Friday, February 6th 2026

Intel Arc B390 Panther Lake iGPU Impresses in Linux Performance Tests

Intel's Arc graphics solutions initially had a rough start to life on the Linux side of things, but it seems as though the tide is changing. Phoronix recently tested the new Panther Lake Intel Arc B390 iGPU in the Intel Core Ultra X7 358H, comparing the new iGPU to the venerable Radeon 890M in the AMD Ryzen AI 9 HX 370, and Intel's iGPU consistently outpaces the Radeon iGPU, which isn't a surprise if you've been keeping track of early benchmarks on Windows. The publication tested the new Xe3 integrated graphics in an MSI Prestige 14 running the latest Mesa 26.0 drivers on Linux Kernel version 6.19 on Ubuntu 26.04.

When it comes to gaming, the only games in which the Xe3 iGPU was bested by the AMD competition were Counterstrike 2, running at 1920×1200, and Quake II RTX. In every other game, the Intel Arc B390 beat both the AMD competition and the Intel Core Ultra 7 155H and Core Ultra 7 258V significantly. Depending on the game and in-game settings, the Arc B390 was anywhere between 15% and upwards of 50%. Hitman 3 was a particularly impressive outlier, with the B390 managing to get over 50% more FPS than the nearest competitor and 64% more FPS than the Radeon 890M at 1920×1200 and low settings, with similar gains at medium quality settings, and a slightly less impressive performance improvement when bumping up to ultra settings.
A similar story played out when it came to gaming benchmarks, with the Arc B390 coming out on top in 3DMark Wild Life Extreme—scoring 20% better than the Radeon 890M—and 3DMark Solar Bay, which was a closer victory, and GravityMark, where the Arc B390 once again dominated in OpenGL, Vulkan, and Vulkan Ray Tracing benchmarks. The Arc iGPU also performed admirably in Unigine Superposition, Valley, and Heaven, where it was never less than 30% ahead of the competition. Curiously, the AMD Radeon 890M still traded blows with the Arc B390 in terms of efficiency, even if raw performance wasn't quite there for Team Red.
Source: Phoronix
Add your own comment

11 Comments on Intel Arc B390 Panther Lake iGPU Impresses in Linux Performance Tests

#1
Cheeseball
Not a Potato
Interesting about the Quake II RTX result, because Intel's current graphics arch is better at ray tracing than what's in RDNA 3/3.5, so weird that B390 didn't beat 890M there. Did the raw 32 ROPs in the 890M help push rasterization in both CS2 and Q2RTX beyond the B390's 24 ROPs perhaps?
Posted on Reply
#2
Visible Noise
Cpt.Janksurprisingly, Intel's iGPU consistently outpaces the Radeon iGPU, which isn't a surprise
Ok, so which is it?
Posted on Reply
#3
Battler624
CheeseballInteresting about the Quake II RTX result, because Intel's current graphics arch is better at ray tracing than what's in RDNA 3/3.5, so weird that B390 didn't beat 890M there. Did the raw 32 ROPs in the 890M help push rasterization in both CS2 and Q2RTX beyond the B390's 24 ROPs perhaps?
This is mostly due to drivers.
Posted on Reply
#4
londiste
I had to look way WAY too close to find out what the memory systems on these were. I could only see this on their stupid specs comparison image that seems to be broken for me as it is not an image clickable to see a larger version. Memory hierarchy and bandwidth have been a limiting factor on iGPUs for a while now and to a notable degree still are so that is important. LPDDR5 on all but 8533MT/s for 358V/258V vs 7500 MT/s for HX370 and 6400MT/s for 155H. A good 13% more memory bandwidth is not a negligible factor in the results.
Posted on Reply
#5
Space Lynx
Astronaut
It's just too expensive. So Dead on Arrival for me.
Posted on Reply
#6
Tek-Check
Cpt.JankCuriously, the AMD Radeon 890M still traded blows with the Arc B390 in terms of efficiency, even if raw performance wasn't quite there for Team Red.
iGPU
So, performance-per-Watt was similar when Panther Lake was tested in Performance mode, which is not out-of-the-box experience. Panther Lake laptops come with Balanced mode by default, in order for Intel to sell the claim of longer battery life. Phoronix did not publish gaming tests in Balanced mode, as buyers would experience it. Had he done this, we would have seen less impressive gaming uplift due to Balanced mode running the CPU at 500 MHz lower clocks.

CPU
Phoronix corrected this mistake in CPU performance teats and published graphs in both modes. Benchmarks and narrative change significantly when tested in Balance and then in Performance mode. In Balanced mode (out-of-the-box, 4.3 GHz max), Strix Point 370 is ~13% faster on average than X7 358H in MSI Prestige, but less efficient. In Performance mode, Panther Lake is only 5% faster than Strix 370 and it reaches the rated 4.8 GHz, but it loses its efficiency. This is 18 months after Strix Point came out...
www.phoronix.com/review/intel-panther-lake-linux-power/9



To be more efficient, X7 358H needs to run in Balanced mode and lower clocks, which hurts performance. To be more performant, it needs to run in Performance mode with higher clocks, which hurts efficiency and seriously shortens battery life. The difference between the modes is 500 MHz on this machine.

It looks like Intel Panther Lake cannot have its cake and eat it. Buyers will only experience long battery life with crippled CPU and iGPU performance in Balanced mode, which is out-of-the-box experience.
Posted on Reply
#7
demonicmaniac
The issue here as they stated is that MSI sets lower than intel suggested PL1/PL2 for that specific laptop so neither is comparable to any other 358H either which runs at 28/64 for balanced. Even in balanced the b390 is much faster than the 890m tho - the only issues remaining is intel driver generally still having more problems in games compared to more mature radeon. The battle will be the non X small igpu models vs strix halo, the b390 competes with 8040s or 8050s smallest strix halo max for gpu performance.
Posted on Reply
#8
Tek-Check
demonicmaniacThe issue here as they stated is that MSI sets lower than intel suggested PL1/PL2 for that specific laptop so neither is comparable to any other 358H either which runs at 28/64 for balanced.
OEMs are allowed to tinker with power limits to hit advertised battery life for intended use. At the end of the day, MSI Prestige is a real implementation of Panther Lake and real consumers will use it with the said parameters. They will not enjoy what Panther Lake could do in theory or on other devices.

Also, gaming laptops tend to have longer Tau values than ultrabooks, up to approximately one minute of running in PL2 state before reverting to PL1 sustained. I would like to see how gaming benchmarks differ between Tau period and post-Tau period. To show real world gaming performance, reviewers will need to have in mind sustained testing in PL1 state, as this is how users will experience games, rather than seeing high fps number from PL2 state benchmark in the first minute of running tests.
Posted on Reply
#9
bug
Visible NoiseOk, so which is it?
Obviously both. It's a quantum world, you know...
Posted on Reply
#10
close
Visible NoiseOk, so which is it?
Well if you expect to be surprised, it's no surprise that it's surprising.
Posted on Reply
#11
demonicmaniac
Tek-CheckOEMs are allowed to tinker with power limits to hit advertised battery life for intended use. At the end of the day, MSI Prestige is a real implementation of Panther Lake and real consumers will use it with the said parameters. They will not enjoy what Panther Lake could do in theory or on other devices.

Also, gaming laptops tend to have longer Tau values than ultrabooks, up to approximately one minute of running in PL2 state before reverting to PL1 sustained. I would like to see how gaming benchmarks differ between Tau period and post-Tau period. To show real world gaming performance, reviewers will need to have in mind sustained testing in PL1 state, as this is how users will experience games, rather than seeing high fps number from PL2 state benchmark in the first minute of running tests.
I fully agree with all you said, I just wanted to point out in general that any test where OEMs tinker with the PL will affect performance so it's hard to talk about Panther Lake as a whole when you have ram and pl and tau affecting results. It's also why I hate that notebookcheck still does not have a sustained cb2024 power and performance test in a loop to showcase where the snapdragons "cheat" by running at 70w for most of the duration of a cb2024 run with their tau before throttling hard down to 20w or so. I recognize Tau and Turbo being valid for bursty workloads like website loading and launching a programm but it sure has become a war with making Tau so that short benchmarks are won :/
Posted on Reply
Add your own comment
Nov 22nd, 2026 17:16 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts