Windows 11 vs. Linux Benchmarks For Intel Arc B-Series "Battlemage" Shows Strengths & Weaknesses

[phoronix]

Phoronix: Windows 11 vs. Linux Benchmarks For Intel Arc B-Series "Battlemage" Shows Strengths & Weaknesses

Last week with the availability of the Intel Arc B-Series Battlemage graphics cards I ran benchmarks looking at the GPU compute performance, Linux gaming benchmarks, and also the workstation graphics capabilities. The Intel Arc B580 graphics showed some nice generational uplift under Linux for most workloads but there were some anomalies where clearly the Intel Linux graphics driver had room to better optimize the new Xe2/Battlemage graphics support. Windows benchmarks of the Intel Arc B580 also showed it performing more competitively to the likes of the GeForce RTX 4060 compared to what I was seeing under Linux. Thus I spent the past few days working on some Windows 11 vs. Ubuntu Linux graphics benchmarks for both Intel Arc Graphics of Alchemist and Battlemage GPUs to see how the performance compares.

https://www.phoronix.com/review/intel-arc-b580-windows-linux

[andre30correia]

tha vulkan driver anv is piece of garbage

[avis]

Strengths: real games
Weaknesses: benchmarks.

BTW, Michael, did you really use the vulkan renderer for CS2 under Windows (CS2 -> Properties -> General -> Launch Options -> -vulkan)? By default it targets D3D11 which is faster and more optimized.

[loganj]

not that great for gaming on linux unless u use opengl

[MillionToOne]

tha vulkan driver anv is piece of garbage

Couldn't agree more. Hopefully they start improving it with 2 generations of discrete cards already out.

[M@GOid]

Judging by the results of A series cards, those expecting any gaming optimizations from Intel for the B series will be left disappointed. Basically their driver only received testing in common Linux benchmarks, but not in any actual game.

[stormcrow]

Strengths: real games
Weaknesses: benchmarks.

BTW, Michael, did you really use the vulkan renderer for CS2 under Windows (CS2 -> Properties -> General -> Launch Options -> -vulkan)? By default it targets D3D11 which is faster and more optimized.

D3D11 would be the more relevant test as most games are going to use D3D by default on Windows not Vulkan or OpenGL. :P I know, you're supposed to compare apples to apples, but the point of making relevant benchmarks isn't to compare the apple to apples (generic), it's to compare a Red Delicious (yuck) to a Pink Lady (yum) specifically. By that I mean comparing the real use cases, not theoretical. When you're comparing the UX between Linux and WIndows you have to therefore compare OpenGL/Vulkan to D3D. Direct Vulkan to Vulkan Windows/Linux comparisons are only relevant if the game is using the same code pathway for both OSes by default.

Also, really really wish the Blender benchmark would stop reporting the time, and report the real benchmark output, which doesn't measure time. It reports on the number of compute units processed for a specific scene. The time element isn't really that important for what the benchmarks scenes are designed for. They're all supposed to complete within about 20-30 seconds regardless of how many units were processed. (Which might be why A770 outperformed the B580, but it's impossible to know because the benchmark isn't reporting the relevant statistics).

[sophisticles]

By now it's clear that the Arc family of cards is great for certain specific use cases.

If you want gaming, buy NVIDIA.

If you want compute, buy NVIDIA.

If you are working with video, buy NVIDIA.

If you are on a tight budget, want respectable gaming, compute and video editing capabilities, buy Intel.

If you are an even tighter budget and are content with decent gaming, compute and video editing capabilities, buy an AMD APU.

And if you are nuts, buy an AMD video card.

It's that simple.

[Quackdoc]

Would love to see more applications eventually since this doesn't show case my experience but well, I know how automated testing goes. I wonder how image recognition would work out hmmm

[blackshard]

Strengths: real games
Weaknesses: benchmarks.

What games? Those ancient games that don't tell anything about the real performances of those cards on current and past titles.