News Intel's flagship Lunar Lake Ultra 9 is slower than an Ultra 7 in the single laptop where it's currently for sale

[Admin]

We tested the new Asus Zenbook S14 with the Core Ultra 9 288V processor, and surprisingly found that the Core Ultra 9 is slower than the Ultra 7 model in threaded workloads.

Intel's flagship Lunar Lake Ultra 9 is slower than an Ultra 7 in the single laptop where it's currently for sale : Read more

 

[RosElks]

Kind of obvious what's happening; Intel did it on purpose so Cristiano can't have the so much craved 'battle of the benchmarks'

 

[JRStern]

Appreciate you guys' work on all this, but it comes out so confusing. It's apparent that Intel themselves are massively uncertain what the E-core/P-core thing is all about. I'd *thought* it was about unloading IO onto E-cores so the P-cores didn't have to get interrupted, upsetting the pipeline and clearing the cache, and all, E-cores being much cheaper to build and run, and yes, I suppose they could run at different, even faster, clock speeds being simpler and all. Yet it never occurred to me that the chip could run different cores at different speeds at the same time - though maybe phones have been doing this for twenty years, so I'm a little behind on the details of some of this stuff, LOL, so was Intel.

I also wondered just how heavy it would be for Windows to factor out what they wanted to run on an E-core versus a P-core, given different core counts of each, and base versus turbo, and yada yada. So much software development, even more than other areas of engineering, seem to consist of people, even the biggest companies, just bumbling around until they trip over something that works really well, but then six other people or projects or departments just won't accept it and keep running in other directions ...

 

[bit_user]

@PaulAlcorn , with release dates slipping and now this, perhaps 2024 is shaping up to be the year of the flubbed launch? Depends on how Arrow Lake's launch goes, but if that's also rough, then I'd say it's a bad sign for the industry.

At least with AMD, I can sort of understand. They've broadened out their product portfolio a lot, in the past couple years. I think they might be getting stretched thin. In Intel's case, I have more trouble understanding why Lunar Lake (not unlike Meteor Lake before it) seems to be having a rough launch.

 

[Kamen Rider Blade]

At least with AMD, I can sort of understand. They've broadened out their product portfolio a lot, in the past couple years. I think they might be getting stretched thin. In Intel's case, I have more trouble understanding why Lunar Lake (not unlike Meteor Lake before it) seems to be having a rough launch.

I think in AMD's case, it's Left-Hand not knowing what the Right-Hand is doing, and suddenly the Middle-Hand shows up and wants to do something.

There are serious internal communications issue and standardization on BenchMark testing that needs to happen within AMD.

Also, messaging to the public & reviewers needs to be adjusted to be concise, uniform, and properly tested before any information goes out.

No more gas-lighting of reviewers over 1-2%, seriously, that's shameful.

Also, no more releasing crap w/o having tested it themselves.

They need to go Slower, Dot their I's, Cross their T's.

This way they can end up going faster over-all, and not trip up on minor issues that get blown out of proportion and cause unnecessary head-aches.

 

[thestryker]

Greatly appreciate this update and look into the performance. I noticed that most of the laptops seemed to be trying to maximize performance putting lower chips into higher power modes so I'd just chalked it up to overall clock limits at the power levels. I'm curious if this is why there haven't been any releases with the 288V or if it was always to be a lower volume product/Asus got some sort of short term exclusivity.

 

[HideOut]

So the new 2xx core ultras have both less P cores AND less E cores? That makes sense. And while we are at it lets take out HT too...

 

[TheSecondPower]

Appreciate you guys' work on all this, but it comes out so confusing. It's apparent that Intel themselves are massively uncertain what the E-core/P-core thing is all about. I'd *thought* it was about unloading IO onto E-cores so the P-cores didn't have to get interrupted, upsetting the pipeline and clearing the cache, and all, E-cores being much cheaper to build and run, and yes, I suppose they could run at different, even faster, clock speeds being simpler and all. Yet it never occurred to me that the chip could run different cores at different speeds at the same time - though maybe phones have been doing this for twenty years, so I'm a little behind on the details of some of this stuff, LOL, so was Intel.

I also wondered just how heavy it would be for Windows to factor out what they wanted to run on an E-core versus a P-core, given different core counts of each, and base versus turbo, and yada yada. So much software development, even more than other areas of engineering, seem to consist of people, even the biggest companies, just bumbling around until they trip over something that works really well, but then six other people or projects or departments just won't accept it and keep running in other directions ...

I had originally heard of phones toggling between the little cores and the big cores and never using them both at the same time. But I don't think that's how phones work today and in hindsight, I don't think that's ever how the phone processors ever worked.

I think that today little cores are primarily about having a lower cost core which can be used to supplement the big course during heavily threaded tasks when they wouldn't have been able to clock all that high anyway. In many cases they're also more power efficient than the big cores if the clock speed is low enough (big cores are more efficient if the clock speed is high enough). So in Lunar Lake, as the number of threads increases, the clock speed of all the cores goes down and eventually reaches the point where the little cores are operating more efficiently than the big cores, so the big cores clock down more than the little cores after this point.

Intel has previously described Meteor Lake as having three classes of cores: LPE cores, E cores, and P cores. In all Intel products, P cores and E cores live on the same ring bus and share the same L3 cache. But LPE cores have no L3 cache and aren't on the ring bus. As far as I know, they only exist because Meteor Lake has multiple dies and the LPE cores are on the I/O and memory controller die so that CPU die can be shut down entirely during light work.

It's also important to understand that in all Intel products a main thread is sent to the LPE cores, then the P cores, then the E cores (skipping any that don't exist). LPE cores are easy on the battery so start there, if the workload is too much send it to the P cores which are responsive, and if it doesn't need to be responsive move it to the E cores.

Lunar Lake follows the same model, the only differences from Meteor Lake being that all the cores are on the same die, and there are no E cores. But I guess for convenience, the LPE cores are often called E cores on Lunar Lake. (There's no separate CPU die to power down when the LPE cores alone are used, but all the P cores, L3 cache, and ring bus can still be turned off.)

 

[usertests]

@PaulAlcorn , with release dates slipping and now this, perhaps 2024 is shaping up to be the year of the flubbed launch? Depends on how Arrow Lake's launch goes, but if that's also rough, then I'd say it's a bad sign for the industry.

We also have port strikes and an impending quartz shortage. Maybe 2025 will be worse.

 

[JRStern]

...

Lunar Lake follows the same model, the only differences from Meteor Lake being that all the cores are on the same die, and there are no E cores. But I guess for convenience, the LPE cores are often called E cores on Lunar Lake. (There's no separate CPU die to power down when the LPE cores alone are used, but all the P cores, L3 cache, and ring bus can still be turned off.)

Interesting. Scary. Moving a thread from processor to processor ... well, but my guess is it probably doesn't, really, maybe an LPE core is dedicated to OS queue management, so the task and the LPE sees it exists and is queued but not even executing, but the first attempt to execute it is on a P-core. Now, that does require you can assume it needs a P-core as opposed to letting it demonstrate it first on an E-core, but I'd think the general class of a job is going to be known in advance, or maybe you can set some flags to have it treated one way or another.

As I said before, making proper use of these new toys/tools, is itself a challenge, just as this article says.

 

[rluker5]

I don't like how Best Buy is obfuscating the CPU model number by calling them Core Ultra series 1 or Core Ultra series 2 and not listing the included CPU by name. There is a pretty big difference.

 

[thestryker]

I don't like how Best Buy is obfuscating the CPU model number by calling them Core Ultra series 1 or Core Ultra series 2 and not listing the included CPU by name. There is a pretty big difference.

If you're talking about online they've done this for years by calling things by their family rather than specific model. It forces you to open up the page for the model you're looking at and find where they've got the actual specifications. I can't imagine it's any better in store.

 

[TheSecondPower]

Interesting. Scary. Moving a thread from processor to processor ... well, but my guess is it probably doesn't, really, maybe an LPE core is dedicated to OS queue management, so the task and the LPE sees it exists and is queued but not even executing, but the first attempt to execute it is on a P-core. Now, that does require you can assume it needs a P-core as opposed to letting it demonstrate it first on an E-core, but I'd think the general class of a job is going to be known in advance, or maybe you can set some flags to have it treated one way or another.

As I said before, making proper use of these new toys/tools, is itself a challenge, just as this article says.

When I'm watching a large single-threaded workload in Task Manager on Windows or the System Monitor on Linux, it seems to jump around between cores (and all my computers have homogenous cores with SMT or hyperthreading), so I'm not sure what's scary if some of the cores are different types. For software developers it doesn't usually matter except for the developers working on the scheduler in Windows or Linux.

 

[das_stig]

So what we can take away from this review is that the Ultra 9 is actually an Ultra 7 with power limitations removed, minor clock boosts and more expensive pricing. Intel no longer the leading CPU manufacturer, handed than crown to AMD again!

Maybe Qualcomm should buy them, may reboot innovation, not that the regulators would allow it.

 

[TheHerald]

So what we can take away from this review is that the Ultra 9 is actually an Ultra 7 with power limitations removed, minor clock boosts and more expensive pricing. Intel no longer the leading CPU manufacturer, handed than crown to AMD again!

Maybe Qualcomm should buy them, may reboot innovation, not that the regulators would allow it.

Yeap, intel handed the crown of the most inefficient battery draining laptop cpu to amd

 

[bit_user]

Yeap, intel handed the crown of the most inefficient battery draining laptop cpu to amd

AMD's Strix Point is more efficient on MT, as we just discussed. Ranging from 11.3% to 30.6% more efficient, depending on whether they're running in stock or low-power configurations.

See:

https://www.notebookcheck.net/Intel...its-everyday-efficiency-is-good.893405.0.html
​

Yes, Lunar Lake has better single-threaded efficiency and lower idle power. However, as we've been through many times, efficiency is a multi-faceted subject and AMD retains its usual lead on multi-threaded efficiency.

 

[YSCCC]

So what we can take away from this review is that the Ultra 9 is actually an Ultra 7 with power limitations removed, minor clock boosts and more expensive pricing. Intel no longer the leading CPU manufacturer, handed than crown to AMD again!

Maybe Qualcomm should buy them, may reboot innovation, not that the regulators would allow it.

What I am wondering is what actually cause that performance difference in the article, if I have to guess would be some kind of protection like CEP in RPL is doing something so while power draw is higher, the performance goes down.

But either way, this took away the expectation of ultra 9 being significantly faster than 7, it sounds like what a k vs ks is doing. Since I am on a phone and didn’t watch the graphs in depth, would be interested to see if the voltage supply have some funny behaviour also

 

[TheHerald]

AMD's Strix Point is more efficient on MT, as we just discussed. Ranging from 11.3% to 30.6% more efficient, depending on whether they're running in stock or low-power configurations.

See:

​

https://www.notebookcheck.net/Intel...its-everyday-efficiency-is-good.893405.0.html​

​

Yes, Lunar Lake has better single-threaded efficiency and lower idle power. However, as we've been through many times, efficiency is a multi-faceted subject and AMD retains its usual lead on multi-threaded efficiency.

There is a whole paragraph in your link testing everyday use case on these cpu's. Lunar Lake absolutely nails strix point on the cross. It's not competitive. Nobody cares about cinebench on a loop on thin and light laptops but about baterry life on normal everyday workloads. Strix points needs 60% more power.

 

[Elusive Ruse]

AMD's Strix Point is more efficient on MT, as we just discussed. Ranging from 11.3% to 30.6% more efficient, depending on whether they're running in stock or low-power configurations.

See:

​

https://www.notebookcheck.net/Intel...its-everyday-efficiency-is-good.893405.0.html​

​

Yes, Lunar Lake has better single-threaded efficiency and lower idle power. However, as we've been through many times, efficiency is a multi-faceted subject and AMD retains its usual lead on multi-threaded efficiency.

Agreed, I do like myself a powerhouse on the go, single thread performance and efficiency is usually not at the top of my list.

 

[Roland Of Gilead]

So the new 2xx core ultras have both less P cores AND less E cores? That makes sense. And while we are at it lets take out HT too...

To add even more sense, they've the E cores running faster than the P cores. Surely, if the P cores are running at max speed, then there will be better results (task dependant)? Or is this more to do with the thermal envelope in laptops? Although Paul said the temps didn't go above 80c. But of a mixed bag.

 

[TheHerald]

Agreed, I do like myself a powerhouse on the go, single thread performance and efficiency is usually not at the top of my list.

Then you are not the target group for thin and light laptops. Those have small batteries (50 to 70wh) and not a lot of airflow + no dGPU, they are not meant to be running blender 247.

 

[ingtar33]

Appreciate you guys' work on all this, but it comes out so confusing. It's apparent that Intel themselves are massively uncertain what the E-core/P-core thing is all about. I'd *thought* it was about unloading IO onto E-cores so the P-cores didn't have to get interrupted, upsetting the pipeline and clearing the cache, and all, E-cores being much cheaper to build and run, and yes, I suppose they could run at different, even faster, clock speeds being simpler and all. Yet it never occurred to me that the chip could run different cores at different speeds at the same time - though maybe phones have been doing this for twenty years, so I'm a little behind on the details of some of this stuff, LOL, so was Intel.

I also wondered just how heavy it would be for Windows to factor out what they wanted to run on an E-core versus a P-core, given different core counts of each, and base versus turbo, and yada yada. So much software development, even more than other areas of engineering, seem to consist of people, even the biggest companies, just bumbling around until they trip over something that works really well, but then six other people or projects or departments just won't accept it and keep running in other directions ...

Intel created the E cores to get AMD out of the laptop/low end desktop market in california. They pushed a bill through the california state senate demanding certain cpus become more "power efficient" and effectively banned all AMD cpu sales in california started in 2025 i think. The "E" cores are supposed to qualify intel chips for this energy efficiency, the hilarious part in all of this, is ever since intel deployed their E cores their cpus have gotten less and less efficiency, meaning the purpose of the california law is being unfulfilled because intel wrote it in such a way as to not hold intel to any energy usage standards, just a core which volts down to a certain number.

As we can see here, intel isn't even trying to pretend E cores are energy efficient anymore. they don't have to, the california law is on the books and supposedly they are in compliance. that is why you see AMD rushing to market their own efficiency core chip. they don't want to be locked out of the cali market.

 

[TheHerald]

So I checked the review thoroughly. The review that is supposed to show that lunar lake ain't that efficient after all.

It's supposed competitor (hx370) was casually drawing 70% more power on average and was also drawing up to 116% more power (LOL!!!) in some workloads. The laptop was running at way higher temperatures (over 59C, youll basically burn yourself touching it) cause it couldn't cope with that amount of heat produced by that very efficient cpu, while the lunar lake laptop was chilling at 46C. Even though the amd chip was on a bigger and heavier laptop it still had to run the fans at 10x (10 freaking times, lol) louder to deal with that insane amount of heat produced.

Yeah, im convinced, lunar lake ain't very efficient

 

[JRStern]

Intel created the E cores to get AMD out of the laptop/low end desktop market in california.

wtf, I didn't know that, what a story, ROFLMAO.
See, I should have been reading TH!

 

[JRStern]

When I'm watching a large single-threaded workload in Task Manager on Windows or the System Monitor on Linux, it seems to jump around between cores (and all my computers have homogenous cores with SMT or hyperthreading), so I'm not sure what's scary if some of the cores are different types. For software developers it doesn't usually matter except for the developers working on the scheduler in Windows or Linux.

Well hush my puppies. OK then. Thanks.