Thursday, October 30th 2025

Leaked Intel Core Ultra X7 358H and Ultra 5 338H Cinebench R23 Scores Reveal Concerning CPU Performance

Intel's upcoming Panther Lake laptop CPU configurations previously leaked, revealing a line-up ranging from the modest Core Ultra 3 320U, with 2 P-Cores and 4 LPE-Cores to the notably higher-end Core Ultra X9 388H, featuring 4 P-Cores, 8 E-Cores, and 4 LPE-Cores. Now, performance figures for the more mid-range Intel Core Ultra X7 358H and Core Ultra 5 338H have both been leaked, thanks to Laptopreview Club. According to the site, the Core Ultra X7 in question managed a Cinebench R23 score of around 20,000, while the Core Ultra 5 338H came in at around 16,000 points or 20% slower than the X7 variant—the site doesn't provide exact scores, only rounded figures, but it's some idea of what to expect in terms of CPU performance for the next-gen Intel CPU platform.

As previously discussed, Panther Lake appears to be focusing more on iGPU performance, with leaked benchmarks revealing RTX 3050-tier performance from the 12-core Xe3 iGPU solution, but this is the first indication of CPU performance for the new CPU generation. The leaker claims that the 358H will be around 10% slower than the 255H at the same power consumption, in this case 60-65 W, with Cinebench R23 scores for the 255H ranging from the low 17,000s to upwards of 22,000. It appears, thus, that Panther Lake may sacrifice some CPU performance in order to improve iGPU performance, although these are still early rumors, and things are subject to change ahead of launch. The idea that Intel would launch a new CPU that underperforms its old platform by such a margin seems unlikely. It's possible that these tests were on the lower side of average or a worst-case scenario, either of which would actually make the performance of the new CPUs reasonably impressive.
Sources: Laptopreview Club, VideoCardz, Notebookcheck
Add your own comment

12 Comments on Leaked Intel Core Ultra X7 358H and Ultra 5 338H Cinebench R23 Scores Reveal Concerning CPU Performance

#2
LastDudeALive
Well, as the article says, there's a lot of unknowns about the bench. Also, it's difficult to compare the scores, given the different core arrangements. The 255H has 6 P cores, 8 e cores, and 2 LPE cores, while the 358H has 4 P cores, 8 e cores, and 4 LPE cores. Having 2 more threads in LPE cores instead of P cores probably makes a good difference in benchmarks.

The article also notes that the iGPU's Time Spy benchmarks are better than the RTX 2060.
Posted on Reply
#3
RejZoR
Seeing how many problems everything has with AMD's different cores (normal and X3D) where penalty is a lot smaller than when game shuffles to Intel's E cores instead of P cores, I'm sticking with single CCD stack of same cores. I was right for going that way with 5800X and 5800X3D and I went with 9800X3D specifically because of this. Scheduling on Windows is a god damn joke and relying on some chipset drivers and windows to schedule shit right is just impossible. With 9800X3D, all cores are equal and it literally doesn't matter which one it runs, each will run at its peak no matter what.

I'd frankly trust Linux scheduler way more for this so having Intel's P/E cores mixed for laptop that's mostly for productivity it would probably by far work the best. Still wouldn't rely on it for gaming...
Posted on Reply
#4
freeagent
CrAsHnBuRnXpIntel is the new (old) AMD.
Pretty crazy eh?

What is going on over there? It seems they have lost their way.
Posted on Reply
#5
holyprof
RejZoRSeeing how many problems everything has with AMD's different cores (normal and X3D) where penalty is a lot smaller than when game shuffles to Intel's E cores instead of P cores, I'm sticking with single CCD stack of same cores. I was right for going that way with 5800X and 5800X3D and I went with 9800X3D specifically because of this. Scheduling on Windows is a god damn joke and relying on some chipset drivers and windows to schedule shit right is just impossible. With 9800X3D, all cores are equal and it literally doesn't matter which one it runs, each will run at its peak no matter what.

I'd frankly trust Linux scheduler way more for this so having Intel's P/E cores mixed for laptop that's mostly for productivity it would probably by far work the best. Still wouldn't rely on it for gaming...
Yup, that's why I skipped everything from Intel and AMD and kept my trusty old 3700X, waiting to 12 core single CCD Zen 6 X3D CPU. If that rumour of 12 P-cores on a single die is true, will be worth the wait.
Posted on Reply
#6
Cheeseball
Not a Potato
What's the concern? The big difference is most likely that the Ultra X7 358H has more cores than the 5 338H.

Its like being surprised that the 7950X3D is faster than the 9800X3D in Cinebench, or that the 13900K is still faster than the 14700K.
The leaker claims that the 358H will be around 10% slower than the 255H at the same power consumption
The 255H is 6C/8e/2LPe and the 358H is rumored to be 4C/8e/4LPe so naturally it should still be able to perform better in multi-threaded workloads. This is only a single generation ahead of Alder Lake-H so I'm not expecting the 4C Panther Lake to suddenly be better than the 6C ADL-H CPU.
Posted on Reply
#7
overclockedamd
It is what it is. But dont mark Intel out of this game. Look at my forum name. They (intel) have heat in the chamber and I pray they firing. All signs point to yes.
Posted on Reply
#8
Shonendo
freeagentPretty crazy eh?

What is going on over there? It seems they have lost their way.
#MURICA
Posted on Reply
#9
BlueScreenJunky
the 358H will be around 10% slower than the 255H at the same power consumption [...] It appears, thus, that Panther Lake may sacrifice some CPU performance in order to improve iGPU performance
That doesn't really make sense to me. I do expect them to sacrifice raw CPU performance by basically allocating less space to the CPU, but why would it make it less energy efficient ?
Posted on Reply
#10
_roman_
Synthetic benchmarks do not count when the processor feels slow. Or when the application just crashes because instructions can not be executed. That does not include extra software and extra processor cycles and wasted energy to put the right tasks on the right cores. Less higher one type single cores perform better, even better with more Cache.
Posted on Reply
#11
Squared
BlueScreenJunkyThat doesn't really make sense to me. I do expect them to sacrifice raw CPU performance by basically allocating less space to the CPU, but why would it make it less energy efficient ?
The models with 12 Xe cores have the big CPU die with up to 4LPE+4P+8E cores, so nothing is sacrificed (except PCIe lanes). But I'm pretty sure 60-65W is a lot of power for a mobile CPU. P cores are designed to be more efficient than E cores at higher power. So for example if P core peak efficiency is 7 watts and E core peak efficiency is 4 watts, Panther Lake's peak efficiency with 4P+8E would be 28W+32W so 50W, and Arrow Lake (255H)'s peak with 6P+8E would be 42W+32W so 74W. Somewhere in between those levels Arrow Lake would be more efficient.

Those are made-up numbers. Whatever the numbers they'll also be affected by the manufacturing nodes, core design changes, tile changes, and LPE cores. New nodes typically struggle at higher power levels more than mature nodes. LPE cores aren't on the ring bus and I've heard Cinebench doesn't even use them on Arrow Lake. Darkmont and Cougar Cove are minor updates over Skymont and Lion Cove.
Arrow LakePanther Lake
P6 Lion Cove on CPU tile4 Cougar Cove on CPU tile
E8 Skymont on CPU tile8 Darkmont on CPU tile
LPE2 Crestmont on SoC tile.4 Darkmont on CPU tile
memory controlleron SoC tileon CPU tile
CPU tile nodeTSMC N3BIntel 18A
SoC tile nodeTSMC N6none
Posted on Reply
#12
Dsirius
SquaredThe models with 12 Xe cores have the big CPU die with up to 4LPE+4P+8E cores, so nothing is sacrificed (except PCIe lanes). But I'm pretty sure 60-65W is a lot of power for a mobile CPU. P cores are designed to be more efficient than E cores at higher power. So for example if P core peak efficiency is 7 watts and E core peak efficiency is 4 watts, Panther Lake's peak efficiency with 4P+8E would be 28W+32W so 50W, and Arrow Lake (255H)'s peak with 6P+8E would be 42W+32W so 74W. Somewhere in between those levels Arrow Lake would be more efficient.

Those are made-up numbers. Whatever the numbers they'll also be affected by the manufacturing nodes, core design changes, tile changes, and LPE cores. New nodes typically struggle at higher power levels more than mature nodes. LPE cores aren't on the ring bus and I've heard Cinebench doesn't even use them on Arrow Lake. Darkmont and Cougar Cove are minor updates over Skymont and Lion Cove.
Arrow LakePanther Lake
P6 Lion Cove on CPU tile4 Cougar Cove on CPU tile
E8 Skymont on CPU tile8 Darkmont on CPU tile
LPE2 Crestmont on SoC tile.4 Darkmont on CPU tile
memory controlleron SoC tileon CPU tile
CPU tile nodeTSMC N3BIntel 18A
SoC tile nodeTSMC N6none
Citing ...28W+32W so 50W. You meant 60W.
Posted on Reply
Add your own comment
Nov 14th, 2026 18:53 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts