Thread 106518987

>intel arc pro b50
>passmark g3d scores of 18000
>16GB vram
>4x displayport
>70W, no PCIe power needed
>dual slot low profile
>fully open source linux drivers
>$350

This thing blows rtx 4060 lp (115w) and 5060 lp (145w) out of the water.

>>106518987 (OP)
too bad intel is imploding

>>106518987 (OP)
I'm hoping it's got a fan but low profile cards like that generally assume they get generous external cooling. Even with a fan I'd still be concerned

does it come with a single slot option?
this could be the replacement to my chinese rtx a2000 8gb

>>106519044
it's a blower design
also it's 70W, they're not furnaces compared to blackwell cards

>>106518987 (OP)
How are the drivers?

>>106518987 (OP)
based

1 slot cards are more important. they made a 1 slot gtx 1070 before.

>>106519202
imagine the noise

where the fuck is my B770? Where is my dual B60? Why does it take this long for Intel, a giant multinational that is struggling to field good products, to update its product stack in an exploding sector where it can make well received products?

wtf are they doing?

>>106518987 (OP)
Hdmi???

>>106519235
It's an interface for poor people. This is a professional grade gpu

>>106519229
Probably too busy trying to make more proprietary x64 extensions

>>106518987 (OP)
I already got the A310 for $100 for my AV1 encoding homeserver needs

>>106519229
>wtf are they doing?
like we said, imploding

sff is only useful in real life for servers and other specific applications. most people here who are interested in sff are doing it for social media/Internet clout. faggots.

>>106518987 (OP)
>displayport
Aryanport chuds... we won.

>>106519229
>wtf are they doing?
They're choking the GPU department probably due to department rivalries over funding. Big corporations like that almost work like government without proper direction and vision (Pat Gelsinger).

>>106519495
they are also useful for repurposing sff optiplex & thinkstation shitboxes for AI/gaymen

>>106519059
>they're not furnaces compared to blackwell cards
just power limit?

>>106519792
they physically cant draw more than 75W from the slot

>>106519824
and that is relevant to what I said how?

Anyone tested SR-IOV yet?

>>106519833
why would you intentionally gimp something that you bought?
fashion?

>>106519229
cpu overhead is prohibitive, litterally no point releasing a higher end card, because the drivers cant handle the high draw calls of new games on most cpus. basically they are stuck with 2014 amd tier driver performance. remains to be seen if its fixable or not, at the very least i would not expect a launch this year, they will probably skip it, and if it does come out, it will be workstation/compute only most likely.

>>106520089
a 70% power limit gets you >90% of the performance, it's just logical

>>106518987 (OP)
>16GB vram
>70W
I wouldn't mind supplementing my 4090 with something that low powered that could keep an LLM up and running 24/7, but what kind of quality can you get at 16GB? Like, what's the best quality model/quant you can cram into that tiny amount of VRAM?

Also... I don't see it for sale anywhere.

>>106519792
That doesn't change the fact that you NEED a PCIe power cable for those cards. This is comfortably the most powerful slot-powered card around, so is perfect for upgrading old office systems and such to turn them into gaymen PCs.


...or it would be if it wasn't $350, meaning it makes no sense for a "budget" build.

they will go bankrupt before this thing hits the shelves

I quite literally just want to purchase 2 x B60;s.
Intel are so slow. Is a good sidegrade from a 9060 XT for LLM use but holy shit just release the fucking B60 already.

>>106518987 (OP)
>idles at 40W
no thanks

>This thing blows rtx 4060 lp
Proof?

>>106522237
they're one firmware flashing away from each other

>>106519018
>too bad intel is imploding
Honestly I'd be okay if everything BUT their GPU division goes under. Sure the GPU's still can't compete with AMD and Nvidia yet in things like gaming BUT they're a great deal for video encoding on a budget and who knows what future fruits ARC will bear.
That's the best outcome. To bad Softbank and US tax dollars are gonna keep the worst aspects of that company a float much like when Bush and Obama bailed out GMC and Chevy numerous times and now people only drive those sheit cars because they're poor and the moment they have enough in their piggy bank or get a raise they almost always end up buying something from Japan that isn't an unreliable teethe rattling piece of shit.

I guess we just aren't allowed to have nice things in Clownworld.

>>106523084
>Sure the GPU's still can't compete with AMD and Nvidia yet in things like gaming BUT they're a great deal for video encoding on a budget and who knows what future fruits ARC will bear.
This is a meme, they're pretty good at budget gaming and for things that are GAAS w/ MMO like titles.
The AAA vulkanDX11,12slopfests that people benchmark with and don't really play they do alright in, not bad, not great but on linux the cards are good. For low tier gaming, even if you've got a 4k display and deal with the VRAM overhead on linux, the cards at least work with gamescope which nvidia cards don't, so you're good to game at 1080p etc.
AI, they're good bang for the buck, good community support and their ecosystem isn't the mess that ROCm & HIP are/were. ROCm outside linux is a disaster for RDNA4. ROCm in linux for RDNA4 is a disaster.

Intel should continue to exist. They're almost in that sweet spot. Timing and that company are what fuck them over.
The B50 existing in may before the 9060 XT came out woulda been good.
The B60 needs to be here now. 16 & 24GB GPUs for inference at an affordable price that shits down nvidias neck, good 2nd cards for video encoding and light 3D work.

The problem now with the B50 is it's only just filtering out, at a time where people really want the B60's 24GB VRAM and initial pricing? The same price as the 9060 XT with slower memory bandwidth pcie5x8 versus 5x16. 70W is killing it, it's a card for people like me who want to throw into their 2nd computer but at the price of the 9060 XT. If compute and VRAM are the end game you'd just wait a few months.
It's in such a weird spot compared to the B60 & B580. The B580 had a good launch window but VRAM deficit it works where it worked, the B60 promises 24GB of VRAM per card, the B50 is this weird inbetweener, filling the low TDP niche with decent VRAM, I'd use 2 of them for inference if they were $100 cheaper, otherwise wait for B60.

>>106519018
They can spin the GPU division into a separate company assuming they can actually sell units.

>>106522167
>idles at <0.5¢/h
the horror

>>106519229
stop manspaining, Intel knows what it is doing

>>106519229
probably just a racist who doesn't like GPUs made by brown people

>>106518987 (OP)
What's the gotcha part? Bad at video games? Can I use it on an older 9th gen intel platform with pcie 3?

How does it compare to 750ti?

>>106524422
The gotcha is that you're a bitch and need to upgrade your ewaste hardware.
These cards will technically work ok on ewaste machines, but even first gen Alchemist Arcs need resizable BAR (you probably don't know how to modify the bios to enable it in unsupported boards) and modern cpus (aka at least new enough to not be stuck in the 14nm+++++++++++ 2c4t/4c4t/4c8t i3/i5/i7 era) to get the full performance out of the cards.
An A380 is "about" the performance of a 1660. This newer card is going to be far better, but it's weaker than the B580 because it's not a gaming card.

>>106524422
>What's the gotcha part?
It's never in stock, the one I want isn't anyway.

>>106518987 (OP)
>>$350
Where?

>>106518987 (OP)
>no PCIe power needed
>low profile
What is the point of this? You take a miserable hit to performance and especially to price:performance, and for what? To save a few litres?
If the idea is to upgrade an old office machine: All the mini cards I've seen are either so shit that there's barely any point compared to a recent APU, or they're so overpriced that they wipe out any saving from not having to buy a PSU and case. There seems to be no in-between.

>>106524116
>Goy
>Poo
>Unit

>>106518987 (OP)
Sriov status? This would be a nice budget card for a virtual machine rig/proxmox server.

>>106525731
>If the idea is to upgrade an old office machine: All the mini cards I've seen are either so shit that there's barely any point compared to a recent APU,
Are you retarded?
That's a rhetorical question, it answers itself.

>upgrade an old office machine
>recent APU
Stop being a fucking retard in public, have some fucking humility.

>>106518987 (OP)
Can it do local AI stuff or do I still need an nvidia gpu for that?

>>106525823
The benchmarks and reviews I all saw say it's coming Q4, likely whenever they finally shuffle B60 out the door.
>>106524837
Why don't Americans just import from other nations that have stock? I don't get it, we're flush with intel GPUs. I saw shit constantly about B580 shortages and they were in stock here the entire time.
Shit, we still have BNIB A770 and most of the Alchemist line too in some form of another.

>>106526018
>Can it do local AI stuff or do I still need an nvidia gpu for that?
That's part of their pitch, it won't crunch compute or be blazing fast but it will be efficient for what it puts out. It's 70W/16GB for $350. A770 would be a bit cheaper but 200W I think, 9060 XT is $350 when it's at MSRP for 170W TDP.
You're buying size/convenience with this. The biggest crutch is the x8 lanes but it's not that huge a nerf.

>>106526008
I guess if you already have said old office machine and want to put a 1050-tier card in there, fine. The scenario I had in mind was the people who buy an old optiplex SFF with the intent of upgrading it.
The card in the OP doesn't seem to be very appealing in either scenario. It's way too expensive for a super-budget optiplex setup but also really weak compared to what you can get for a few more litres and an extra hundred watts.

>>106526018
Local LLMs and other tools approved by Intel R TM AI department. Outside 99% it will not work.

>>106518987 (OP)
>SFF

>>106526024
its not coming this year most likely, their driver is not in a good state.

>>106524485
>>106524422

9th gen supports resizable bar, but bios implementation is spotty depending on the board., the bigger problem is that intel has large cpu overhead , so older cpus struggle. litterally anything nvidia or amd is a better choice than intel for older platforms.

>>106526075
Depends though there's other economics at play. Modern cards have modern features, encoding, decoding, support for later versions of Vulkan/DirectX.
I have a 2nd PC that could use one, an HP box with 32GB of RAM and I could get a 1050, 1060/1660 tier card, use the PSU and I could save a bit of money on it.
On linux driver support only matters in terms of nvidia support being terrible. On windows, those cards are on their way out too. There's use cases for a modern low TDP card, they gain in efficiency too and whether you like it or not the 100W difference between a card when run full tilt at a year or thereabouts can dwarf the savings too when you go for older hardware.

This card's price is higher than what Intel were going to be selling it for, the price was supposed to be $300, for 16GB of VRAM, that's local LLM tier and access that nvidia just can not offer you at that price point, TDP or anything really. The closest is AMD's 9060 XT with a 100W TDP premium.
100W isn't much, it seems cheap enough but in certain energy markets 1 years use outstrip the savings on an old card, with less features, less memory, less flexibility or support down the line. Over 2 years the cost between it and the RDNA4 card are called into question. That assumes constant use and compute but for a heavy user or specific scenario it comes in.
If you intend to use it for 5 years or more, it pays for itself.

In case you think I'm coming from cuckoo land.
I have an R7 250 (not x), good ol' GCN 1. It's 65W, it can do some of the things it needs to, it's short on memory, runs at load, doesn't have great media encode/decode support. Really sits about GTX 550 Ti tier. PCIe3 x8, on a PCIe3 machine the B50 will be constrained to the same bandwidth limitations. but for the same watt for watt output, you're getting much more. Somewhere around 8-9x the performance. Can make sense. Not to you maybe.

>>106527058
Does battlemage have the same driver overhead issue with MESA? I only used the B580 as a stop gap between my current GPU change.

The B60, the cards value is entirely in the pro market and the AI inference market.
Delaying is painful for them. It'll still do well if there's stock purely because both AMD and Nvidia are not going to release a 24GB card in that price range, at all. It's just not going to happen.
I get it though, if the problem exists only on windows, it leads to the most common user complaining. Just so happens to be that Battlemage's pro line is more suited towards tasks outside of just gaming.

>>106526075
>>106527114
there are versions of the rx 560 without the pcie power connector, the one i have has a ppt limit of 43w, if you are willing to use 1 pcie power connector, the rx 570/580 is the best choice for the price, decent efficiency, reducing the ppt limit can put it in the 75-100w range easily and you retain most of the performance. otherwise the rtx 3050 or 5050 are the way to go. 750ti or 1050ti are a bit slow now for the price.

>>106527114
Thanks for the thoughtful reply anon. I have to admit I never really take power consumption into account, even though electricity is quite expensive in my area.

>>106527137
havent really seen much in terms of benchmarks since launch, but windows was faster and the last linux benchmarks I saw showed some improvement , but not enough i think to support a faster gpu. in compute only its probably fine, but they need customers to sell it, and i think as a workstation card it would make sense, since 3d rendering is a big part part of it, and ai is atough sell for anybody serious ,even with extra memory since nvidia is so much faster and better supported. even amd has a hard time selling cards and they have much faster gpus than intel.

>>106527316
*wouldnt make sense

>>106518987 (OP)
>This thing blows rtx 4060 lp (115w) and 5060 lp (145w) out of the water
Yeah i don't give a fuck cunt Intel can fuck off forever and die (like they already are)

>>106526574
>Cramped mobo with a single PCIe slot
DTX format is forgotten, as usual.

Still rather pick laptop + eGPU though, SFF was more akin to 2010's before Ryzen showed up and made 6 core cpus mainstream, since i5's have always been a huge POS till 10th gen, and now they are the only decent stuff intel makes lmao.

>>106527361
>I want Nigvidia to become an even stronger monopoly because... JUST BECAUSE?!?!
Unhinged, turn yourself into the psych ward anon before you hurt yourself or someone else.

>>106526574
They’re good for server pc. I have one for running my security cameras.

>>106518987 (OP)
>SR-IOV

>>106526574
Imagine being poor

>>106520251
For workstation it's pretty great. 16GB of VRAM for $350 is crazy for AI too.
The B60 is supposed to go up for 48GB in some third party builds (Sparkle), and not be too crazy of a price. Not sure if they've released any price estimates for that yet though.

>>106519240
It's quite cheap for what you get. I would have though poor people liked cheap things.

>>106527316
Intel's AI position is pretty good atm with community support, both it and ROCm/HIP have come far. For image or video gen there may be issues with support, speed for image/video gen is where compute will hammer it but inference? The $300 price they gave is good, the 70W TDP 16GB card is, a niche but there's nothing else out there with that sort of TDP > VRAM trade off. If you were using Vulkan or one, you could pair it with an existing A770, 32GB split over two cards with 70W for 16GB is pretty big.
Could see more focus on Intel compatibility when the B60's hit if they're good supply. They're a decent size for low scale local LLM.

>>106527543
$1200 is what the maxsun dual was quoted as, $1-1.5k for the 48GB dual boards.
The 24GB was initially quoted at $500, 120-200W and 24GB VRAM on a single card is good shit, combine it with a 16gb and aside from the overhead you're in 40GB VRAM territory and 46-48GB with two.

>>106518987 (OP)
black screen on everything expect last expensive intel scam

>>106526136
So probably only Israeli trained models. DROPPED.

>>106527774
>$1200 is what the maxsun dual was quoted as, $1-1.5k for the 48GB dual boards.
I thought it was around that but I couldn't quite remember. Absolutely crazy for 48GB. That's new as well so the used price will be even lower.
It puts 96GB of VRAM in the reach of enthusiasts who are willing to get a 2x PCIe x16 board and two of these cards. Probably puts you at, what, $4,000 for new parts? A Blackwell at 96GB is about $10,000 from what I can see.

>>106526018
Intel has a tool for getting CUDA to work on Arc GPUs, called "Intel Extensions for Pytorch" or "xpu".
They also maintain a fork of Ollama that makes use of xpu.
Can't comment on whether it works well or not, but there are some signposts for you.

>>106526018
>>106528174
I forget to mention, the fork is called IPEX-LLM.

>>106528142
Yup, about that.
It's worth noting that the same scales apply for the 24GB cards. It's 24GB NEW not a used 3090 that's been through 5 years of misery and chugs an additional 150W of power. Slower in compute and memory bandwidth yes, but still a load of VRAM. If you put that on a new system like you said, pair it with as much RAM as you can and you're in big model territory.
For people who have the idea that old cards are good enough, buy one of the many 3090's that are still in demand. It's availability BNIB that will impact prices across the board too.

>>106528342
I think the supposed downsides to used cards are overstated. The one thing of value to me that LTT ever produced is his video on used mining cards, and he didn't find any flaws with the majority of the cards. Chips don't really fail in that way. Age related failure is more of a mechanical wear thing, but a conductive piece of silicon remains a conductive piece of silicon with usage.
I actually just recently bought a used 3090 because NVIDIA's sales strategy for the 50 series here has caused the price of 3090s to fall pretty hard. It's still in the post but I'm looking forward to it arriving for my server build.
That said, even though I like used, I'm going to be looking for used Arc next time around for sure.
I don't really feel any brand loyalty to any components company, so for me this is great because it's disruptive and hurts the notion that high VRAM should be really expensive. Feels like a bit of a price bubble might be about to burst.

>>106519846
>>106525823
>SR-IOV?
Thirding this, I just need proper confirmation it's a feature. Something slot-powered with 16GB VRAM and SR-IOV sounds perfect.

>>106518987 (OP)
Very interesting, will probably buy on release to replace my aging nvidia card.

>>106518987 (OP)
Looks perfect for a home server. Local LLM (for home assistant voice assistant), object detection on video streams, maybe even a simple game server for local streaming with moonlight/sunshine so the gf can play vidya on her laptop/tv while I'm using my desktop.
Will have to step up from the 1L mini PCs I've been using up until now though.

>>106518987 (OP)
that's very cool anon, if i didn't have my current gpu i'd buy that

It's listed for preorder on Newegg: https://www.newegg.com/intel-arc-pro-b50-16gb-workstation-sff-graphics-card/p/N82E16814883007?srsltid=AfmBOoriEKWhA1fvvtoK64WDEUwolMvWsC25GAPp1rvpc5ygFT5AQmDR

>>106518987 (OP)
I would wait for this instead
https://www.nvidia.com/en-au/products/workstations/professional-desktop-gpus/rtx-pro-2000/

>>106531081
That won't be cheaper than the intel offering.

>>106524104
Please tell me how one mega-corp is better than the other mega-corp.

>>106531225
I buy all of my video cards locally hand-crafted from an anarcho-syndicalist commune of Lesbian open-source hardware designers. Avoid the corporations and crush the patriarchy!!!

>>106531269
And PoC-owned too!

>>106518987 (OP)
Can I use it despite hating Intellaviv?
>>106519018
Dayum. I have mixed feelings.

>Intel is imploding
two more weeks
fucking retards

>>106531384
>Can I use it despite hating Intellaviv?
The upcoming RTX Pro has way more TO/S.

>>106531423
>EVERYTHING IS FINE
>NEXT NODE IN 2 WEEKS
>TOO BIG TO FAIL ANYWAY
>NOW PLEASE IGNORE THE CEO PAYRISE
a nice mindset to have

>>106528174
> Intel has a tool for getting CUDA to work on Arc GPUs, called "Intel Extensions for Pytorch" or "xpu".
This is misinformation, it does not make cuda work on intel gpus, ai software developers have to add support for intel cards.

>>106518987 (OP)
>>$350
no way
SHUT UP AND TAKE MY MONEY

>>106522167
Your Ryzen CPU idles at more than 40W compared to monolithic chips which are single digits yet I don't see you complaining about those.

>>106531502
My bad. What do the devs have to do to add that support? Is it in depth or fairly simple?

>>106531791
It depends. Sometimes they have just to add few if else. Sometimes hundreds lines of code. Sometimes there is no way, because low level things written in cuda lang and target cuda hardware, like sage attention.

>>106518987 (OP)
i would rather not buy a chip made by a company owned by the US government, tbdesu

>>106518987 (OP)
>sff
>takes up two slots

>>106531815
Oh right, so on that low level stuff is it like as long as it's written in Pytorch it's an easy port, but otherwise it requires more in depth work?

>>106532017
Sometimes. There is ABI still exists in Pytorch, like torch.cuda._C. But even plain torch.cuda to torch.xpu conversion might look like an easy port on paper, but in practice there could be wrong output, out of memory errors, performance drops, etc.
It's still very raw, but Intel had polishing AI software stack goal in q3-q4 on their road map iirc.

>>106518987 (OP)
there is no SLI for intel no? Kinda pointless to buy a intel card if $50 more gets you a 5060ti

>>106518987 (OP)
It has 200GB/s bandwidth
it's garbage
At least wait for the B60

>>106531466
Two more weeks. Jewtel still has over 20B in cash and short term assets. Over 150B in long term assets. The only nuclear scenario for Jewtel would be if they have to write off Intel 16A in 2027/2028 and then ditch their Fabs and go full TSMC which all of their current gen products are already on as a stopgap for Intel 16A

this isn't really relevant, but is there any distinction between implementing gpu architecture for arm vs x86 CPU in an SOC context?

>>106532798
secondary accelerators are ISA agnostic

essentially T4's but accessible to non enterprise, just feels like intels still 2 generations behind

>SFF
>dual slot
? Why is this thing even dual slot to begin with? It's less powerful than a B570 with more VRAM and a smaller memory bus

>>106520248
for reference a 12B Q6 quant fits nicely into my RTX 3060 (though I use Q4 quants right now)
So around 12-13B family would make the most sense
I'm personally quite interested in these cards myself, though more for "hmmm I wonder if these are viable for imagegen" because the Norwegian market fucking murders me with VAT and *any* lower price higher VRAM card that I could feasibly get-

>>106531918
>>106532924
I had the same thought, but apparently it's a half-height card. This picture with a full-height bracket makes it easier to judge the size.

>>106531113
It won't , but it also won't have the bugs/issues, and will work with everything. Paying the nvidia tax makes sense if you actually value your time, I would recommend an amd alternative over nvidia because compatibility is ok, better prices usually and you can usually get shit to work, but it doesn't look like anything from them exists yet in the same formfactor.

>>106538096
This argument no longer applies with Vulkan backends existing now. Paying 5x is no longer a worthwhile "Tax"

>>106538158
it won't be 5x the price, maybe double. this is still low end. the a2000 goes for ~$600-650, if you're not extremely poor this isn't a big issue.

>>106538158
>>106538308
The specs of the rtx pro 2000 blackwell put it close to the rtx ada 4000,

https://www.phoronix.com/review/intel-arc-pro-b50-linux/5
and if we look here , the ada 4000 is alot faster than the b50 in many areas, so if it comes in at 600-$700 that would be only a little more expensive perf/$ vs the b50. gonna depend on your workload whether its worth it.

>>106518987 (OP)
>16GB
>dual slot
DOA
all they had to do was slap 24GB in a single slot card

>>106531081
I wouldn't. Its not even close to proportional unless Nvidia suggests abandoning their current pricing formula entirely and choosing to attack intel on one very specific ground, with room to attack the B60 later.
That however would through the rest of their pricing out of line and skew it. It wouldn't make the Nvidia offering more popular in general or lead to a surge of nvidia sales, at most it captures the market where intel resides and intel's price comes down a little down the line and people no longer buy the nvidia offering.

Remember, the nvidia offering still has shit linux support. They'd have to drop the price significantly lower.

>>106538515
>provides benchmark
>bigger is betterer
>forgets that price is a factor
>forgets that VRAM capacity to $ ratio is the defining metric
>would only be a little more expensive
lel

>>106518987 (OP)
>WE NEED MORE GPU COMPETITION...
>... so Nvidia cards will become more affordable again

This is why gamers suck.

You should be aware that the linux driver is now developed by troons.

https://rosenzweig.io/resume-en.pdf

When Donald Trump will hear about that, he will definitively put sanctions on Intel.

>>106539466
>This is why gamers suck.
I'm currenting gaming with Intel's garbage UHD graphics so I'm an exception to this rule. Maybe one day I'll get a RX 7600 if I can find a good deal.
t. Linux gaymer

>>106539682
>the linux driver is now developed by troons.
Tranny's have been involved in the Troonix + GNU project for a while. This is old news.
>When Donald Trump will hear about that, he will definitively put sanctions on (((Intel))).
No he won't, the US is banking on those Intel fabs being completed in Arizona and New Mexico just in case Xi Jinping is dead serious about chimping out and invading Taiwan unprovoked as a means to re-integrate them.
Sadly they aren't doing this for the consumer market but fear of losing out on TSMC's fabs fueling the gAIy hype that I wish would die already.

>>106539682
Probably why it sucks, because they bring their political drama along side, which emerges during the course of any interaction with them.

Scores 7671 (graphics) in Time Spy with pre-release drivers, seems pretty solid for a slot-powered card. I have an RTX 3060 12GB in my server now, and it manages a 8913 graphics score for comparison. But it also eats 3 slots at full height and pulls 2.5x the power.

Seems like a really nice all-around server card. 16GB VRAM is plenty, low wattage, SR-IOV support, and gaming performance at about RTX 2060 level if you want a streaming box.

>>106540191
Wait...
If this is true (and applicable to actual things) then that's pretty damn good
I might try to grab one actually

>>106539960
>Xi Jinping is dead serious about chimping out and invading Taiwan unprovoked

lmao. you are aware that china has only two active aircraft carriers at most, right? the seventh fleet alone can stop china. Also, if they attack, Taiwan will destroy the Three Gorges dam (>200 millions deaths).

>>106540191
I am waiting for news on SR-IOV. Once that is confirmed working fine, I am 100% picking one up.

>>106538776
The B60 is 24gb and bigger bus width
B50 is a strange card to me.

>>106518987 (OP)
16gb?
how is it for ai?

>>106518987 (OP)
Does local diffusion work on this?
An old $50 Dell optiplex + this would make an amazing always-on SD server

>>106541912
works on existing arc gpus so it should

>>106541612
China doesn't need aircraft carriers to invade an island next door.

>>106541922
Also just saying, the A770 will give you much better LLM results for the same prices (or less by now). The only thing B50 does better is power efficiency.

>>106518987 (OP)
>>106518987 (OP)
>>106518987 (OP)
Will it work for kvm virtualization and containers? I know params for Nvidia and know similar exist for amd, but I think I never seen anything like that for intel.

Wouldn't it be seen as 2nd Integrated gpu considering a lot of skftware matches "intel" string with with igpu?

>>106539682
The competition is literally a Puerto Rican named James Jones who whines on all day about every single other person in the software community for not meeting his lofty standards at nvidia where they use AI to develop their graphics driver.
The man can't even get 2 back to back driver releases where gamescope runs on every mainstream card on a single GPU generation. The hurdle is low and he still falls flat.

Bring on the trannys developing drivers, they're at least autistic enough to be competent. That's better than Jones' aspergers.

>>106540191
Honestly, it's pretty appealing in the now.
In the UK it's about the price of the 9060 XT 16GB. You can obviously power limit it down to 70W and undervolt too but the size and not needing to use a better PSU is the big deal. If it were only a little cheaper at the $300 they announced it'd be a lot better a deal. Hopefully pricing drops a little before the B60 releases to what they'd announced it as. I'd have no issue being excessive and buying one of them with the goal of moving it to another server down the line.

>>106541971
Can you powerlimit arc down to a specific TDP like you can with RDNA on linux?
Nvidia has their 50% cap on PL, AMD allows for precise TDP limits on linux. If I could run the A770 at 70-100W I'd seriously consider it, I can get one for around 62-71% of the price of the B50.

>>106542061
Never tried desu so no sure.

>>106519229
Jews too busy shoveling gold into their bags on the way out the door.

>>106518987 (OP)
Raja is back to guide a company into making RX480 (performance tier wise) equivalent cards while the rest of a company tries to get their shit together. RIP Intel

>>106542772
I honestly forgot Raja was at Intel

>>106539448
>2x faster for 2x the price
> same vram
>supports cuda
>no cpu overhead issue
>fewer bugs

price difference is probably ~$300, that is small enough to justify it, if you're gonna built a stupid overpriced SFF system, why would you gimp it.

the correct thing to do instead of buying a overpriced POS, is to buy a cheap POS, you'd litterally be better off buying an a770 (16gb) and lowering the power limit to whatever the minimum is, and use a riser if you have a stupid SFF case.

>>106538916
intel just cut a bunch of their linux developers in their recent layoffs, so good luck, nvidia on linux can be a pain , but it works, and has for a long ass time. I used a geforce 7 on linux and the linux drivers were supported until like 2017

>>106518987 (OP)
For a second I read that as $150, then I realized it was $350. Waste

>>106542996
Worthless comment.

>>106519018
Not if trump has anything to say about it

>>106518987 (OP)
They discriminate against hiring White people