Bionic_Squash

silicon enthusiast, has broken humour #intel #AMD #Nvidia

[Longhorn @never_released]

The RTX Spark NPU software stack appears very broken right now

https://t.co/3ol6t86md5

Cc @anemll

[Xeno @XenoPanther]

Here is how the new "Advanced Performance Extensions (APX)" setting would appear on compatible hardware on 29695

[Wccftech @wccftech]

AMD FSR 4.1.2 now runs on Intel Arc XMX cores, delivering better image quality than XeSS upscaling with similar performance. 🔗 https://t.co/e9r5XjKR2C

[tsuki @tensorcore]

OzII on Xe2
https://t.co/FtFsE1w1fR

[Phoronix @phoronix]

The @Intel Core Ultra 9 Arrow Lake On Star Labs @starlabsltd StarFighter Shows Nice Gains On Ubuntu 26.10

~5% geo mean improvement over Ubuntu 26.04 LTS, real nice improvements for @IntelGraphics Arc Graphics

https://t.co/Hshejy46Sa

[Bionic_Squash @SquashBionic]

New Graph Compiler for intel GPUs!
https://t.co/avroE6oyme

[Bionic_Squash @SquashBionic]

@XYang2023 Yea I saw, the PR's are already up in the oneDNN repo

[Bionic_Squash @SquashBionic]

AMD will be upstreaming Zen IP optimisations from ZenDNN into oneDNN library to reduce redundancy.
AMD has also joined the UXL consortium as a Steering member.
https://t.co/3MVhxAibGQ

[Navi24-chan 🌠 @IridiumWings]

It's that time again?

[Haze @Haze2K1]

Intel Graphics Extensions SDK Released https://t.co/UA1UrLIO16 #opensource #github @inteldevs @intel @IntelGaming

[Bionic_Squash @SquashBionic]

@intelligenz_b @mooreslawisdead It's 96 EU's, same as PTL IGPU for EU count.
Intel does this weird shit where they count EUs using the SIMD8 way in their internal docs.

[Bionic_Squash @SquashBionic]

@intelligenz_b @mooreslawisdead It's different

[Bionic_Squash @SquashBionic]

@mooreslawisdead so this 12Xe/192EU SKU is different from RZL-AX fka NVL-AX or *is* that going to be the rebranded halo SKU?

[Bionic_Squash @SquashBionic]

@M0_R4_H4 I just found the funniest image lmao

[Bionic_Squash @SquashBionic]

Wtf is this copium

[Hassan Mujtaba @hms1193]

Exclusive: NVIDIA's Vera Team Says Chiplet Designs Sacrificed Single-Threaded Performance for Core Counts, and Agentic AI Is Now Forcing A Reversal https://t.co/2dZwVhGZ1E

https://t.co/2dZwVhGZ1E

[Bionic_Squash @SquashBionic]

@XYang2023 @jaykihn0 @realLeeMax The shores lineup is missing from roadmap entirely (this was after his tweet).
So no Rackscale for now (well you can technically do Rackscale CRI with Altera Labs switches)

[Bionic_Squash @SquashBionic]

Atleast we still have the island product line

[Bionic_Squash @SquashBionic]

@Silicon_Fly >Jaguar Shores

I got some news for you.....

[Bionic_Squash @SquashBionic]

@Silicon_Fly So 10 months ago?
Fyi it's been gone from their internal roadmap for a while, only island line-up remains.
That's why I'm assuming it's dead.
Intel pretty much has never referenced Jaguar Shores recently

[Bionic_Squash @SquashBionic]

@Dangerman1337 Island product line is still alive, that will use Xe4

[Bionic_Squash @SquashBionic]

@Dreaming_Man2 @pers0naluni0n It does have competitive memory bandwidth, idk about the other specs.
Its not been in the internal roadmap for sometime and Intel has avoided talking about it publically.
So I'm assuming it's dead.

[Bionic_Squash @SquashBionic]

@M0_R4_H4 Dawg i already said deployments usual do NUMA stuff AND the cache aware scheduling makes sure cross CCD communication is minimised.
It's even less of an issue on Intel too.

[Bionic_Squash @SquashBionic]

@M0_R4_H4 Still not addressing the claim that chiplets affect ST perf