Wednesday, December 10th 2025

Upcoming Flagship "Panther Lake" APU Geekbenches Above AMD's Best Ryzen AI Mobile Chip

The upcoming "Panther Lake" mobile processor series seems to consist of (at least) fourteen distinct SKUs, according to a major mid-October and November leaks. Since then, a bunch of these elaborately named models have been fleshed out in greater detail—courtesy of benchmark data appearing across online repositories. The supposed flagship option—"Core Ultra X9 388H"—has appeared again; courtesy of a mysterious "Default string Default string" test system posting results and basic specifications onto the Geekbench Browser database. Almost two months ago, the not-yet-launched Core Ultra X9 388H chip's integrated graphics solution seemed to do quite well in 3DMark Time Spy tests.

Looking at intriguing Geekbench 6.5 info, a 16-core CPU design—hosted on an "NM14PTL" board—scored (overall) 3057 points in single-core stakes, and 17687 points in multicore scenarios. These tallies seem to place the next-gen "flagship" above the "Arrow Lake-H" Core Ultra 9 285H APU, with an approximate 15 to 19% performance advantage over the "direct predecessor." The leaked Core Ultra X9 388H candidate seems to perform in the region of Team Blue's beefier "Arrow Lake-HX" Core Ultra 9 275HX model. Looking at current-gen competition, AMD's "Strix Halo" range topper—Ryzen AI Max+ 395—about matches the forthcoming "Panther Lake" flagship in terms of multicore, but Team Blue's futuristic design seems to win in the realm of single core performance. Additional "Core Ultra X9 388H" fundamentals include a measured 5.078 MHz maximum frequency, 4.0 GHz base frequency, and 18 MB of L3 cache. Cluster 1 consists of four cores, presumably "Cougar Cove" Power units. Cluster 2 sports twelve cores in total; official slides designate "Darkmont" architecture across eight E-cores and four LP-E cores.
Sources: Geekbench Browser, CPU Monkey Tweet, Notebookcheck, Tom's Hardware, VideoCardz
Add your own comment

16 Comments on Upcoming Flagship "Panther Lake" APU Geekbenches Above AMD's Best Ryzen AI Mobile Chip

#1
Antique4106
so matching the performance of a nearly 1-year old cpu, nice
Posted on Reply
#2
ToTTenTranz
Antique4106so matching the performance of a nearly 1-year old cpu, nice
Strix Halo is a much bigger, more expensive, more power hungry SoC with twice the memory channels.


Reaching Strix Halo's multithreaded score with a monolithic 4P + 8E + 4 LPE solution is actually a big step up.
Posted on Reply
#3
R0H1T
ToTTenTranzStrix Halo is a much bigger, more expensive, more power hungry SoC with twice the memory channels.


Reaching Strix Halo's multithreaded score with a monolithic 4P + 8E + 4 LPE solution is actually a big step up.
Did you forget Halo is a monolithic die as well, also a node(half?) behind?

Also why does "memory channel" come anywhere near this equation wrt Geekbench?
Posted on Reply
#4
ToTTenTranz
R0H1TDid you forget Halo is a monolithic die as well, also a node(half?) behind?
No, Strix Halo is always a multi-chip module. In the case of Ryzen MAX 395 it's 3 chips, 1x IOD + 2x CCD.

R0H1TAlso why does "memory channel" come anywhere near this equation wrt Geekbench?
Because effective memory bandwidth influences the Geekbench multicore score a lot, especially for higher core counts.



Strix Halo has 256GB/s from the RAM whereas Panther Lake has 137GB/s.
Posted on Reply
#5
R0H1T
Ok, looks like you're right about MCM part but that chart doesn't prove much. You're basically going nowhere from 3600 MT/s to 4800 MT/s although that's a much older benchmark.
Posted on Reply
#6
ToTTenTranz
R0H1TYou're basically going nowhere from 3600 MT/s to 4800 MT/s
Now do 3600 vs. 6400MHz (1.77x higher bandwidth) which is closer to the 256bit STX-Halo vs 128bit PTL-H difference (1.87x higher bandwidth).
Posted on Reply
#7
R0H1T
Alright but comparing performance across different chips would never make sense when you're talking about GB scaling (performance) due to higher bandwidth ~ it has to be the exact same chip & various capacity/speeds for RAM.
Posted on Reply
#8
Brijoshoo
So am I reading this right - there is not enough PCIe lanes to run a GPU at full speed PCI-E 5.0 16x?
Posted on Reply
#9
Cheeseball
Not a Potato
BrijoshooSo am I reading this right - there is not enough PCIe lanes to run a GPU at full speed PCI-E 5.0 16x?
AFAIK, none of the mobile (255H/285H/258V) CPUs/APUs have enough PCI-E 5.0 lanes for x16. Its only the HX (275HX, 285HX) ones due them being power-limited desktop variants (along with only having 4 Xe-cores for graphics).

255H:



285HX:


AMD's HX 370 has x16 PCI-E 4.0 lanes, but you would expect 4 of those dedicated for NVMe SSD usage:


9950HX is as expected, since its basically a power-limited desktop 9950X:
Posted on Reply
#10
Visible Noise
BrijoshooSo am I reading this right - there is not enough PCIe lanes to run a GPU at full speed PCI-E 5.0 16x?
In a laptop???
Posted on Reply
#11
Squared
BrijoshooSo am I reading this right - there is not enough PCIe lanes to run a GPU at full speed PCI-E 5.0 16x?
CheeseballAFAIK, none of the mobile (255H/285H/258V) CPUs/APUs have enough PCI-E 5.0 lanes for x16. Its only the HX (275HX, 285HX) ones due them being power-limited desktop variants (along with only having 4 Xe-cores for graphics).

255H:



285HX:


AMD's HX 370 has x16 PCI-E 4.0 lanes, but you would expect 4 of those dedicated for NVMe SSD usage:


9950HX is as expected, since its basically a power-limited desktop 9950X:
It's also true that the Core X variants only have 4 PCIe 5.0 lanes. But the bigger CPU core variant with the smaller iGPU has 12 lanes (for a GPU and SSD) as a slide in the same article shows.
Posted on Reply
#12
dyonoctis
R0H1TDid you forget Halo is a monolithic die as well, also a node(half?) behind?

Also why does "memory channel" come anywhere near this equation wrt Geekbench?
I mean, that's a (16) 4+8+4 CPU competing against a "real" 16-core CPU. Over half of that CPU is made of little cores
Posted on Reply
#13
Cheeseball
Not a Potato
SquaredIt's also true that the Core X variants only have 4 PCIe 5.0 lanes. But the bigger CPU core variant with the smaller iGPU has 12 lanes (for a GPU and SSD) as a slide in the same article shows.
I believe the Core X's are still technically the H-variants since its X9 3xxH but they most likely cut down the built-in iGPU to free up more PCI-E lanes for gaming/workstation laptops with a dGPU, unless they don't plan to make a high-end HX models (based off a desktop CPU, usually for 16"+ desktop replacements) this time around.

Something like how some AMD laptops still pairs HX 370 with a dGPU despite only having x8 lanes dedicated to it (with x4 for NVMe).
Posted on Reply
#14
ToTTenTranz
BrijoshooSo am I reading this right - there is not enough PCIe lanes to run a GPU at full speed PCI-E 5.0 16x?
Chips designed for laptops haven't had 16x lanes for dGPUs at the latest PCIe gen for a while.
Truth be told, thin and light laptops wouldn't be able to pair with GPUs big/performant enough where 8x is a bottleneck, and desktop replacements use power-limited desktop CPUs anyways.
SquaredIt's also true that the Core X variants only have 4 PCIe 5.0 lanes. But the bigger CPU core variant with the smaller iGPU has 12 lanes (for a GPU and SSD) as a slide in the same article shows.
It makes sense that Intel doesn't really want OEMs to pair their big iGPU SKUs with Nvidia dGPUs. Doing so would probably dilute their marketing efforts for propping up ARC iGPUs.
Posted on Reply
#15
Squared
ToTTenTranzIt makes sense that Intel doesn't really want OEMs to pair their big iGPU SKUs with Nvidia dGPUs. Doing so would probably dilute their marketing efforts for propping up ARC iGPUs.
Wouldn't it also be pointless? I've never had a graphics switching laptop so I might be missing something, but I would expect automatic graphics switching to kick before a workload reaches the point where it would benefit from the 8 extra Xe3 cores on the X models. I've always been a little annoyed while looking for laptops with big iGPUs that a lot of them also have a big dGPU and as a shopper I'm only looking for one or the other.

I also think the Steam Deck and Lunar Lake chips were great because they're the only small CPU big iGPU chips which is ideal for handhelds (from a cost-to-performance and size-to-performance perspective).
Posted on Reply
#16
ToTTenTranz
SquaredWouldn't it also be pointless?
Yes, but ⬇️
SquaredI've always been a little annoyed while looking for laptops with big iGPUs that a lot of them also have a big dGPU and as a shopper I'm only looking for one or the other.
This.

OEMs just can't help themselves putting a nvidia GPU everywhere. I think some OEMs even took the hilarious decision of pairing a Ryzen 7 8840U with a slower RTX 2050 dGPU.
Posted on Reply
Add your own comment
Nov 19th, 2026 19:55 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts