chip that's 20% faster at 125W, if you can get to 40% faster at 250W, why not?

Except ADL-S doesn’t scale like that. They could have limited it to 125W and it would have probably narrowly beaten the 5900X.

This is what MLID claims about Raptor Lake performance gains:

MLID is wrong as usual.

 

[jpiniero]

A Q4 launch of Raptor Lake mobile is rather unlikely; given that Alder Lake mobile isn't even available yet and won't be until March. Dell has a couple laptops available to preorder but the earliest won't ship until like the 8th.

 

[Carfax83]

8-15% is very good for a refresh architecture if you ask me and ensures that Zen 4 won't just walk right over Raptor Lake. It's probably a combination of clock speed, IPC enhancements and cache increase to boost single thread performance by that much, but I'd be interested in knowing what you guys think the actual breakdown is. Clock speed is likely the least factor of that performance boost IMO, followed by either IPC or cache for the number 2 or 1 spot in terms of impact.

 

[igor_kavinski]

8% is low. 15% would be much better. Even with 15% ST perf increase, Zen 4 might walk all over it with a better memory controller or V-cache.

 

[Carfax83]

8% is low. 15% would be much better. Even with 15% ST perf increase, Zen 4 might walk all over it with a better memory controller or V-cache.

AMD's memory controller will be in their I/O die unlike Intel's which is on the same die as the CPU so I doubt it will be better; at least not in terms of latency. Also Raptor Lake should be capable of supporting higher memory speeds than Zen 4 out the gate.

 

[igor_kavinski]

I don't know. Something tells me Zen 4 will be good with DDR5. AMD will show Intel how it's done.

 

[mikk]

MLID is wrong as usual.

Actually MLID is usually right, he has good sources. It's one of the best leakers when it comes to Intel (not so much for AMD).

 

[Saylick]

Actually MLID is usually right, he has good sources. It's one of the best leakers when it comes to Intel (not so much for AMD).

Given his recent accuracy with respect to Intel products, you have to wonder if Intel is intentionally feeding him info given that Intel is the underdog and they could use a mouthpiece to stave off AMD. Given that MLID usually mentions WCCFtech, Videocardz, and Tweaktown on Twitter whenever he posts a leak video, and those people are generally quick with writing an article based on his video, it's a rather effective way to spread news of whatever Intel need spreading.

For Intel, because they are the underdog, they have to play an attacking game; if you're AMD and are the leader you have to play a defending game. AMD not having as many leaks falls under that theory.

 

[mikk]

Given his recent accuracy with respect to Intel products, you have to wonder if Intel is intentionally feeding him info given that Intel is the underdog and they could use a mouthpiece to stave off AMD. If you're the underdog you have to play an attacking game; if you're the leader you have to play a defending game. AMD not having as many leaks falls under that theory.

He knows all of his Intel sources personally he said a couple of times.

 

[Tarkin77]

This is what MLID claims about Raptor Lake performance gains:

He calls this leak? LOL. Every human with a functiton brain can come up with these numbers. If they add 8 more e-cores, yes, that could be good for 30-35% more MT performance (40% is a little optimistic). And ST - well... ~5% more IPC and ~5% more clocks... there you have your +10% (again... 15% is a little on the high side IMO)

Leak Update, yeah .. what amazing sources he has

 

[IntelUser2000]

He calls this leak? LOL. Every human with a functiton brain can come up with these numbers. If they add 8 more e-cores, yes, that could be good for 30-35% more MT performance (40% is a little optimistic). And ST - well... ~5% more IPC and ~5% more clocks... there you have your +10% (again... 15% is a little on the high side IMO)

Leak Update, yeah .. what amazing sources he has

I disagree. Lot of us here disagree that it'll be that fast in MT. And Intel presentation says otherwise. It's on the same process anyway. Personally I think 30% is possible because the P cores take up all the power and slight reduction can free up quite a bit for the E cores.

But we'll see what Intel actually does.

That said MLID has been spot on with Intel leaks. I don't know why @eek2121 is so harsh on him but he's been right on so many things.

Late Quarter 3 is August-September. That's not hugely earlier than when Alderlake launched.

A Q4 launch of Raptor Lake mobile is rather unlikely; given that Alder Lake mobile isn't even available yet and won't be until March. Dell has a couple laptops available to preorder but the earliest won't ship until like the 8th.

March 8th?

That's actually normal. The mobile chips are launched at CES but to consumers it usually takes until mid to late February or March to arrive. And that's for the first wave. Launches will happen throughout the year, some even after the launch of the next gen chips.

This is why mobile needs the full year cycle, since you can't just plop the chips in or replace the board. With Desktops CES launch means people can buy the chips few days after that.

 

[LightningZ71]

It feels like Raptor lake is just throwing more silicon at the problem. With Intel7 improving yields, they can afford to go for larger die, meaning more room for cores and cache.

 

[IntelUser2000]

8-15% is very good for a refresh architecture if you ask me and ensures that Zen 4 won't just walk right over Raptor Lake. It's probably a combination of clock speed, IPC enhancements and cache increase to boost single thread performance by that much, but I'd be interested in knowing what you guys think the actual breakdown is.

At twitter, Davidbepo has a chart with simulations of cache hierarchies and their impact on performance.

It says the cache alone will improve performance by 2-4%. So a typical tick like gain of 4-6% plus 2% clock boost and we'll end up with the 8-12% number. 15% might be a corner case scenario.

What about MT performance? And here's the point I am wondering. Whether Intel is sandbagging or not.

Let's assume Alderlake 8+8 as being 1+0.3, with 0.3 being the E cores.

Raptor Cove: 1.1

Let's say the 8 Gracemont cores need to clock down by 30%. You still end up with 1.1+0.4, or 1.5 for a MT boost of 15%. Not sure why they'll need to clock the E cores so low as it only uses 50W out of the total 250W. I think lowering P core clocks by 5% will offer them huge power budget for the extra E cores without impacting the clocks much. This assumes there's zero gain from a refined stepping, and the impact of better cache hierarchies and ring bus.

20-30% boost in MT seems pretty doable to me. Since Golden Cove is already faster than Zen 3, that is enough to be competitive with a 16 core Zen 4 in both metrics.

 

[DrMrLordX]

AMD's memory controller will be in their I/O die unlike Intel's which is on the same die as the CPU so I doubt it will be better; at least not in terms of latency.

Eh. Depending on whom you ask, Alder Lake has a rather shoddy memory controller. Raptor Lake is an opportunity for Intel to improve the situation, but . . . Rocket Lake-S also had a crappy memory controller. Do not hold your breath.

 

[IntelUser2000]

Eh. Depending on whom you ask, Alder Lake has a rather shoddy memory controller. Raptor Lake is an opportunity for Intel to improve the situation, but . . . Rocket Lake-S also had a crappy memory controller. Do not hold your breath.

Not necessarily related to Raptorlake but Davidbepo says the tile approach in Sapphire has integrated-like latencies for L3, so at least performance-wise their mesh-over-EMIB approach is working.

If they can do that then no reason Raptorlake can't be better.