Intel Core Ultra 9 275HX Defeats Core i9-14900HX In Cinebench R23 Multi-Core, While Losing Out in Single-Core

[InVasMani]

Where you saw 18% multi? It is compared to the worst of the worst examples for the previous gen with unknown CPU that can be power limited or overclocked. When compared to not gimped samples of the previous gen its barely faster in multi so it is on par with the previous gen. Maybe you should learn the meaning of "crushing"

11% MT for 2% is reasonable uplift isn't to shabby I'm not here to bicker over the use of the word crushing and how suitable or turn this into a Intel vs AMD nonsense. They made a comparison to two chips old and new and the new one is doing a sizeable amount better which if those figures generally hold consistent and true between the two chips is good to see it's progress. I'm not saying AMD can't come up with something equally compelling or better or here to debate that. I'm just looking at the figures and saying sure not bad good compromise the uplift.

 

[lexluthermiester]

18% MT

Except it's not 18%.

 

[usiname]

18% MT for 2% is reasonable uplift isn't to shabby I'm not here to bicker over the use of the word crushing and how suitable or turn this into a Intel vs AMD nonsense. They made a comparison to two chips old and new and the new one is doing a sizeable amount better which if those figures generally hold consistent and true between the two chips is good to see it's progress. I'm not saying AMD can't come up with something equally compelling or better or here to debate that. I'm just looking at the figures and saying sure not bad good compromise the uplift.

I ask you again, because you obviously have no idea how the things works, how 1% faster than 2 year old CPU is good?

 

[Contra]

The median for the 14900HX is ~30k, thats what I used....

This looks like an attempt to make a burger out of nothing.
The median value is good to compare with the median value, which we do not know, and we also do not know in what part of the spectrum both samples tested by them are. Therefore, here is a little fantasy on your part. We need to compare either the values in the picture and there is no talk of any 18% for them, or wait for reviews and a large-scale release for median values. Otherwise, this smells like clickbait

 

[InVasMani]

Except it's not 18%.

Was merely quoting the article on that %, but 2% was rounded. Also I was only comparing the two CPU's in the article and quoted uplift. It's actually worse than it was first presented though for a number of factors when you dig into it further incorrect figures and being apparently virtually the same as 2 year old CPU from Intel really isn't progress. Anyway I didn't verify the maths, but at least I spelled razberrie correkt!

 

[lexluthermiester]

Was merely quoting the article on that %, but 2% was rounded.

No, it was wrong altogether, as detailed in an earlier post. The actual percentage is 11.38%.
From post #20;

Because they aren't doing the math properly. This is basic 5th grade math people.

Percentage Increase Calculator

Percentage increase calculator finds the increase from one value to another as a percentage of the first value. Shows you how to find percentage increase with percent increase formula.

www.calculatorsoup.com

To the find the percent increase, first subtract the initial value from the final value. Then take the difference and divide it by the initial value. Finally, multiply this number by 100 to convert the number to a percentage. This final result will represent the percent increase between the two values. So;

35481 - 31854 = 3627
3627 / 31854 = 0.1138632511
0.1138632511 * 100 = 11.38632511

The difference is 11.38% aprox.

EDIT:
All that said. 11.3% is a solid increase gen on gen.

 

[InVasMani]

No, it was wrong altogether, as detailed in an earlier post. The actual percentage is 11.38%.
From post #20;

Yeah I noticed that detail after the fact oh well. In any case relative to the 13900H it's like error of margin. So this whole topic thread was in essence error of margin bickering about %'s and no real performance uplift of a incorrect article. We did get some sage wisdom about crushing it though error of margin certainly isn't. As for what qualifies as crushing it I'm going with <= 42% because that's what the guide has informed is the answer.

 

[lexluthermiester]

Yeah I noticed that detail after the fact oh well. In any case relative to the 13900H it's like error of margin. So this whole topic thread was in essence error of margin bickering about %'s and no real performance uplift of a incorrect article. We did get some sage wisdom about crushing it though error of margin certainly isn't. As for what qualifies as crushing it I'm going with <= 42% because that's what the guide has informed is the answer.

11.3% is still a decent uptick from the last gen. The 13th gen thing is like from the hyper-threading. It does show that Intel is able to recover from the performance lost from the removal of HT, which I see as an good thing overall.

 

[b1k3rdude]

So greate at synthetic benchmarks, but crap everywhere else, where have we seen this before...

 

[Clawedge]

Performance per watt?

 

[GGforever]

Except it's not 18%.

Well, I'd still say its best to stick to the median value because I am 90% sure the scores will be much better towards launch.
Saying this because I 100% expect the single-core score to improve (considering previous leaks).

 

[dyonoctis]

I ask you again, because you obviously have no idea how the things works, how 1% faster than 2 year old CPU is good?
View attachment 383996

RPL is a thorn in intel side tbh.
-It wasn’t supposed to exist was meant to be a stop gap for MTL.
- It spend an additional year being a stop gap
- Suffered from being a scrambled together arch with degradation.
- The new arch are now suffering from the comparisons because RPL was still a fast cpu despite being a last minute release. ADL> MTL>ARL was the original plan. But Intel took too long to swallow their pride and go external

 

[lexluthermiester]

Well, I'd still say its best to stick to the median value because I am 90% sure the scores will be much better towards launch.
Saying this because I 100% expect the single-core score to improve (considering previous leaks).

And those are fair points. The benchmarks and reviews will tell the tale.

 

[TumbleGeorge]

The main problem is not in a 1-2-3-5% difference in the calculations, but in using big words to drag out readings when it is a negligible difference. Negligible! This is the right term and the attempts to justify the "achievement" so as not to ruin the advertising impact are tragicomic.

 

[Daven]

35481 - 31854 = 3627
3627 / 31854 = 0.1138632511
0.1138632511 * 100 = 11.38632511

The difference is 11.38% aprox.

I like to just divide the high number by the low number and visually pull out the answer in one step:

35481 / 31854 = 1.11386325
11.39%

 

[Tek-Check]

"Crushes" with score barely faster than 7945hx and 7% faster than Raptor Lake (non refresh). When 9945hx come and wipe the floor with it, will we have article 9945hx crushes everything? I doubt
View attachment 383973

It's Zen5 - 9955HX ;-)

https://www.amd.com/en/products/processors/laptop/ryzen/9000-series/amd-ryzen-9-9955hx.html

I've described how I've calculated the difference. Next, I've just used the Microsoft Copilot to verify if my way is wrong. Here is a log of my "conversation" with Microsoft Copilot.

Gosh... it's not that complicated. No need for AI tool... There are two ways of comparing two things, point of view 1 and 2 for products A and B

Example: Apples are $2. Bananas are $3.
1. Bananas are 50% more expensive than Apples - math 3/2=1.5; Bananas' price is 1.5 times higher than Apples' price, therefore 50% more expensive
2. Apples are 33% cheaper than Bananas - math 2/3=0.67; Apples' price is 67% of the Bananas' price, therefore it's 33% cheaper (67%+33%=100%)

Numbers depend on comparison reference point:
- from the point of view of 275HX/Banana (faster/pricier), it will appear with higher percentage number in performance/price gain in comparison to something slower/cheaper that has '100%' value
- from the point of view of 14900HX/Apples, it will appear with lower percentage number in missing performance/price in comparison to something faster/pricier that is now a new '100%' value

Therefore, this benchmark:
1. 275HX is 11.3% faster than 14900HX (here, 'the 100%' sits with old 14900HX)
2. 14900HX is 10.2% slower than 275HX (here, 'the 100%' sits with new 275HX)
Hence difference in percentage numbers. It depends on which item is a reference point of '100%'.

So, 100 is 25% faster than 80 <=> 80 is 20% slower than 100 i.e. simple math

This.

 

[Vya Domus]

Um...

Yup..

I think you might have misread..

?

 

[tpuuser256]

Most LLMs are sh*t at calculating exact values. They're much better at explaining concepts and processes teachers do a terrible job at.

 

[Tek-Check]

A 14900HX can reach 35,000 points in R23. There is lots of room to go beyond that if you use ThrottleStop to do some undervolting.

Sure, but tuning is not a good reference point. Imagine if Intel officially advertised this CPUs in this way... It's a no go.

the result indicates that the 275HX handily outperforms the Core i9-14900HX in multicore performance by a whopping 18%

Where did you get 18% from? Please correct the article. 35,481/31,854=1.113 x100=11.3, so it's 11% and not 18%.
It's neither 18% nor it's 'whopping'. Tame the language of the narrative, please, to keep high standards of tech reporting.

 

[Vya Domus]

This is getting stupid

35481 - 31854 = 3627
3627 / 31854 = 0.1138632511
0.1138632511 * 100 = 11.38632511

The difference is 11.38% aprox.

Numbers depend on comparison reference point:
- from the point of view of 275HX/Banana (faster/pricier), it will appear with higher percentage number in performance/price gain in comparison to something slower/cheaper that has '100%' value
- from the point of view of 14900HX/Apples, it will appear with lower percentage number in missing performance/price in comparison to something faster/pricier that is now a new '100%' value

Therefore, this benchmark:
1. 275HX is 12.5% faster than 14900HX (here, 'the 100%' sits with old 14900HX)
2. 14900HX is 10.22% slower than 275HX (here, 'the 100%' sits with new 275HX)
Hence difference in percentage numbers. It depends on which item is a reference point of '100%'.

There is no need for all for all this, you just do 35481 / 31854 = 1.11386325108, then you read what's after the decimal point. In my first comment I only did 35 / 31 in the back of my head which is 1.12...something, that's why I said 12% as a rough estimation. Jesus Christ.

 

[Tek-Check]

All that said. 11.3% is a solid increase gen on gen.

Sure, but where does the article take 18% from?
Also, we need to see the power usage and scaling, as Arrow Lake HX has higher turbo of 160W than Raptor Lake-R. That's not good for starters.
Arrow Lake HX was supposed to fix the snappiness of CPU at lower power usage. I really hope they were able to achieve this.

A 18% MT uplift for a 2% ST drop off is a very good compromise

It's not 18%. It's 11%. The article author made a mistake in calculation.

- there is a good reason why Intel does not name their Arrow Lake i9 CPUs with number 9 this generation.
- classic i9 should have '9' in processor number. There are no 290K/295K or 290/295
- likewise, mobility CPUs adopt the same, without number 9 in the names; they have i9 275HX and 285HX, but not 295HX

Well, I'd still say its best to stick to the median value because I am 90% sure the scores will be much better towards launch.
Saying this because I 100% expect the single-core score to improve (considering previous leaks).

You cannot do this in a published article without making explicit reference in the text as to where 18% is coming from. It's misleading because readers can easily check and divide the two scores on the screenshot 35,481 with 31,854 to get 11.3% and not 18%. Please correct this.

Also, deriving 18% by dividing one single score of 275HX from one test laptop with median score of many 14900HX from many laptops from Notebookcheck is a really bad extrapolation. You don't even mention this in the article. Why? If you want to compare a single test score with median score, then you need to publish the screenshot of those median scores as a main part of the article and say something about it.

The only fair solution is to correct the face value number and edit to 11.3%. If you already want to play the game predictions and still mention 18%, on the top of 11.3%, then you need to mention this in the narrative, as stated above.

 

[lexluthermiester]

Sure, but where does the article take 18% from?

No idea. Math mistake is my guess.

I really hope they were able to achieve this.

If not, I think they got close. The reviews will show us that data.

 

[marios15]

11% Vs 18% is a huge difference, 11% could be sample variance while 18% is more solid
To keep in spirit with the math in the thread, you are overestimating the gains by 70%

We don't know if the same power target is kept, what if the extra 11% is from increased TDP???

Also what @Tek-Check is saying, if you want to show a 18% difference, show the numbers that make the difference

All I see is 11% in that screenshot from the reviewer/leaker

 

[Tek-Check]

11% could be sample variance while 18% is more solid

We don't know this. This 275HX sample could also be an upper range sample and its median could fall in future too, as different laptops with this chip will perform differently and will have different power budgets, design and cooling performance.

That's why we should never calculate percentages from one single sample with a median from another, as it gives us distorted, false picture, detached from specific devices.

Most LLMs are sh*t at calculating exact values. They're much better at explaining concepts and processes teachers do a terrible job at.

I had several really good teachers who transformed my life. Sorry if you did not have one.

 

[GGforever]

We don't know this. This 275HX sample could also be an upper range sample and its median could fall in future too, as different laptops with this chip will perform differently and will have different power budgets, design and cooling performance.

That's why we should never calculate percentages from one single sample with a median from another, as it gives us distorted, false picture, detached from specific devices.


I had several really good teachers who transformed my life. Sorry if you did not have one.

True, there's a lot of controversy surrounding this one. Let me clarify my reasoning once again.

We know that ARL-H/X will bring only a modest SC improvement over 14HX. The fact that this sample underperforms clearly indicates it's not a high-end variant.

Take the Scar 16 - a fairly standard gaming laptop with the 14900HX. It pulls around 30K points in R23. Could it do more? Sure, but so can the 275HX.

I still believe it's best to compare a fresh benchmark against the previous generation’s median value, simply because we don't know the exact configuration in question. These leaks should always be taken with a grain of salt, and our performance "leak" articles consistently include similar disclaimers.