Intel really doesn't have a great product range for AI, at least compared to NVIDIA and AMD.
Their server CPUs are quite competitive and relevant for agentic stuff, but AMD CPUs are arguably better and every hyperscaler (and NVIDIA) has their own ARM based CPU nowadays.
On the GPU/AI accelerator side, they have basically nothing. Gaudi is dead, all the Ponte * accelerators are heavily delayed, Xe based GPUs are not competitive.
On client (think gaming and office PCs), memory has gotten so expensive that nobody buys new machines right now.
I doubt it'll be enough to carry the company but I really hope Crescent Lake works well for them. A 120-480GB bigger Xe3p GPU is not going to be a speed demon but hopefully gives Strix Halo and Mac M5 Ultra a very good run for the money.
Again, not going to be "the" product to turn things around. But a very smart offering that no one else is doing, if they can price it ok. Which is barely in their control, given that it's once again overwhelmingly ram (by cost share).
Such a pity. Panther Lake and Wildcat are both really excellent chips. What a bad run of timing for an Intel that finally seems arising.
> fixing corrupted input handling in a C/C++ code base would not be too hard
The best programmers on the planet have tried and failed with this task for 50 years now, so I don't think this is true.
The main disadvantage of Rust right now is not supporting some more obscure platforms, but because mold wouldn't support them anyway I don't see that as a problem.
> The main disadvantage of Rust right now is not supporting some more obscure platforms
Last time I checked, I got impressed by the wide platform support, once you go down the tier list (https://doc.rust-lang.org/nightly/rustc/platform-support.htm...). What "obscure platform" specifically are you thinking about, that is currently missing from those lists?
LLVM doesn't have an equivalent annotated list, but it is missing alpha, bfin, c6x, fr30, frv, gcn, h8300, ia64 (aka Itanium), lm32, m32c, m32r, mcore, mep, microblaze, mmix, mn10300, moxie, nds32, nios2, pa (aka PA-RISC), pdp11, pru, rl78, rs6000, rx, sh (aka SuperH), storm16, v850, vax, and visium. For its part, LLVM does have some targets that GCC doesn't have (mostly related to GPU compilation).
The "big" targets that GCC has that LLVM lacks are Alpha, Itanium, PA-RISC, and SuperH, with Itanium being sufficiently weird that it's pretty firmly in the "fuck this" category from a maintainer's perspective, and people are actively ripping out support for it.
Just to be clear, I think this is a very small disadvantage. GCC and therefore C/C++ supports some old stuff like SuperH, Intel Itanium, PA-RISC and a bunch of microcontroller archs that LLVM does not.
However, there is now a Rust codegen plugin for GCC, so even this disadvantage is now basically moot.
I think the claim is that GCC supports a few obscure platforms that LLVM doesn't. Don't ask me which, that is just the claim that I heard multiple times. So it's not Rust that limits platforms, but LLVM. And they say the GCC backend efforts of Rust are meant to deal with that.
There are many. Two big ones are Alpha and PA-RISC. NetBSD and Linux continue to support both. Linux distro choices are pretty much limited to Gentoo though.
1. You've got the causality backwards. LLVM backends don't come for free, you have to write code to enable support for them.
2. LLVM does not support as many backends as GCC does, so even if you did get 100% of the LLVM supported backends up and running, you'd still be missing some.
Worth adding to this: there may also be code needed in the rust part (not just the llvm part) to enable support for a target triplet. There certainly will be for the OS and libc part of the triplet: what specific syscalls exist and should be used to implement the standard library on a particular flavour of BSD etc.
But even for the architecture part there can be differences. For example, my understanding is that a lot of the calling convention details need to be handled by the frontend in LLVM, leading rustc to duplicate logic from clang here. Things like varargs FFI with C code can be particularly gnarly.
I am also running some old AMD datacenter cards, 2x MI25 in my case. Getting around 30 tokens/second with short context.
Tensor parallel in llama.cpp using RCCL (disabled by default in llama.cpp for some reason). Surprisingly, for these cards HIP is actually faster than Vulkan, unlike the 9070 XT where Vulkan still wins.
ROCm nightlies do actually support these old cards, just not the ROCm stable releases.
Their server CPUs are quite competitive and relevant for agentic stuff, but AMD CPUs are arguably better and every hyperscaler (and NVIDIA) has their own ARM based CPU nowadays.
On the GPU/AI accelerator side, they have basically nothing. Gaudi is dead, all the Ponte * accelerators are heavily delayed, Xe based GPUs are not competitive.
On client (think gaming and office PCs), memory has gotten so expensive that nobody buys new machines right now.