First vega cards are pro sadly but we fucking called it

first vega cards are pro sadly but we fucking called it..
1.5ghz monsters
videocardz.com/64677/amd-announces-first-vega-accelerator-radeon-instinct-mi25-for-deep-learning
50% more tflops than fiji

Other urls found in this thread:

images.nvidia.com/content/pdf/tesla/184457-Tesla-P4-Datasheet-NV-Final-Letter-Web.pdf
blogs.nvidia.com/blog/2016/11/16/tesla-p100-on-google-cloud-platform/
cloud-computing-today.com/2016/11/28/1075598/
amd.com/en-us/press-releases/Pages/amd-radeon-technology-2016nov15.aspx
cloudplatform.googleblog.com/2016/11/announcing-GPUs-for-Google-Cloud-Platform.html
anandtech.com/show/10209/amd-announces-firepro-s9300-x2
anandtech.com/show/10905/amd-announces-radeon-instinct-deep-learning-2017/2
youtube.com/watch?v=4_oI1K6rt48
twitter.com/AnonBabble

>passively cooled.
Wot?

Welcome to data centres.
Passively cool everything. Then slam the air temperature down to below 0.

50% more than fiji..if amd found a way to actually utilize every single bit of the shader output and not fuck it up like fury i bet the consumer cards will be monsters...

These are put in servers with a bank of powerful fans at the front of the chassis that push air through the case.

You're a fucking idiot if you think they cool the air to below 0.

Gtx 1080 is a 180w tdp card usually pulling 173watts and has a 9tflops of performance. So, 400tflops/9=44.4 1080s. 44,4x173= 7688watts from just the gpus...and a fully gp100 is even worse

can you imagine if that specific vega comes out on sale? literally double the perf of a 1080p for less than a double of its power..

a true 4k beast

hallo

>9tflops
In FP16 or FP32?

The Falconwitch is claiming 400tflops from 16 of the Mi25 cards.
So we've got 25tflop per Mi25, 16 of the fuckers does put it at 400tflop. Assuming they run at the stated TDP, then those 16 cards are pulling 4800watts.

Depending on the software packages AMD can ship this thing with, it could see some action.

>300W

AYYMD HOUSEFIRES

their double is 4.4 4.5 so 9 is single
more than double the horsepower less than double the power req..

Tesla P4 is 75W, it is significantly more power efficient than any AYYMD HOUSEFIRES junk

images.nvidia.com/content/pdf/tesla/184457-Tesla-P4-Datasheet-NV-Final-Letter-Web.pdf

>Nvidia 75W 5.5 TFLOPS

>AYYMD 150W 5.7 TFLOPS

AYYMD IS FINISHED & BANKRUPT

>Tesla P4
are you actually comparing something with 5.5 tflops fp16 with something that has 25 tflops? fp16?

>Comparing AYYMD'S TFLOPS with Nvidia TFLOPS

We all know AYYMD's TFLOPS are meaningless, Fury X gets killed by GTX 980 Ti and killed by GTX 1080

Yes.
Yes he is.
He's surprised to learn that something with nearly 5 times the performance, consumes 4 times as much power.

you literally compared a p4 with an amd card moron and you call me out for doing the same? KEK

AYYMD TFLOPS are WORTHLESS

They get killed by superior Nvidia power efficient GPUs

nvidiots right now

Its too late for any damage control. Nvidia already covered said market and AMD is as always too late to get any relevant share. Feel free to read Nvidia financial report, their professional class division boomed like supernova.

>BUTTMAD AYYMDPOORFAGS DETECTED

considering the wins amd has done those past months id say a big NO especially when they snatched google

Are you stupid or retarded?

AYYMD didn't snatch anything

blogs.nvidia.com/blog/2016/11/16/tesla-p100-on-google-cloud-platform/

STAY MAD, CLUELESS AYYMDPOORFAGS

>when they snatched google
Did I miss something here?

I knew AMD secured an IP lease deal with Intel, should see some reports of that soon enough out of AMD.
And the SSG seemed to attract some attention, but I didn't hear anything about any google deal.

AYYMDPOORFAGS, not very smart people

Nvidia has won the market while AYYMD won't ship anything until next year

cloud-computing-today.com/2016/11/28/1075598/
amd.com/en-us/press-releases/Pages/amd-radeon-technology-2016nov15.aspx
fucking nvidiots dont even know what snatched means..

blogs.nvidia.com/blog/2016/11/16/tesla-p100-on-google-cloud-platform/

DEBUNKED

STAY MAD, AYYMDPOORFAGS LIARS

>Promises
>Will be available

Typical AMD with their "its shit now, but might get better in future" shit

It seems you're all unable to read
>Amazon Web Services, Microsoft Azure and the IBM Cloud, for example, use Nvidia GPUs. Google’s decision to use a combination of Nvidia and AMD GPUs for its data centers empowers Google Cloud to avoid vendor dependency and obtain greater negotiating power in the procurement process.

Google is going with both m80s.

god damn it its like talking to a wall built by gtx480 ready to catch fire
cloudplatform.googleblog.com/2016/11/announcing-GPUs-for-Google-Cloud-Platform.html

No, he's the one that can't read

Google uses Nvidia, but he said AYYMD snatched Google, which is totally false and wrong

Smart companies uses Nvidia, not even Microsoft which uses AYYMD HOUSEFIRES junk in their console uses AYYMD for mission critical apps like datacenters

anandtech.com/show/10209/amd-announces-firepro-s9300-x2

>FP64 0.8 TFLOPs (1/16)

AYYMD IS FINISHED & BANKRUPT

Are you that big of a brandfag? You must be paid to shill that much otherwise you are fucking stupid and have nothing to do all day

so basicly even google is lying?kek such a massive idiot

talkng about vega posts a link about fiji from march

cant be more nvidiot

You're the one that can't READ

AYYMDFAG liar claims Google only uses AYYMD's for datacenter which was thoroughly debunked

Google uses Nvidia way more than AYYMD's worthless FP64 0.8 TFLOPs, but you AYYMDPOORFAGS have to lie to save your worthless dying, bankrupt company

I'm seeing some major salt here.

Is this a FP64 GPU?

All shills or the same Nvidia Shill

>Sup Forums is so relevant to market that it has shills

Keep dreaming you autistic snowflake

300W for a double precision card is completely normal, the P100 is 300W as well.

Consumers will probably get one with FP64 cut to 1/16 or 1/32, less cores + higher clocks.

Primary focus seems to be on FP16/32, we've got limited numbers at the moment. I imagine we'll see more details in the coming weeks. If not sooner if AMD get some large preemptive orders.

>b-but my baseless statements are more valid than the values stated by both manufacturers

>AMD and Nvidia use retarded graphs that start from 200 and increment by .10 so a 10% lead looks like a 200% lead for the consumer market
>but use normal graphs for professional markets that don't look inflated

Really makes you think, they think we're idiots.

i said snatched but ok i know that nvidiots are total strangers to the word truth afterall they keep buying cards like 970

>I base GPU power off gaymen benchmarks

so, no 490.
how much of these numbers will survive consumer transition?

Zen confirmed for for 36 PCIe 3.0 lanes? Is my math wrong?

>they think we're idiots.
And for the most part. They're right.
I mean, fuck. Look at Sup Forums. They gobble that shit up.

>Implying 12 year olds that lap up whatever nvidia shits out aren't

It's likely running with a number of xeons. When pushing new professional GPUs AMD tend to pair it with a xeon.

>Dying bankrupt company
Is that why the stock is up over x3 this year?

...

it's Napels 32core

Depends on what features are packed onto the PCB.

This is good news for their future consumer cards right? I don't even play video games and I've never bought a GPU but I've always been interested in the technical side of this shit for many years.

Nvidia basically having a monopoly on the high end consumer GPU market isn't great for even their own customers in the future.

they do it for free

Is it a 4 die MCM with 8 cores or a 2 die MCM with 16 cores each?
Does every die have the southbridge enabled? Is that even necessary? For most setups that don't need a GPU/accelerator or SATA ports that's just a waste of power.

>75watt nVidia
>150watt Radeon
>Four times the power

75+75=150... So five times the performance at only double the power.

He was shitposting in regards to the Mi25. The 300w card.

I found some more leaked benchmarks.

You sure? These are ROPless cards so no display outputs, these are Tesla competitors and Teslas have historically been primarily FP64

Not sure I follow m8. Mind expanding?

they arent ropless they are full fledged gpus they only lack a video firmware

You're a fucking moron

He's comparing it to the 5.7TFLOPS 150W card

AYYMD'S power efficiency is pure garbage and it shows

Every full fledged FP64 GPU lacks ROPs or has them disabled for obvious reasons, no display output = no need for ROPs

nice post pajeet

this is the firstyear nvidia actually had 2 lines of production...
amd never actually creates a specific hpc chip (although this could very well be the first time for amd also)

Accelerators have no need for them, if they didn't remove them, then common sense dictates that they'd at least disable them.
Little reason to keep them since there's nothing for them to connect to.

But... Why? That doesn't make sense, they're not designed to compete. That's like mocking a BMW m5 for using more fuel to make more power than a Honda accord...

It says less than 150 though.

>That doesn't make sense
It doesn't have to.
One number is bigger than the other, that's ground for relentless unadulterated shitposting.

cut down the cost obviously but yeah they are obviously disabled at least

AMD tends to soft disable hardware through the card BIOS. Nvidia and Intel laser cut. So it's technically possible that you could get the AMD crap to work as graphics cards if they are based on the same die. Assuming you can attach an output interface.

They have no output ports. No idea if they have the soldering points for the outputs either.

What is condensation

GP100 has ROPs

They get away with passive cooling because they have decent airflow and AC units. Your below 0 comment is pants on head retarded though.

Do these new graphcs/slides show 20+tflops anywhere, I can only see the bar graphcs which look like they're still giving teh fp32 numbers. Especially if the mi8 is a fiji card and those don't expose fp16 I think.

>Nvidia chip is 300W and 10.6 TFLOPS
>AMD chip is

Nvidia TFLOPs are 40% faster than AMD TFLOPs you stupid fucking shill

So this is the reasons for the stock jump 2 days ago.

what

YOU HEARD ME

Mi6 = 5.7TFLOPS = Polaris
MI8 = 8.2TFLOPS = Fiji
MI25 = ~25TFLOPS (16bit) = ~12.5TFLOPS (32bit) = Vega 10

Polaris is more efficient with its TFLOPS than Fiji. Vega is likely to be more efficient, again.

muh dik

Based AMD saving us again.

"AyyMD" Nvidiot permanently BTFO.

I hope you all bought at 1.6.

>nvidia still 8 times bigger

Lel

It just jumped to 11

Fuck. Knew I should have went long on AYYMD when they were in the trash.

>AMD shares 920k
>Nvidia shares 530k

Hurf, Nvidia is currently overvalued.

My point is I can't see anything in the slides stating what the actual performance is. 1.8x maxwell titan is it, a card which doesn't even have fp16 capability.

The 25tflop number is still just coming from earlier rumours.

These numbers are also pretty close to an also rumoured dual polaris 10 card clocked a bit higher, 512gb/s memory is just double again,

>Bought at 1.70
>Sold at 2.90
>tfw
Never try to time the market. If you think a stock is going to rise long term, stick with it.

the slide literally says "MI25 Vega"

It's in the name.

anandtech.com/show/10905/amd-announces-radeon-instinct-deep-learning-2017/2

Vega in Doom 4K ultra settings

youtube.com/watch?v=4_oI1K6rt48

Around 55% faster than a Fury X then.

nShillia is finished

about 50% faster than Fury X, which would put it at about 15-20% faster than a GTX 1080 and slightly below a Titan X.
Fits with expecations. Only question: is that full Vega or cut down Vega?

the real question is it $400 card or $600 card

$600 nobody would actually care except people that buy amd anyway, they need to break market

>implying that the titan X is more than 20% over a 1080

>Under 300w
>12.5TFLOP

So big Vega sure as fuck isn't a 64 CU design like previous Videocardz "leaks" claimed. A clock of 1500mhz would only be 12.28TFLOPS.

There are no 16 core dies. Everything is based off a single die design called Zeppelin.

Naples is 4 Zeppelin dies in an MCM.
Summit Ridge is a single Zeppelin die.
Snowy Owl is two Zeppelin dies in an MCM.