NVIDIA TITAN VOLTA With NVLINKIS

wccftech.com/nvidia-titan-v-volta-gpu-hbm2-announcement/

>110 Teraflops
>9x that of its predecessor

nvidianews.nvidia.com/news/nvidia-titan-v-transforms-the-pc-into-ai-supercomputer


also the rumors was real
wccftech.com/nvidia-titan-volta-nvlink-hbm2-graphics-card-leak/

Other urls found in this thread:

guru3d.com/news-story/nvidia-titan-v-will-not-support-nvlink-and-does-not-support-sli.html
sheepit-renderfarm.com/
wccftech.com/nvidia-titan-v-volta-gaming-benchmarks/
twitter.com/NSFWRedditGif

>actually citing the tensor meme performance
Go shill somewhere else.

>poor volta

Kek

>$3,000
>new titan that didn't replace the old titan slot
>a whole new titan tier
why do i feel like volta is NOT coming out for the "mainstream" consumer any time soon?

WOW, imagine 50 of them in server

How huge is the IC?

I love nvidia just laying their dick on amd

>HBM2 memory
get ready for more housefires

Yeah, I'm sure the only way AMD will be able to attempt to compete is by raising their power draw even higher.

?

Guess I'll do a "quick & dirty" estimation on performance:
>Titan V = 5120 CUDA Cores / 1455 MHz Boost Clock / 652.8 GB/s Total Memory Bandwidth.
>Titan Xp = 3840 CUDA Cores / 1582 MHz Boost Clock / 547.6 GB/s Total Memory Bandwidth.
>33.3% increase in CUDA Cores / 8% Decrease in Clockspeed / 19.2% Increase in Bandwidth.
>Total Estimated Possible Performance Increase = 44.5% faster than Pascal Titan Xp.
Meaning in this Benchmark (Witcher 3 Ultra Settings @ 4K) pic related:
>Titan Xp scores 59.4 fps average.
>Titan V would score 85.6 fps average based on these numbers.
Mind you this is without an overclock on HBM or core, and additionally a low-end average estimate. Performance increase could be anywhere from 35-50%, 40-45% being about the average performance increase when coupling memory bandwidth and CUDA core increases.

that's actually pretty awful if the titan v will only be ~40% faster than the 1080 ti / titian xp. if we cut that down for consumer parts i don't see any sort of major increases at all. it will be a pascal over maxwell all over again......

my guesses would be:
>1060 replacement
"1160" will get something between 1070 - 1070 ti level of performance
>1070 / 1070 ti replacement
"1170" will get something around 1080 ti level of performance
>1080 replacement
will get something +20-25% over the 1080 ti

that's really nothing exciting. at all.

anyways $3,000 is utterly ridiculous. anyone who buys this for "gaming," this is why people want communism.

It's a Titan though

The Titan is certified hardware on a lot of professional apps. It sits between the Quadro and gaming line.

>It's a Titan though
so?

>he thinks performance scales linearly with clocks
smuganimugirl.jpg

well i was a shit poster back before pascal launch telling people pascal wasn't going to be a big leap over maxwell and people like you shat all over me.

and i was right. so smug anime girl all you want, prepare to be disappointed in volta mainstream consumer line.

Shouldn't you take into account any increase in performance from the new architecture?
Granted, we have no idea what that figure is, but still.

>110 Teraflops
Only for single precision, aka "ai applications". Games don't use that. But the sheer number of cores should give a decent boost.

i could, and i did to a degree, but nvidia hasn't done big leaps from pure architecture alone since keplar over fermi. with fermi being the true last big from fermi over tesla. maxwell was mostly reduced power consumption + higher clocks over keplar, and pascal was nearly the same. small bump in cuda performance + higher clocks, further power reduction, and better async performance to at least not suffer a regression.

looking at volta, all the new architecture improvements seem to solely fall on improved performance for ai / deep learning / pure compute. tensor will be all but useless for gaming. but nevertheless i did. with the 50% estimate. so change 40-45 to 50-55.

how many bitcoin can this mine for me?

Its literally available to order now you retarded fucking troglodyte. Why is everyone here so god damn stupid?

it is ~15 TFlops for 32-bit operations, the 110 TFlops is for machine learning application

3k is crazy, but the die is over 800m^2 on 12nm sitting on top of interposer with HBM2. The margins are still large, but that is not a cheap die to produce.

Zero. Pic related is a mining farm that only gets a couple of bitcoins a day now. You have to buy bitcoin.

>Only for half precision
FTFY

>9x that of its predecessor

you mean the titans that didnt had any matrix computing abilities?

LOL

>look ma', i'm a raging idiot xD
>i post on 4chaners.org ma'!
>i so edgy xD xd XDDDD!!!1

cause we all know that there is a pc out there that has nvlink on it RIGHT?

fucking idiotic shills mate

>NVLINKIS
Yeah, another retarded marketing term that means exactly nothing.

Also, lel grabiks cart

>its
How many its faggot?

Volta-Powered GPU Delivers 110 Teraflops of Deep Learning Horsepower — 9x Its Predecessor — to Researchers and Scientists
Conference on Neural Information Processing Systems — NVIDIA today introduced TITAN V, the world’s most powerful GPU for the PC, driven by the world’s most advanced GPU architecture, NVIDIA Volta.

Not for gaming.

Educate yourself

x9

because titan x had matrix cores as we all know

fucking hell that nvidia pr really needs to stop making idiotic statements

Where is the HBM on the PCB?

sad thing is the current pc gayming market sees a 20% increase as maasive

Normies praised pascal even though the performance increase was shit over Maxwell (1060 replacing 970 etc due to the price hikes)

>110 Teraflops
>9x that of its predecessor
half precision is irrelevant shit

pascal was absolute horseshit. never understood why its gained so much praise.

1060 being an overclocked 970 was nothing to brag about

1070 being an overclocked 980 ti was nothing to brag about

the 1070 ti was beyond stupid and made no damn fucking sense. it hurts nvidia's own 1080's than vega 56.

1080 being 20-25% faster than the 1070 was absolute shit and worse value out of all the pascal lineups.

the 1080 ti is only a good value because of how shit of a value the 1080 is.

then it doesn't help founder edition cards being $100 over "msrp" drastically fucked up retail pricing of all series.
>hurr but i can find a few 1070 for $380 if you look around you can find!
wow, 3 cards out of how many fucking 1070 models? before the founders edition shit you could find, take the 970, the "best selling" midrange card to date for example, for under $280 - $320. under msrp. but thanks to founders edition and the amount of raging retards that bought it, it showed both board partners and retailers you're all fucking retards.

i mean just look at pic related i just took a second ago. where are those $350 fucking 1070's?? fucking drones. all of you.

This is a new arch, it is not pascal which is basicaly maxwell shrunked to 16nm.

>look ma'
>i'm an idiot on 4chaners xD xD xD xddddd

nVidia SlideWorks™ forgot to render CoWoS interposer + HBM2 stacks.
That happens.

Is NVlink physically compatible with PCIe?

Energy efficiency, but I didn't like the temps, 80ºC as normal operation is bulshit. What sucks at pascal is the ridiculous high price for a mid range chip, the 1070 was cut down was ridiculous and not giving the 1060 SLI.

So
Volta is pascal + tensor cores.
That is disabled features for plebs.
inb4 maxwell ++ 2080Ti.
Don't get me wrong, it will still be good.
But the real question is, how overpriced will it be?

pfft, it only cost 0.15 BTC

oh please don't give me that energy efficiency bullshit. majority of people don't give a fucking two shits above energy. nvidia can release another fx, tesla, and fermi and people will still have explosive orgasms over it and act like its the best thing since slice bread. if it has the nvidia logo, people will buy it, even if its pseudo gold plated rotten, std infected feces.

people only bring up energy efficiency just to shit on amd. back when amd had the better performance per watt and cared about power draw no one gave a shit. having a 480 that was 50-$100 more cooking an egg for 5-10% faster performance over the 5870 was a badge of honor to these fucking drones. back when microsoft themselves admitted during vista and early 7 days that 30% of bsod where caused by nvidia drivers and amd actually had more stable drives no one gave a fucking shit about drivers.

nvidia can literately slap on a 5-$10 monthly fee to geforce experience and they would have overwhelming praise and adoption by their user base over night. only a small, vocal minority would complain but nvidia knows full damn well that it would just be a minority.

>it's ok when Nvidia do it

That would make an epic 3D CGI render farm.

I think they meant Ethereum.

>110 Teraflops
those are not the teraflops you're looking for. nvidia has already stated, "tflops" is now considered tensor flops for their measurement.

yes and its utterly pathetic. when nvidia finalizes its monoply only then i hope these fucks realize the fucking mistake they made.

it wasn't because amd couldn't into drivers
it wasn't because amd couldn't into power efficiency

it was because no one gave a fucking damn under nvidia cared. amd did all of those things long before nvidia and no one gave a shit until nvidia did it themselves.

>7970ghz faster than the 680
>oh who cares. its only 5-10% difference. anyways 680 uses less power!
>so?? the 6970 was as fast as the 570 and used less power but no one bought the 6970.
>the 4850 was faster than the gtx 250 and used less power but nvidia sold more gtx 250's
>the 5870 was 5-10% behind the gtx 480 but used drastically less power but no one bought the 5870
>and back then driver quaility was pretty equal
>yeah, but its not nvidia. you need cuda for games. techpowerup says so.

AMD just needs a better marketing team.

yeah they need to release pimp my computer videos like nvidia did and stop it with the stupid communist 1% radeon uprising crap and pic related.

Hbm2 runs cooler and uses less power than ddr5 and ddr5x.
That's part of the reason AMD used it.
Imagine the Vega 64 card with 50w more power usage

AMD's HBM2 is oversalted at 1.3V and probably isn't giving them shit for power benefits.
Hynix yields must be pretty awful, thankfully with Nvidia, Intel and AMD using it it should improve.

Why

do

you

type

like

a

retard?

110 TENSOR flops you dense mother fucker

That's Reddit spacing man.
The Reddit fags subconsciously type like that.

..but can it run Crysis?

>According to Nvidia themselves, "Titan V does not support SLI or NV Link"

guru3d.com/news-story/nvidia-titan-v-will-not-support-nvlink-and-does-not-support-sli.html

mostly to trigger autistic twats like you

...

Get a V100 for $10000

it was confirmed that nvlinks are there because it use same pcb as quadro and tesla, but you cant use it.

*Breathes in...

AHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA!!!!!!!

...

Ah, the "I now have a life" card, funny will it be when he comes back in a few months saying how "life" wasn't as fun as he thought it was.

exactly. he'll be back and a full supporter of mgtow when stacey chops off his balls when she finds a bigger, blacker chad.

>using a TV as a monitor
>using a fucking OLED TV as a monitor

I do this, once you go with the deep blacks, you can never go back.

so does this mean that an 1170 will make a 1070 look like a chump?

Probably, Nvidia wants to be first in AI, so the fast trickle down to GPUs will accelerate in the next 5 years.

So I had a hd 5770 then a HD 5970 and there NO way Nvidia drivers were worse than AMDs at the time.
That fucking series NEVER ran right
I've never encountered a more crash happy Series.

Why do AMD arguments always devolve to bringing up GPUs that are 7+ years old?
your arguments not valid for 2017

>1170 will make a 1070 look like a chump?
That happens every generation.

see
/ 1070 ti replacement
>"1170" will get something around 1080 ti level of performance + / - 5-10% on top of that
>it be another pascal over maxwell - maxwell over keplar all over again.
from reading more about titan v and volta, it really seems like average for gamer's in terms of increases. all the new arch stuff in it is all focused for compute / ai / deep learning stuff. once you start removing that crap like tensor cores that's meaningless for gaming, all you're left with is increased cuda core count. all the under the hood stuff for the core itself was all improvements for their context switching for async, reduce power consumption primarily for the server / professional market, and better compute. very little for actually gaming. so all the stuff that would benefit gaming is pretty much cuda core count as more cuda cores = faster gpu. like its been with nvidia since cuda came out.
also don't expect hbm2 on consumer, gaming cards from nvidia. the ONLY card i can see getting that treatment is the gaming oriented titan for the $1,200 teir and MAYBE the "xx80 ti" variant. just MAYBE. and if they do, you will pay for it. nvidia isn't here for charity. most likely consumer gaming cards will get gddr6 on the higher end and possibly gddr5x on the midrange - low high end.

>volta is a supercomputer/ai gpu you fucking mongloid

that pic was used when EVGA GPU cards where catching fire

I use this and also render my 2 server pcs here
sheepit-renderfarm.com/

>building a pc
>not real life
>you build it digitally, I guess?

Instinct will rip this apart when released. That's it.

Amd doesn't care

>t. triggered redditard.

C7 is incredible as a monitor, input lag is basically on par with a standard 60Hz monitor and the GTG Response Time is 0.1ms

I guess it's time to sell my 1060 6g

you can't be serious can you? people have been spacing posts like that since the beginning idiot

hello redd!t

>815mm2 die
what the FUCK how are they gonna cool this?

>So I had a hd 5770 then a HD 5970 and there NO way Nvidia drivers were worse than AMDs at the time.

AMD had bad drivers too but they quickly fixed them where as Nvidia took over a year.

Underclocking, chocolate syrup.

You know that NVlink isn't enabled on Volta Titan and tracings for it aren't there?

You need to get Tesla Volta if you want NVlink.

That's really the only difference between Titan Volta and Tesla Volta

dear god its fermi all over again

wccftech.com/nvidia-titan-v-volta-gaming-benchmarks/

>At stock, the card pumps out a GPU score of 32,774 points and when overclocked, the card delivers almost 36,000 points. A stock Titan Xp which uses the full Pascal GPU configuration has a graphics score of 28,000 points while the GTX 1080 Ti ends up with a graphics score around 28,300 points. So yeah, the card was faster in stock to stock comparisons and is just a whole another level when overclocked

back2school with you

>only 26% faster than a stock 1080 Ti when overclocked
>overclocked 1080 Ti gets ~31000 points
>15%
>much faster
>$3000

That's isn't that great at all, I might skip this gen if we're only looking at around 15% extra performance in games

This card is like the 1st Titan: a pure compute card

The card is a classic PCIe card.
NVlink is replacement for SLI bridge.

>110TFLOPS
>TITAN V’s incredible power is ideal for developers who want to use their PCs to do (...) high performance computing.

not so fast, half-precision is virtually useless

Not designed for gaming, but you can game on it because it is still a graphics card.

Just like you can game on Quadro.

>NVlink is replacement for SLI bridge
Not really.
It's not designed for gaming and it isn't for gaming.

And allegedly the TitanV, while having the NVlink headers, actually has NVlink disabled.

>only gets a couple of bitcoins a day
>bitcoin over 15 grand
>it only generates ~50,000 USD a day

>Can't into gaming SLI
I do not want this piece of shit!