Of course Intel has been designing and selling GPUs for years, I guess Lip-Bu means they're going to start manufacturing them as well? Or they're going to be data-center focused now?
Since he was touting that they recently hired a well-known GPU architect, it seems unlikely that this is merely about them using their own fabs for discrete GPUs instead of having integrated GPUs being the only ones they fab themselves. Some kind of shift in product strategy or reboot of their GPU architecture development seems more likely, if there's anything of substance underlying the news at all.
But this news is somehow even less comprehensible and believable than usual for Intel, whose announcements about their future plans have a tenuous connection to reality on a good day.
Intel has been using a fair bit of TSMC in their CPU manufacturing recently, yes. Most recently they’ve been assembling “tiles” of silicon from many process nodes into a single CPU package and IIRC they have been using TSMC for the GPU tiles.
Intel has been designing GPUs manufactured on TSMC nodes across client and datacenter for at least the past 5 years. The client chips are price competitive but not performance competitive with AMD/NVIDIA/Apple. The data center roadmap has historically been a huge mess with cancelled products left and right. But, to say "Intel will start making GPUs" seems misleading. Perhaps "Intel to try to inject sanity into its GPU roadmap" would be a better headline, though I am skeptical one hire will do anything to fix 10+ years of mismanagement.
I have a B580 in my desktop. Unfortunately AMD still has broken PCIe level reset and so their GPUs don't work well for assignment to a VM, Intel and Nvidia cards both work fine.
The perf is fine - it was a $350 CAD GPU after all.
I am certainly interested to see where Intel ends up going with their lineup. Having a third player in the GPU space is definitely a good thing.
I have a B580 too. The cool thing about it is architecturally speaking it is basically a mini version of the Ponte Vecchio (PVC) datacenter GPU. You can run most of the datacenter GPU workloads, albeit scaled down to fit the compute/memory constraints of the B580. It's a great vehicle for software development. But you can't buy PVC anymore so it's unclear what you are developing for...
intel has been making graphics silicon since the 90s, the current discrete graphics effort has been going for at least a decade, and in areas like low power video decode and encode it could be argued intel is class-leading. the concept of the "GPU" is a quarter of a century old. this is an especially poor article, especially for a publication running as long as techcrunch.
The most rapid path that Intel has to selling competitive GPUs, would be to licence designs from Groq, and apply all effort to getting them working on 14a.
Hyperscalers would bite their hand off and would be a viable alternative to TSMC.
Nvidia has left the door open with the non-exclusive license in the acquisition
Intel focused on SyCL which not many people seem to actually care about. It looks far enough removed from CUDA you’d have to think hard about porting things as well. From what I understand ROCm looks very close to CUDA.
Exactly. CUDA is huge moat and all competitors must be adopting SOFTWARE first approach similar to what tinycorp is trying to do.
Find one single thing that makes CUDA bad to use and TRIPLE DOWN on that.
Why doesn't AMD make a similar framework than CUDA? Is this so much of a task? But if that increases their market share that should be financially viable, no?
ROCm is their CUDA-like and imo it's been a buggy mess, and I'm talking bugs that make your entire system lock up until you hard reboot. Same with their media encoders. Vulkan compute is starting to recieve support by stuff like llama.cpp and ollama and I've had way better luck with that on non-nvidia hardware. Probably for the best that we have a single cross-vendor standard for this.
That may be true, but assuming you meant "within 30% of the performance" ... can we just acknowledge that is a rather significant handicap, even ignoring CUDA.
It's also complicated by the notion that raster performance doesn't directly translate to tensor performance. Apple and AMD both make excellent raster GPUs, but still lose in efficiency to the CUDA's architecture in rendering and compute.
I'd really like AMD and Apple to start from scratch with a compute-oriented GPU architecture, ideally standardized with Khronos. The NPU/tensor coprocessor architecture has already proven itself to be a bad idea.
The customers are players that can throw money into the software stack, hell, they are even throwing lots of money in the hardware one too with proprietary tensors and such.
And the big players don't necessarily care about the full software stack, they are likely to optimize the hardware for single usage (e.g. inference or specific steps of the training).
It is my understanding that this isn't happening in any meaningful capacity, they're simply using the kit no longer relevant to R&D.
I'm still not entirely convinced they actually did Arc themselves. It has all the hallmarks of a project that was bought or taken. Every meaningful iteration keeps getting pushed back further out towards the horizon and the only thing they've been able to offer in the meantime is "uhhhh what if we used two"
It's a confusing article. It's strongly implies that Intel will make GPUs for data centers. It says Intel will produce GPUs without saying whether they are manufacturing them in house or not.
Intel started making and selling their own gpus many years ago, this news is just that they are going to fab the chips themselves, instead of outsourcing to TMSC.
Good to hear. More than two players in the GPU market is a really good thing and their recent dedicated consumer GPUs are really good value in their segment. It will take a few generations until they might catch up to Nvidia, but I am hopeful. This is a good thing.
I am a little confused, I thought Intel had a big push the last couple of years to create its own GPU:
https://www.intel.com/content/www/us/en/products/docs/discre...
The original article is here, and it unfortunately is just as confusing: https://www.reuters.com/business/intel-ceo-says-company-will...
Of course Intel has been designing and selling GPUs for years, I guess Lip-Bu means they're going to start manufacturing them as well? Or they're going to be data-center focused now?
Since he was touting that they recently hired a well-known GPU architect, it seems unlikely that this is merely about them using their own fabs for discrete GPUs instead of having integrated GPUs being the only ones they fab themselves. Some kind of shift in product strategy or reboot of their GPU architecture development seems more likely, if there's anything of substance underlying the news at all.
But this news is somehow even less comprehensible and believable than usual for Intel, whose announcements about their future plans have a tenuous connection to reality on a good day.
I think this is about them creating the silicon. AFAIK their GPU chips were produced by TSMC.
Even the gpus inside their cpus?
Intel has been using a fair bit of TSMC in their CPU manufacturing recently, yes. Most recently they’ve been assembling “tiles” of silicon from many process nodes into a single CPU package and IIRC they have been using TSMC for the GPU tiles.
This year's laptop chips use TSMC for the 12-core GPU parts but Intel 3 for the 4-core GPU parts.
I wasnt aware of that, thanks for correcting.
Intel Arc seems to be well liked, this seems to just be bad writing by Reuters. Unclear what is news here exactly as Demmers was hired a month ago…
Agree, extremely poorly reported out across numerous outlets.
One can only assume the original Intel press release was just as confusing.
Their newsroom website has nothing more recent than Feb 2... I wonder where this came from.
Intel has been making GPUs since the late 90s:
https://en.wikipedia.org/wiki/List_of_Intel_graphics_process...
They did but then it stopped and was scrapped (chief involved also left). So it's more like "re"-start than start.
Any official source? YouTubers have been saying that Intel is shuttering Arc at every minor setback for years.
Deja Vu all over again.
2016 Nervana. Intel would lead in AI training. The "Nervana NNP" was the future.
2019 Habana Intel announced the Gaudi and Goya chips as their new official AI strategy, effectively killing the Nervana project.
2021 Xe general HPC/AI GPU (Ponte Vecchio) Intel said they will be shifting to the "AI chip" market.
2023 The "AI PC". every consumer CPU would now be an "AI Chip" with NPU (Neural Processing Unit).
2024 Intel is now "AI Systems Foundry" to focus on making AI chips for other people (like Microsoft and Amazon).
2026 Intel will start making GPUs
If you discount recent AI-specific developments, you can trace this back further to Larrabee, Intel MIC, Xeon Phi, etc., etc., etc.
Wasn't Itanium marketed as being good for some kind of AI?
Don't beat a dead horse, that's rude ;). They did kinda try to market it as good for everything though, didn't they.
"Intel keeps starting to make GPUs"
Intel has been designing GPUs manufactured on TSMC nodes across client and datacenter for at least the past 5 years. The client chips are price competitive but not performance competitive with AMD/NVIDIA/Apple. The data center roadmap has historically been a huge mess with cancelled products left and right. But, to say "Intel will start making GPUs" seems misleading. Perhaps "Intel to try to inject sanity into its GPU roadmap" would be a better headline, though I am skeptical one hire will do anything to fix 10+ years of mismanagement.
I have a B580 in my desktop. Unfortunately AMD still has broken PCIe level reset and so their GPUs don't work well for assignment to a VM, Intel and Nvidia cards both work fine.
The perf is fine - it was a $350 CAD GPU after all.
I am certainly interested to see where Intel ends up going with their lineup. Having a third player in the GPU space is definitely a good thing.
I have a B580 too. The cool thing about it is architecturally speaking it is basically a mini version of the Ponte Vecchio (PVC) datacenter GPU. You can run most of the datacenter GPU workloads, albeit scaled down to fit the compute/memory constraints of the B580. It's a great vehicle for software development. But you can't buy PVC anymore so it's unclear what you are developing for...
intel has been making graphics silicon since the 90s, the current discrete graphics effort has been going for at least a decade, and in areas like low power video decode and encode it could be argued intel is class-leading. the concept of the "GPU" is a quarter of a century old. this is an especially poor article, especially for a publication running as long as techcrunch.
The most rapid path that Intel has to selling competitive GPUs, would be to licence designs from Groq, and apply all effort to getting them working on 14a.
Hyperscalers would bite their hand off and would be a viable alternative to TSMC.
Nvidia has left the door open with the non-exclusive license in the acquisition
They've been trying for nearly 30 years:
https://en.wikipedia.org/wiki/Intel740
So they are just moving GPU production in-house? Weren't they already designing GPUs for contract with TSMC?
The silicon is just one piece of the puzzle. CUDA and the rest of the software stack is huge advantage for NVIDIA.
Intel focused on SyCL which not many people seem to actually care about. It looks far enough removed from CUDA you’d have to think hard about porting things as well. From what I understand ROCm looks very close to CUDA.
Exactly. CUDA is huge moat and all competitors must be adopting SOFTWARE first approach similar to what tinycorp is trying to do. Find one single thing that makes CUDA bad to use and TRIPLE DOWN on that.
Why doesn't AMD make a similar framework than CUDA? Is this so much of a task? But if that increases their market share that should be financially viable, no?
They do. It's called ROCm. It works, it's open source, but CUDA is so entrenched it's like a Windows vs. Linux kind of thing.
ROCm is their CUDA-like and imo it's been a buggy mess, and I'm talking bugs that make your entire system lock up until you hard reboot. Same with their media encoders. Vulkan compute is starting to recieve support by stuff like llama.cpp and ollama and I've had way better luck with that on non-nvidia hardware. Probably for the best that we have a single cross-vendor standard for this.
They even make two! rocm and hip. And intel has one…api.
yup, which is why AMD struggles so much even though its hardware is usually within 30% of the performance (give or take) of NVIDIA.
(Replaced "with 30%" with "within 30%")
That may be true, but assuming you meant "within 30% of the performance" ... can we just acknowledge that is a rather significant handicap, even ignoring CUDA.
It's also complicated by the notion that raster performance doesn't directly translate to tensor performance. Apple and AMD both make excellent raster GPUs, but still lose in efficiency to the CUDA's architecture in rendering and compute.
I'd really like AMD and Apple to start from scratch with a compute-oriented GPU architecture, ideally standardized with Khronos. The NPU/tensor coprocessor architecture has already proven itself to be a bad idea.
AMD is doing that for their next gen, no idea if khronos is involved.
The customers are players that can throw money into the software stack, hell, they are even throwing lots of money in the hardware one too with proprietary tensors and such.
And the big players don't necessarily care about the full software stack, they are likely to optimize the hardware for single usage (e.g. inference or specific steps of the training).
Intel people would point at SYCL if you told them this
It is my understanding that this isn't happening in any meaningful capacity, they're simply using the kit no longer relevant to R&D.
I'm still not entirely convinced they actually did Arc themselves. It has all the hallmarks of a project that was bought or taken. Every meaningful iteration keeps getting pushed back further out towards the horizon and the only thing they've been able to offer in the meantime is "uhhhh what if we used two"
> I'm still not entirely convinced they actually did Arc themselves
Raja ex AMD / Radeon ran the project?
It's a confusing article. It's strongly implies that Intel will make GPUs for data centers. It says Intel will produce GPUs without saying whether they are manufacturing them in house or not.
“We realized that we were too focused on the consumer market. “ - Intel
Didn't know my A770 wasn't a GPU.
The CEO of Intel already specifically said Intel has given up competing with Nvidia.
That’s the spirit!
Wtf Intel is doing, they can't even communicate well !
I'd be more excited about Intel making HBM.
…again.
3 years too late to get serious about GPUs.
Not at all. There is no indication that the world won't need more GPUs going forward.
Intel started making and selling their own gpus many years ago, this news is just that they are going to fab the chips themselves, instead of outsourcing to TMSC.
Good to hear. More than two players in the GPU market is a really good thing and their recent dedicated consumer GPUs are really good value in their segment. It will take a few generations until they might catch up to Nvidia, but I am hopeful. This is a good thing.