>32GB VRAM>600 GB/s memory bandwidth>SR-IOV support which normally only server grade cards have>price estimated $1000>B65 has same memory but only $650https://www.techpowerup.com/347703/intel-announces-arc-pro-b70-and-arc-pro-b65-gpus-maxes-out-xe2-battlemage-architecture#g347703-1Fuck, this might be a great grab for trying AI shit, or using with VM's thanks to SR-IOV
AI shit favors cuda. your performance will be diminished if not error prone.
>large scene ready>robust pro softwareIncredible
wow, have fun paying a scalper 800% msrp
>>108461534I'm a bit of a tourist here, is this only good for 'other' productivity, home gamer nerd shit, and some games?
>>108461534anything with fast ram is fine for inference in practice.
>>108461606If someone has written an application that supports your GPU. As soon as you leave the CUDA happy path, you will realize that barely anything does and you will chase down forks that don't really work and break with meaingless errors.
>>108461534vulkan works fine for AI. I use it for qwen 122b all day.It's interesting for sure and I hope it does well because a 5090 for $3-4k is fucking stupid.
it's a trap, the successor for Battlemage has been cancelled already
>>108461534>>108462303Nvidia shills in shambles.
>>108461522nice
too bad they're 5 years too late with sr-iov support and you can just run everything in wine now instead of needing gpu passthrough to a windows vm, still good for vdi though I guessseems decent enough as an AI card according to Wendell from level1techs' video
>>108462303Could be.I really liked the B580, but the absence of a B770 and no news what so ever about the successor have me worried.I really hope they stick with GPUs.If they manage to push out a 770 for Celestial (or Druid if they skip C) I will definitely buy it.
>>108461522>B65 has same memory but only $650that's insane
>>1084615225090s are ridiculously overpriced so this is welcome, but there's no way around intel super being behind in terms of software.>>108461593The user experience isn't great for gaming. They're not as rough as they were a few years ago, but they still remind me a lot of the 2000s with random driver issues that may or may not annoy the shit out of you depending on your patience.
>>108461535It's bleak, but they write that shit in big letters for a certain demographic to pretend they understand why it's a good product. Doesn't mean the product isn't actually good. Keep in mind how many marketing RETARDS get their hands on these announcements before they put them out.
I just don't understand how $300 dollars gpus aren't the norm, like who the fuck would even pay more than that for video games and why the fuck would you make games that require $6000 dollar hardware to run and limit yourself to 0.0001% of the potential market
>>108462895The market will happily carry on milking that 0.0001%. If it runs at 144p 15fps upscaled via dlss framegen slop on playstation that's good enough.
>>108462895I could play any game I tried on my laptop 1050ti. I never tried any very new games like the last 3 years or so, but I think most of them would've run too. It'd get solid enough frame rates at 1080p on low to medium settings. 1050ti is like ewaste by now and it still works for most things, they rely on people thinking they need 200fps at 4k. If you want a $300 GPU then buy a $300 GPU.
>>108461522Can't they do 64GB or 96GB already.
>>108461522>>B65 has same memory but only $650wow.
>>108463721I think they are testing the waters.>>108462861>there's no way around intel super being behind in terms of software.One thing Intel doesn't understand is that nvidia aggressively uses guerrilla tactics to cause llama.cpp and derivatives to suck on amd (and intel)
>>108461522How good is the path tracing performance? This seems like it could be magnificent for Blender.
>>108461522well shit... maybe i'll just buy intel instead. amd cpu intel gpu... my have the tables turned.
>>108463809it might match an rtx5060 since its on-paper specs are 60% higher than the b580
>>108461522Tk/s using Qwen 3.5 27B?
>>108461522why announce a product you cannot buy
>>108461522is this competing with the Radeon AI PRO R9700 ? Always wondered if those cards are worth buying if I want image gen.
>>108462895a burger costs close to $10 these days $300 for a gpu is simply not worth it anymore.
>>108461534That is not important !That is just testing platform!Important is what is not mentioned here ...Of course if you understand main goal!!!Let's keep Nvidia asleep!!!
https://www.newegg.com/intel-arc-pro-b70-32gb-graphics-card/p/N82E16814883008?item=N82E16814883008&Source=socialshare&utm_campaign=snc-copy-link-_-sr-_-14-883-008-_-03272026&utm_campaign=snc-copy-link-_-sr-_-product-_-03272026&utm_medium=social&utm_source=copy-link
https://www.microcenter.com/product/709007/intel-arc-pro-b70-single-fan-32gb-gddr6-pcie-50-graphics-card
>>108464518It's the Intel way.
>>108461522I can't use this to prompt porn with Stable Diffusion, what is the point? Regretting not buying a 3090 every day.
>>108461522>$1400XDlol get lostthis still makes no sense over a 3090
>>108461522Nvidia will do a sisterfucker special and buy all the cards so their empire of slop is protected
>>108465272It's cheaper to bribe industry partners to not sell raw components to Intel but yeah, that seems like something Nvidia would do
If Intel is better than AMD's shitty crashy drivers and a compute platform that is usable, unlike ROCm, then I'll buy. I'm sick of AMD's shitty everything and tired of pretending they're good. Even their CPUs have bizarre erratas and microcode fixes.
>>108465315What do you mean you need 1044 days of uptime? 522 days is more than what any rational man would need
>>108461522As long as some actually useful ML library works with it. I don't need to spend my life trying to get the tool chain working. Within a certain narrow set of variables, Nvidia is plug and play for ML. Hence I paid €1200 for a 20GB card a couple years ago. Would be great to have more but not for €4000
Are these any good
>>108461522I get they only did this because they had the designs ready, but its not really much better than the b50they really should have gone with double the bus size and 48gb, then this would have been a massive hit and insta sold out
>>108461522are they designed to cluster? that would be the only value
>>108462895APUs are unironically good enough for most people now. AMD's last few APUs were already running modern games at 1080p30, and Intel's newest integrated solutions are even faster. Nvidia wants you to spend thousands for memetracing but it's really not necessary
>>108461522>600 GB/sLmao. 2080 Ti called.
>>108464551Not even close. The Intel card is faster and between 2/3 or 3/4 of the cost. On top of that, AMD is "what are you doing" tier for software support when it comes to attention optimization nodes. The setup is fiddly, and I get crashes on AMD unless I pin my drivers to specific releases.
These look nice, I think I am going to sell my PS5 and my current rx 6600 xt PC and get a prebuilt with this gpu in it
>>108464356Wendell got around 500Tk/s with 4 of them in his system
>>1084713586700XT owner here, I play everything I like at 2k
this shit will never be bought for 1k kek
>>108461534latest turboquant shit landed for metal firstcuda is now second class citizen
>>108461606you have to wait months for your implementationto land tho
>>108471595For comparison I get 50 tk/s of single user throughput with 2 3090s on pcie x16 gen 4 without p2p on qwen 3.5 27b fp8 with the marlin w8a16 kernel at 0 context under vllm.
>>108461535>>108462876Large scene ready is kind of true though
