Bionic_Squash

@Haze2K1 I wonder if it has the old intel IGPU detected message box that pops up when running the game. Pretty weird NGL.

Crimson Desert currently doesn't support Intel Arc GPUs, per their FAQ crimsondesert.pearlabyss.com/en-US/News/Not… is this the first time a game has blatantly said they don't support an entire brand of GPUs? lol

The @Intel Compute Runtime 26.09.37435.1 Brings More Features & @IntelGraphics Performance Optimizations phoronix.com/news/Intel-Com…

QuickLogic Announces Contract for High Density eFPGA Hard IP Optimized for Intel 18A (...) its eFPGA Hard IP in a new customer ASIC based on Intel 18A technology. ir.quicklogic.com/press-releases…

Intel enables Precompiled Shader Delivery on Arc B-series and Core Ultra iGPUs in new driver, up to 3x faster loading times videocardz.com/newz/intel-ena…

@FelixCLC_ @IntelSoftware It's just different name they use in IGC for one of the FP8 varients afaik

Hey, uhhh @IntelSoftware, whats the definition of "HF8"?!!?!?!?

@mango_raw Yes some look decent (especially the leon ones) but the other grace shots scream AI filter. It's changing the face entirely or it changes the art style entirely

@SquashBionic Some shots aren't so bad.

Looks ass, it should respect the games art and design choices. It shouldn't be a "hyper realistic" filter

openVINO backend has finally been merged into llama.cpp! github.com/ggml-org/llama…

Intel & Microsoft collaborate closely on the Future of PC Gaming intel.com/content/www/us…

.@AIatMeta #MTAI Roadmap: MTAI 300: - 1x Compute - 2x Network - in production MTAI 400: - 2x Compute - 2x Network - 1x SoC-Chiplet - path to being deployed MTAI 450: - 2x Compute - 2x Network - 1x SoC - faster D2D compared to MTAI 400 - early 2027

Custom silicon is critical to scaling next-gen AI. We’re detailing the evolution of the Meta Training and Inference Accelerator (MTIA), our homegrown silicon family designed to power the next era of AI experiences. Traditional chip cycles span years, but model architectures change in months. To close this gap, we’ve accelerated MTIA development to release four generations in just two years. See our roadmap and tech specs here: go.meta.me/16336d

>it's back again >It's working now >It's 20MB bigger(~90MB, back to normal size) They definitely forgot to bundle the MTL binaries last time didn't they, also it's got the extractor thingy that launches when you open the installer. It was missing in the last one

Intel has removed the driver from their site, idk how they missed such a big issue

@pers0naluni0n Hmmm, don't know what's causing it exactly but I've seen the same issue on other 185H systems

@SquashBionic Did a clean install on my XMG Meteor Lake 155H and it worked.

newest intel NPU driver is failing to install on MTL devices (haven't checked LNL and PTL) intel.com/content/www/us…

@chucker @Sebasti66855537 @SebAaltonen Wants ≠ does It's representative in smartphones, but laptops and desktops have plenty of workloads that scale well with more threads.

@SquashBionic @Sebasti66855537 @SebAaltonen Yes, because it wants to represent _realistic_ workloads.

The first M5 Max CPU benchmarks look promising. M4 Max already had a notable ST advantage over desktop CPUs. Now the MT performance is clearly ahead of 9950X3D too. Tiny efficiency cores are gone. Now M5 Max has 6 super cores + 12 performance cores (18 total). It's a MT monster.

@chucker @Sebasti66855537 @SebAaltonen For MT? I disagree. It's by far the worst for benching MT perf. It doesn't like more threads and stops scaling at relatively low core count

@Sebasti66855537 @SebAaltonen It's still a good, perhaps the best, overview across multiple disciplines, available for common platforms.

MSVC v14.51 is up to 6.5% faster on SPEC CPU 2017 integer benchmarks from improved optimizations such as a new SLP vectorizer and major SROA improvements on both x64 and Arm64. Learn more about the performance improvements: devblogs.microsoft.com/cppblog/c-perf…

@dcominottim 3.1 pro is extremely fucked(the model has very weird quirks), and availability wise their API is in shambles. It's almost always failing in Gemini code assist and Gemini CLI

Gemini models are really, *really* weird right now. Did a Deep Research with Pro (not sure if it’s using 3.1 Pro or DeepThink behind the scenes, because the whole Gemini naming is also an absolute mess) about a medical subject and it returned the most long-winded actual slop I’ve ever read. Can’t trust even 3 sequential words from it.

@aschilling IO tile is same as GNR (reused)

So we now have the Compute Tile (Intel 18A, first wafer) and Base Tile (Intel 3, second wafer), but we are still missing the I/O Tile (Intel 7) for Clearwater Forest.