AMD blown the fuck out, murdered in the face, and buried.

File: amdvega.jpg (156KB, 633x758px) Image search: [Google]

156KB, 633x758px

>>58910409
DELET THIS

>>58910409
(You)
Intelshill on clockwork right now eh?

I've seen moms troll harder than your bitch ass.

Didn't realize Ryzen affected the Shill-Stations in India so badly, Intel must be putting you guys on overtime.

I'd like to go ahead and acknowledge that this is bait.

However, I might as well bite.

AMD's new Vega lineup was showcased at CES this year, and they demonstrated pre-optimization performance somewhere between a 1080 and a titan X.

Secondly, intel is running out of ideas. Each new generation is only a tiny performance increase over the last. I'm still sticking with them for now, but AMD is lined up to eat their lunch with their latest lineup. Of course, it's all speculation and nobody knows for certain what will happen, but AMD has a few more tricks up it's sleeve.

Next, as far as the ML scene, you could always just use, you know, CPUs. You aren't required to use CUDA acceleration, at least for tensorflow. Not sure about the others.

The only reason nvidia and intel are beating up AMD (and have been for nearly 10 years) is because when AMD actually was ahead, intel and nvidia both did massive licensing deals to prevent AMD from gaining any popularity.

File: nvidia-tattoo-09242010.jpg (57KB, 600x603px) Image search: [Google]

57KB, 600x603px

>>58910749
>Next, as far as the ML scene, you could always just use, you know, CPUs.

So you acknowledge our superiority.

If CPUs were good enough we wouldn't use GPUs, but I don't care what you think because you're a shit smeared AMD neckbeard manbaby gaymer faggot cuck whose mommie bought him a Radeon to play child porn games with.

Kill yourself.

File: Nvidia saddening.png (85KB, 800x726px) Image search: [Google]

85KB, 800x726px

>>58910409
t. scared of vega

>>58910650
>bitch ass

Remember the Riva TNT?

File: meme.png (118KB, 1907x628px) Image search: [Google]

118KB, 1907x628px

>>58910792
Kys yourself jewtel + nvidiot shill.

File: cntk-speed-comparison.png (15KB, 626x352px) Image search: [Google]

15KB, 626x352px

>>58910856
I don't care about price you faggot cuck. Maybe when you get a real job you'll realize that cutting edge technologies are worth an extra 100 bucks and no hike in your fire insurance.

File: nvidia-tesla-p100-pascal-graphics-100654537-orig.jpg (1MB, 2000x1278px) Image search: [Google]

1MB, 2000x1278px

>this triggers the poojeet gaymur baby

>>58910867
>cutting edge
See the pic related again
>>58910856

Didn't I see some shit that said new AMD cards will be able to run CUDA code?

>>58911479
http://venturebeat.com/2016/03/09/otoy-breakthrough-lets-game-developers-run-the-best-graphics-software-across-platforms/
http://techfrag.com/2016/03/11/nvidias-cuda-language-can-now-run-on-amd-intel-gpus/

>>58910409
Bro AMD is rising and I am making the bank investing in it. Come together and make money too m8.

Accept and will hurt less.. ;>)

>>58910749
I'm not OP but the point was that all the ML frameworks are built around CUDA, not OpenCL because CUDA is nicer to write in.

In other words, even if AMD's hardware is comparable it still doesn't have the same software support. Of course it doesn't help that Nvidia has been going balls deep into ML applications for years, arguably moreso than gaming.

Even if AMD starts supporting CUDA it will still lag behind Nvidia.

>>58910749
>>58910856
ML on a CPU will always be indefensibly slow. The advantage of using a GPU is that many of the operations lend themselves to parallelization and GPU's can deal with large amounts of memory at once (provided said card has plenty of ram and bandwidth).

>>58910409
explain why any of this matters to a home user.

To me this is like saying Apple is fucked because most servers run on windows and then posting "Hey MACFAGS, what's it like being completely excluded from the server scene?"

KYS

>>58910409
I dunno how did it feel to be left out of buttcoin mining?

>>58910409
I don't know what any of that means
Does it have 60 frames in Crysis?

>>58910409
>what's it like being completely excluded from the machine learning scene?

holy shit who cares?

AMD owns the console space, guess which space is more commercially relevant lol?

>>58913562
ML is, actually.

>>58913634
commercially? More than cucksoles?

no

>>58910409

>machine learning

It is what it ever was, a playground for phd's creating research for the round storage device.

>>58913673
You actually think there are appreciable profit margins in consoles?

>>58910409
When will AMDfags ever (machine) learn? Kek

>>58913773
I don't think AMD's making all those processors for free

>>58913673
It is. You just have no idea how widespread it is and how much money goes into it.

>>58913849
more commercial money in a cutting edge developing theory than there is in consoles?

ProTip: no

>>58913877
There are a lot of algorithms involved in machine learning. Even within just the subset of deep neural networks there are lots of different types and more constantly being developed/discovered. That said, there's a great wealth of research that is already well developed and being applied fucking everywhere.

In a timespan of a day you probably interact with more machine learning algorithms than you can count.

>>58913562
>holy shit who cares?

Companies who want to automate nearly everything (for example). So, all of them.

Margins are extremely high for ML because its so sought after and there are very few players in the market.

File: 1482048229419.jpg (139KB, 1000x1000px) Image search: [Google]

139KB, 1000x1000px

>>58913026
>most servers run on windows

>>58910409

if half of what you said is true then that's pretty pathetic since CUDA is nonfree garbage that only works for nvidia.

it would be pretty stupid to depend on such a low quality company like nvidia.

>>58913962
also good luck with FPGAs if all your garbage libs depend on CUDAids

>>58913962
You're not wrong but CUDA owns the market. Their proprietary linux drivers work better than AMDs and CUDA is nicer to work in than other comparable languages. Not to mention that Nvidia is doing a TON for machine learning while AMD is sitting around with it's thumb up its ass. It's really not surprising that almost all frameworks out there are designed around Nvidia, it's like AMD isn't even trying.

>>58914012
FPGAs are slow.

>>58910409
>Tensorflow? Same.

>what are TPU's

>>58913877
>what is Tesla
>what is Netflix
>what is Google
>what is Microsoft
>what is Apple

can go on forever tbhfäm

>>58912256
Swap "ML frameworks" for "DL frameworks" just to be safe

>>58913026
It matters only indirectly to a home user. Doesn't take away from the fact AMD is getting BTFO'd though

It's one market out of many AMD hasn't entered yet, AMD is in plenty of markets Nvidia has no relevance in as well.

Big deal.

>>58916521
The way I think about it is:

It's a profitable market that AMD could potentially have had a foothold in, but they basically ceded all ground to NVidia. So a huge wasted opportunity

>>58916549
Yeah, I agree.
But AMD is stretched too thin already, they should focus on markets they know they can do something since they don't have money to spend willy-nilly like Intel and Nvidia.

>>58916565
Yet they're still trying, which is odd.

I guess it's one of those things where you have to put resources into just so the competitor doesn't actually own the whole market, even if your own product sucks

>>58916521
>AMD is in plenty of markets Nvidia has no relevance

Such as(excluding the CPU market of course)? Asking out of curiosity.

>>58916605
Their entire CPU side of course, most notably embedded where they really shine until Zen wins some server market back.
Then there's miscellaneous markets like DRAM and SSDs, but they're rebrands but apparently still a source of profit,
All GPU markets are covered by both of them, machine learning is the newest hip one and Nvidia got to it first.
While before that was whatever market Radeon SSG was addressing like professional graphics that have to decompress a hundreds of gigabyte or larger CAD files.
Nvidia still hasn't entered this market, and I don't even know if you can call it a market, but the SSG concept itself is a pretty unique product with no current competitor.

>>58916725
>SSG concept itself is a pretty unique product with no current competitor.

It indeed is, I hope opencl replaces cuda. Vendor specific APIs should have been a thing of the past years ago.

>>58916725
Is it weird that these ML chips are called GPUs? And the non-graphical ones are called GPGPUs?

Like, even the TPU name makes more sense than GPU when applied to ML.

>>58917128
GPGPU is pure numeral calculation though, ML is not.
That's why GPGPUs are done with Teslas or the like with no video output.

File: alexandre-bissonnette-18.jpg (498KB, 1704x1944px) Image search: [Google]

498KB, 1704x1944px

AMD BTFO

>>58918513
Who is this cuckbird

>>58918537
pol's boyfriend

>>58917192
>GPGPU is pure numeral calculation though, ML is not.

Could you clarify? My understanding was that GPUs used in DL are already fairly "general-purpose", but it seems like GPGPUs go one step further?

>>58910409
>tfw nvidia fags won't get freesync monitors for cheap

>>58919070
>>58917192
>>58917128
I'm not going to read through the chain of posts, but GPU = Graphical Processing Unit and GPGPU = General-Purpose Graphical Processing Unit.

Modern GPUs are all GPGPUs, because you can run your own software on them (CUDA, OpenCL, <insert generic shader stuff here>). It has nothing to do with video output.

sup fags
structural engineer here. If I made an algorithm to solve finite-element method problems using GPGPU, would it be faster or slower than using the CPU?

>>58919676
Ah, there comes the confusion I guess. I know that CUDA and so on can already be run on regular GPUs. I just assumed that GPGPUs referred to entirely separately devices.

>>58919828
Not at all. In fact, the term "General Purpose GPU" is actually considered a bit archaic and has fallen out of mainstream use. I haven't seen it been used in ACM or IEEE publications for at least 10 years.

>>58919870
Yea as far as I'm concerned, GPUs nowadays means "Fast Matrix Multiplication (and other operations) box", rather than graphical processing unit. At least in the context of non-gaming/media.

>>58919627
"cheap" is a great word to describe AMD

>>58919900
Yes-yes, good goy, now buy Titan X.

>>58919891
Yeah, same. I have a bit more abstract point of view than you I guess, and see them as "moar layers" Deep Learning[c][r][tm] boxes.

File: deep_learning.png (374KB, 947x946px) Image search: [Google]

374KB, 947x946px

>>58919926
relevant

>nvidia bla bla bla bla ml bla bla bla bla
Wait until qualcom starts to become a threat to intel and nvidia, maybe then you'll come crawling back to amd.

>>58910409
https://github.com/benoitsteiner/tensorflow-opencl

>>58919818
If you can parallelize it, it would be much faster. Finite-element methods are solved by matrix multiplications as far as I know, GPUs are good at that.

Don't forget Volta which promises a 50-75% performance boost over Pascal

>>58921634
Cool 2018 Onii-chan.

>>58921658
Bitch, Volta will be out before AMD even pre-paper-non-launches Vega, they are dragging that shit out slower than a sloth with concrete feet.

>>58921673
You get Pascal refresh this year and maybe a single Volta HPC SKU by the end of the year.