Well, this is a system I keep upgrading for AI/ML shit. Its kind of funny collection of shit: its got an X11 SuperMicro server board which I put in this monster Digital Storm Aventum case I bought from a thrift store. I've actually got a second 3090 to go in it, but I have to use risers to bring both cards up to connect NVlink. To save VRAM (it really doesn't need a display anyways), I had thought of just using the motherboards graphics chip, but the I thought "well how can I make this more insane" and bought an FX5200 like the one I had as a youth to drive the display. The board is decked uo with some dumb 2TB Samsung top end NVMe shit and fully loaded with ECC ram, but it has another unoccupied PCI slot, and I figured I would find a floppy controller and drive or an PCI -> ISA bridge plus whatever esoteric bullshit I can find and procure or write drivers for.
What ever retarded nagger bought that Digital Storm system has it outfitted with an assloads of LED fans (or put them in), and a bunch of LED strips along with a big radiator. They're annoying as fuck, but now my GF and animals are afraid to go down in my basement since that shit glows harder and more violently than an angry FBI agent, so I guess it's a win.
Tell me you're a fucking youtube guy, this is complete insanity and just the type of LULZ shit that i actually love to see. None of the clean, sane, sterile builds everyone makes. This? this is true technology.
AMD is not catching up with ML. Their pathetic attempts at MLIR, HIPS, ROCm, and whatever the fuck are not going to pan out and none of them are merged into master.
They should have championed Vulkan. They should have written industry-leading papers on ML and open sourced the results. They didn't. Now they're left to rot.
Intel, I am not so sure about. They're so far out of the game that it's not so obvious what they want to do.
If Intel can release the battlemage series with out of the box support for text local models and stable difussion, Nshitia will be buttfucked so much their stock will hit concrete in less than hours.
They are catching up where it matters (CDNA) not for the gayming crowd (RDNA).
Nvidia isn't doing much for their gayming crowd either and are intentionally gimping their gayming SKUs to ensure nobody uses them as "poor man's Telsa" for real work. They are doing this via small VRAM capacity and narrow memory bus. Nvidia regretted releasing the Titan brand.
Oh fuck off, it's just another "nVidia fabricates another artificial GPU shortage" episode. At least I'm back on full team red and won't be affected. :^)
>Sir, the customers aren't buying the corona argument anymore, because corona is already gone for like 1.5 years, what should we do? >Just tell them it's due to AI. People have already forgotten the quarter report of TSMC in which they stated that they will have a reduced revenue in the following year because all the major GPU manufacturers have 90% full warehouses of unsold GPUs, so they will soon forget that this is all a plot to rake in more money from them
>the quarter report of TSMC in which they stated that they will have a reduced revenue in the following year because all the major GPU manufacturers have 90% full warehouses of unsold GPUs
Sauce?
Pic related is from the Q2 2022 Earnings Conference Transcript
>The semiconductor maker reported that customers were holding onto excess inventory, a factor that could have led to its record profits but also signal a looming adjustment, as predicted in May.
https://www.theregister.com/AMP/2022/07/15/tsmc_record_profits/
This is TSMCs quarterly report site. Check for yourself everything after Q2 2022. "Inventory adjustments" and "Inventory correction" on the customer side is a code-word for "Customer tries to reduce excess inventory".
>In China, where the company is offering a modified version of its A100 and H100 GPUs to bypass the US sanctions, known as A800 and H800, these chips are being offered at prices 40% more than the original MSRP.
Pajeettech at it again. That's not even what the original source says. >https://news.mydrivers.com/1/909/909778.htm
I don't know a single goddamn person buying a GPU so I suspect this is just apple or nintendo about to release a new consumable™ or someone is about to release a third party steamdeck
Could also be ai bullshit.
>down across the board >except Data Center >except Auto >expect actual Revenue
Besides Q1 FY23/Q4FY22 which were carried by miners they are doing perfectly fine
>amd and nvidia battle for the ai market >only a small amout of consommer gpus get made that are expensive, the rest of the is used for enterprise gpus
its unironicly over, our only hope is intel battlemage
Why need a GPU for AI? Because our approach to AI is a shitty brute force one maybe?
Before 2000, AI was all CPU. In late 2001, the Athlon XP launched. This does a whopping 270 MOps/sec on a single core according to cpu benchark.
Now, the recent 7950 does 4300 Mops/sec, almost 16x as much. And it has 16 cores at that, so 256x times the compute.
How much compoooot do you need for a text generator?
Thanks anon, so it is usable but I think this sentence explains it best. >Such [algorithm-intensive] tasks are not common in modern AI applications, given that most companies choose the efficiency and speed of GPUs over the specialization of CPUs.
So, everybody is too dumb for algorithms, so we end up throwing compooooot at it.
Honestly at this point I see Nvidia winding down gaming GPU sales eventually.
It's really clear they don't care about userbase growth but just extracting as much from the existing base.
LAMA models suck for anything practical. Until more advanced architectures push the field forward, RLHF is a necessity and practically all of them are closed-source and optimized for CUDA.
What are you waffling about?
Google has been producing their own chips for years now.
And you can bet your ass that other tech giants will do that too eventually.
Nvidia's AI monopoly essentially only exists because Alphabet allows it.
If AI chips start going up in price guess what they will do?
I have no use for a GPU
this feels calculated
of course it is. not that a 40 grand h100 was in any of our reach anyway. this is to cut off small businesses.
Nah they're just fucking with israelite
Yeah. Calculated to make a buttload of money.
Pretty sure Nvidia saw this coming at least five years ago.
OH NOOO, NOT THE GARBAGE 40 SERIES, WHAT WILL I DO
AMD doesn't have this problem
>AMD doesn't have this problem
Of course not. They have ample stocks of GPUs nobody wants to buy.
>can't buy a shit product
oh no
no one is buying 40series anyway
>Release an unbelievably shit card that everyone calls you out on
>Claim there's going to be a chip shortage so you should buy now before its too late
So Nvidia also manufacture false stock levels. Lucrative business I hear
after being a waiter™ i caved and have a used 3090
comfy
i hate DLSS 3 anyway, and FSR saved my ass so many times running my gtx1070.
i guess AMD earned a slot as my next gpu, 7700 when
>The reason the 4000 series is so expensive is because of the chip shortage that will happen in a few more months
Do nvidiots really?
Battlemage waiters don't have this problem
A770 user here, GPU it great, but driver is dogshit buggy. Everyday I have to reinstall it stop reporting that my GPU can do anything about DX9.
Yes. I've got both a 3090 and an FX5200 in the system I'm building, I should be set on GPUs for a while.
>FX5200
This one?
https://www.techpowerup.com/gpu-specs/geforce-fx-5200-pci.c1906
Yes
If you're wondering how, this system runs Debian. FX5200 gets Noveau drivers, 3090 gets Nvidia drivers.
I fondly remember playing doom 3 on that one in 320x240 with 10-15 fps
>3090 and an FX5200
Well, this is a system I keep upgrading for AI/ML shit. Its kind of funny collection of shit: its got an X11 SuperMicro server board which I put in this monster Digital Storm Aventum case I bought from a thrift store. I've actually got a second 3090 to go in it, but I have to use risers to bring both cards up to connect NVlink. To save VRAM (it really doesn't need a display anyways), I had thought of just using the motherboards graphics chip, but the I thought "well how can I make this more insane" and bought an FX5200 like the one I had as a youth to drive the display. The board is decked uo with some dumb 2TB Samsung top end NVMe shit and fully loaded with ECC ram, but it has another unoccupied PCI slot, and I figured I would find a floppy controller and drive or an PCI -> ISA bridge plus whatever esoteric bullshit I can find and procure or write drivers for.
What ever retarded nagger bought that Digital Storm system has it outfitted with an assloads of LED fans (or put them in), and a bunch of LED strips along with a big radiator. They're annoying as fuck, but now my GF and animals are afraid to go down in my basement since that shit glows harder and more violently than an angry FBI agent, so I guess it's a win.
Tell me you're a fucking youtube guy, this is complete insanity and just the type of LULZ shit that i actually love to see. None of the clean, sane, sterile builds everyone makes. This? this is true technology.
How many waifus per second can you generate?
until amd or intel catches up to the ML side of things, it'll be NVIDIA all the way for me regardless of the price
AMD is not catching up with ML. Their pathetic attempts at MLIR, HIPS, ROCm, and whatever the fuck are not going to pan out and none of them are merged into master.
They should have championed Vulkan. They should have written industry-leading papers on ML and open sourced the results. They didn't. Now they're left to rot.
Intel, I am not so sure about. They're so far out of the game that it's not so obvious what they want to do.
If Intel can release the battlemage series with out of the box support for text local models and stable difussion, Nshitia will be buttfucked so much their stock will hit concrete in less than hours.
also, if this happens, Intel will be the new gpuisraelites which could be worse
They are catching up where it matters (CDNA) not for the gayming crowd (RDNA).
Nvidia isn't doing much for their gayming crowd either and are intentionally gimping their gayming SKUs to ensure nobody uses them as "poor man's Telsa" for real work. They are doing this via small VRAM capacity and narrow memory bus. Nvidia regretted releasing the Titan brand.
AMD already won in the future
Oh fuck off, it's just another "nVidia fabricates another artificial GPU shortage" episode. At least I'm back on full team red and won't be affected. :^)
JUSTWAIT(tm) mode since 6 years
>cryptominer shortage
>pandemic shortage
>AI boom shortage
lololololol get fucked gamers
>PS5/SEX are around $470
get fucked scalper cuck
>paying $470 for sex
a blowie is like $50 dumbass
you overpaid
1080ti still going strong
3080 and 3090 are STILL in clearance here.
But yes, jenseng will israelite the AI for shure.
Yes I'll be fine, thank you. (I have a GTX 970).
>Sir, the customers aren't buying the corona argument anymore, because corona is already gone for like 1.5 years, what should we do?
>Just tell them it's due to AI. People have already forgotten the quarter report of TSMC in which they stated that they will have a reduced revenue in the following year because all the major GPU manufacturers have 90% full warehouses of unsold GPUs, so they will soon forget that this is all a plot to rake in more money from them
I may be subhuman IQ but I have the non midwit unconcious nose for sniffing out this kind of trickery.
>the quarter report of TSMC in which they stated that they will have a reduced revenue in the following year because all the major GPU manufacturers have 90% full warehouses of unsold GPUs
Sauce?
Srsly, please deliver.
Pic related is from the Q2 2022 Earnings Conference Transcript
>The semiconductor maker reported that customers were holding onto excess inventory, a factor that could have led to its record profits but also signal a looming adjustment, as predicted in May.
https://www.theregister.com/AMP/2022/07/15/tsmc_record_profits/
This is TSMCs quarterly report site. Check for yourself everything after Q2 2022. "Inventory adjustments" and "Inventory correction" on the customer side is a code-word for "Customer tries to reduce excess inventory".
>have a 2080ti, a 2060, and two 1050ti's
i have windows 7 so anything above a 3090ti is utterly useless and irrelevant to me so i don't care
I'm a 1080p fag, I own a 3060. I'm set until it breaks.
>upgraded from a R9 270 to a GTX 960 for free
>got my wife a 1070 a few months ago
Yeah we're set for at least 4 years.
Kek, they already told they are stopping the production of 4070 for a while because the card doesn't sell.
Already planning to buy 7900xtx so this won’t affect me. Upgrading from 2060, I can’t support Nvidia israeliness any longer.
I got a 4090, I was not worried about AI but by wars in HK. I made sure I have a computer I'm happy with for the next 5 years if needs be.
Nobody's fighting over HK
it's not about fighting, but about supply chains for best nodes getting utterly fucked
nvm read HK thought of TW, im tired
>In China, where the company is offering a modified version of its A100 and H100 GPUs to bypass the US sanctions, known as A800 and H800, these chips are being offered at prices 40% more than the original MSRP.
Pajeettech at it again. That's not even what the original source says.
>https://news.mydrivers.com/1/909/909778.htm
I don't know a single goddamn person buying a GPU so I suspect this is just apple or nintendo about to release a new consumable™ or someone is about to release a third party steamdeck
Could also be ai bullshit.
Nvidia is down across the board everything tanking. Hmm.
>down across the board
>except Data Center
>except Auto
>expect actual Revenue
Besides Q1 FY23/Q4FY22 which were carried by miners they are doing perfectly fine
I don't think you know what across the board means
intel bros now is our time
finally
maybe I'll be able to afford one now that they are gonna be 40% cheaper
and can upgrade my gtx3.5
>AI GPU
shut the fuck up you retarded animal
Can't wait for the AI hype to die down and Nvidia to go "there was no future in LLMs. they're just a waste of resources".
>tfw A770 guinea pig
How do you like it?
I have been thinking about buying one to play with
>amd and nvidia battle for the ai market
>only a small amout of consommer gpus get made that are expensive, the rest of the is used for enterprise gpus
its unironicly over, our only hope is intel battlemage
All my gaming is on Steam Deck now, a device that costs half of an RTX 4070.
seems like i'm keeping my GTX 750 Ti for 10 more years
Why are these nvidia chinks so scummy? Aren't there laws against this kind of blatant manipulation?
They mean AI chip shortage
Nvidia DGPUs are sitting on shelves right now
anyone can buy them
except poorfags
I stopped playing video games when I was sixteen. Perhaps you kids should grow up. You're what, in your late 20s and 30s? Kek
S-shut up
Shouldn't you be at work and have children by now?
why are you like this?
incest
i got a used rx 570 for 60 dollars off ebay and it works just fine kings
Why need a GPU for AI? Because our approach to AI is a shitty brute force one maybe?
Before 2000, AI was all CPU. In late 2001, the Athlon XP launched. This does a whopping 270 MOps/sec on a single core according to cpu benchark.
Now, the recent 7950 does 4300 Mops/sec, almost 16x as much. And it has 16 cores at that, so 256x times the compute.
How much compoooot do you need for a text generator?
shhhh stop noticing things, goy!
https://www.run.ai/guides/multi-gpu/cpu-vs-gpu
parallel processing
Thanks anon, so it is usable but I think this sentence explains it best.
>Such [algorithm-intensive] tasks are not common in modern AI applications, given that most companies choose the efficiency and speed of GPUs over the specialization of CPUs.
So, everybody is too dumb for algorithms, so we end up throwing compooooot at it.
My RX580 8GB still works.
>another shortage
Good. I just went through the last of my retail arbitrage money from 2020
not a consumer cuck, I'm ready for at least the next 10 years. get fucked goymers, not our problem.
now back to >>LULZ homosexual
>sales in the fucking dumpster
>WHAT SHOULD WE DO? REDUCE PRICES? INCREASE SPECS?
>no lol lets just fabricate another shortage to get those FOMO bux
Fuck I hate israelitensen with all my guts, what a terrible "man".
Honestly at this point I see Nvidia winding down gaming GPU sales eventually.
It's really clear they don't care about userbase growth but just extracting as much from the existing base.
i got my 4080 i'm chillin
Why do I need a GPU I can ssh into a cluster with 4 boxes each containing 2x A1000s from my cell phone.
Just ordered a mini pc with AMD Ryzen 9 7940HS APU (Radeon 780M iGPU) both for shitposting and gaymin'.
Not my problem
Already have my XTX, so I'll be set for a few years
imagine all this ai dotcom bursting overnight
LAMA models suck for anything practical. Until more advanced architectures push the field forward, RLHF is a necessity and practically all of them are closed-source and optimized for CUDA.
What are you waffling about?
Google has been producing their own chips for years now.
And you can bet your ass that other tech giants will do that too eventually.
Nvidia's AI monopoly essentially only exists because Alphabet allows it.
If AI chips start going up in price guess what they will do?
Goldrush and shovels.
Why do I have a feeling that my GTX 1070 Ti is going to be very valuable in the future.