I'm sorry what?
It just reads in the article
"In response to a question on Twitter, Koduri said,"
followed by Raja Koduri's reply
"we are".
What was the question? Where is the source?
If he did just write "we are" and nothing else in the tweet and with no context in his response. Wouldn't that be the opposite of denying it when in all probability the question would have been "Are you cancelling Arc Graphics?".
There's also the sources in Hilbert's post, at the top, where you can get onto the twitter post in question (last here on the screenshot).
And clearly it's NOT what he implied to say, somebody's just too stupid to read it as it was intended.
"We are [not sure what to think] about these rumors as well. [...]"
There you go, that's the actual quote. Sometimes I wonder why such things are even issues when people, me included, do not check sources, and do not even read a single twitter post.
Hilbert was kind enough to link the source. Which is good journalism, imho.
Honestly, I hope not. What kind of CEO expects to enter a difficult and competitive market and succeed first try on getting ahead of competition?
They need to invest heavily on user experience, aka drivers. If the price is competitive, they will sell.
If they nail the drivers, its just a case of keep on improving the hardware generation after generation.
And if they can take the lessons learned to make better iGPU or even APUs, wouldn't that be a sweet deal?
Any company starting a business fresh is going to be in the red, how long they can be there depends on many things.
Intel has no reason to cancel these GPUs right before launch.
Even if these are of no interest to enthusiasts that build their own PC, Intel has enough dealings with pre-built companies, to put these in millions of PCs around the world.
A possible decision to cancel Arc will not impact the development and launch of the the first series of cards as they are fully designed (maybe badly) and most likely already produced (at least fabbed in silicon). What might happen is that Intel decides to cancel further development of Arc or to refocus Arc away from the gamer/consumer markets and just develop professional cards instead (think Nvidia Quadro equivalents).
So while this last option technically means that Arc is not doomed, it might not be aimed at the gamer/consumer market anymore, with its heavy need for game specific optimizations in the driver stack. In the professional market the required optimizations are much less widespread and much more manageable.
A possible decision to cancel Arc will not impact the development and launch of the the first series of cards as they are fully designed (maybe badly) and most likely already produced (at least fabbed in silicon). What might happen is that Intel decides to cancel further development of Arc or to refocus Arc away from the gamer/consumer markets and just develop professional cards instead (think Nvidia Quadro equivalents).
So while this last option technically means that Arc is not doomed, it might not be aimed at the gamer/consumer market anymore, with its heavy need for game specific optimizations in the driver stack. In the professional market the required optimizations are much less widespread and much more manageable.
Arch is done. All it's missing is the official launch, and improve drivers.
And Alchemist is the next step, which Intel already said already started development.
Intel would be naive if they thought they could be better than 3rd place for the first gen, regardless of the driver situation. They didn't even attempt to outperform the 3070 even in theory. Granted, you don't have to have the fastest GPU to necessarily be better than 3rd place - if you've got something with a great price and great performance-per-watt, it can still be the better product; an architecture isn't necessarily inferior just because there isn't some supersized version of it. The average person isn't going to spend $400+ on a GPU so sometimes, something like a 3090 Ti or a 6900XT don't really have much appeal.
Anyway, Intel's core competency is their platform as a whole, rather than any of the individual parts. Intel still sells well in the server and workstation markets to more conservative customers who need something to work reliably and consistently rather than quickly. Intel either makes everything you need to run a server or is partnered with those who will. This is also where IBM makes their money, except they take reliability to a whole new level.
I'm certain Intel knew they would be in third place. We can assume that because they didn't try to make a large die to compete at the higher end.
I do expect them to get this gen to 3070 perf after 6-9 months of driver tweaks. I also expect Intel to keep investing they need to have a dGPU if anything for the server space for AI etc as you are saying for complete package systems.
A possible decision to cancel Arc will not impact the development and launch of the the first series of cards as they are fully designed (maybe badly) and most likely already produced (at least fabbed in silicon). What might happen is that Intel decides to cancel further development of Arc or to refocus Arc away from the gamer/consumer markets and just develop professional cards instead (think Nvidia Quadro equivalents).
So while this last option technically means that Arc is not doomed, it might not be aimed at the gamer/consumer market anymore, with its heavy need for game specific optimizations in the driver stack. In the professional market the required optimizations are much less widespread and much more manageable.
I´ve also thought about that possibily but from what i understand the hardware for gamers and professional is almost the same, the biggest difference is the software.
So it makes sense to continue to make GPUs for gamers as long as they are also making them for professionals.
If i`m wrong with this, feel free to correct me.
Arch is done. All it's missing is the official launch, and improve drivers.
And Alchemist is the next step, which Intel already said already started development.
I´ve also thought about that possibily but from what i understand the hardware for gamers and professional is almost the same, the biggest difference is the software.
So it makes sense to continue to make GPUs for gamers as long as they are also making them for professionals.
If i`m wrong with this, feel free to correct me.
Yes, though in the professional market you see (at least with NVIDIA, but I think with AMD as well) the use of ECC memory i.s.o regular GDDR. So the hardware design is from a board perspective different (don't know if this also requires changes in the GPU itself or not).
The basic problem that Intel is struggling with (apart from the supposed hardware issues that make the use of Resizable BAR essential for Intel) is that in order to get good traction on the consumer/gamer market requires an enormous effort from the driver team, as each game supported requires specific driver level optimizations. Both AMD and NVIDIA have worked on these since the beginning of their GPU product lines and as such have these in place for their current and future offerings (barring a drastic hardware change that would make these unusable). Intel is starting from scratch in this regard and as we've seen from benchmarks has quite some work to do.
This would not be the case if they just service the professional and data center markets. Of course there they have the same problem as AMD has: CUDA is highly entrenched in these markets and as such makes entering these markets challenging. AMD has opted to try and achieve this by creating ROCm and HIP. HIP uses a similar programming model to CUDA and AMD supplies tools to convert CUDA code to HIP (within limits of course, functionality that doesn't exist in AMD's GPU's or is drastically different will require some manual modification). HIP is so similar that there is even an option to compile the HIP code via the HIP compiler to NVIDIA cards via the CUDA compiler with similar performance as a native CUDA implementation.
Intel on the other hand has created something called OneAPI, which is supposed to let you run the same code on CPU, GPU and FPGA. This uses a programming language that is a mix of C++, SYCL (a collection of templates and lambda functions in C++ for heterogeneous processing) and OpenCL, called DPC++. This is a completely different programming model from CUDA (though some of the constructs due to their origin in OpenCL look similar) and thus forms a much bigger hurdle for adoption than ROCm does (there you can almost get away with replacing all cuda parts of function names with hip and have compileable and runnable code).
This will make the use of Intel GPU's for High Performance Computing or other GPU intensive tasks like rendering or video processing less simple as it requires extensive rewriting of code. Which will form a big hurdle for Intel to overcome.
This coupled with the fact that up till now Intel has only announced the lowest level cards in the professional range (think A380 equivalent), which aren't typically the most used professional cards as they have little performance to offer. Intel really needs to launch the mid and high end professional cards soon in order to capture any market share.
Arch is done. All it's missing is the official launch, and improve drivers.
And Alchemist is the next step, which Intel already said already started development.
Shouldn't that be Batllemage as the first series of cards is Alchemist (so the current crop of Arc cards are Alchemist). Arc is just a family name, a bit like Geforce or GTX was.
There's also the sources in Hilbert's post, at the top, where you can get onto the twitter post in question (last here on the screenshot). And clearly it's NOT what he implied to say, somebody's just too stupid to read it as it was intended. "We are [not sure what to think] about these rumors as well. [...]" There you go, that's the actual quote. Sometimes I wonder why such things are even issues when people, me included, do not check sources, and do not even read a single twitter post. Hilbert was kind enough to link the source. Which is good journalism, imho.