$5 next week buy now or forever be a poorfag

.35

30k brainlet who just fomod

Just sold my stack for LINDA. No regrets.

By way of background, I'm a postdoctoral research fellow at Harvard developing novel neural network architectures and training methods primarily for computational medical imaging.

Deep Brain Chain aims to distribute neural network training over computers worldwide. However, one the biggest problems in distributing training, whether over multiple GPUs or multiple computers, is communication bandwidth. Even when training is done on a single machine (let's call this Distribution Level 0), data needs to be transferred between the CPU to the GPUs to transfer training data and iterative gradient updates. If your model is reasonably big (for example, I regularly work with ~10GB models), that communication is happening after each training batch and is often the main bottleneck for training time. PCIe 3.0 x16 supports ~16 GB/s max transfer rates which means I'm waiting about one second each iteration JUST for data transfer... this transfer delay is one of the primary reasons why you might hear deep learning models taking days/weeks/months to train. This is such a big problem that NVIDIA has developed a specialized I/O bus called NVLINK (which will improve those speeds 5-10x but is only currently implemented on IBM Power systems).

Parallelizing this training over multiple computers (Distribution Level 1) typically occurs in a local cluster or a cloud system that is intraconnected with 10-100 Gbit/s ethernet (AWS is 25 Gbit, or about 3GB/s). Even though you're adding more computers, this 3-5x speed transfer loss, in addition to algorithmic losses due to imperfect parallelization and asynchronous updating etc (en.wikipedia.org/wiki/Amdahl's_law), results in a poorly-scaled system, which is a widely-studied issue in the research community (e.g. arxiv.org/pdf/1609.06870.pdf). You'll still get gains better than single-machine, but they won't be linear (10 machines doesn't equal 10x compute speed), and it'll be worse the slower your transfer bandwidth

Let's say 0.45 next week. Now that's reasonable. 5$ for next summer. 10$ EOY

...

>10 EOY

I wish i wish in reality it will be more like 3$

BRAINY GANG, BRAINY GANG, BRAINY GANG, BRAINY GANG, BRAINY GANG, BRAINY GANG (BRAINY GANG!)
SPENT .00014 ETH ON BRAIN CHAIN
MY DBC LOVE DO MOON GAINZ, OOH
BOUGHT THE DIP, FUCK THE WHALES THIS IS NOT A GAME, YEAH
CAN’T BLAST THRU SELL WALLS WITH THESE BAGS MANG, AYE
BRAINY GANG, BRAINY GANG, BRAINY GANG

MY STACK COST MORE THAN YOUR RENT, OOH
WHALES GOT ME SELLIN MORE THAN I SPENT, YEAH
STILL SHILLING POSTS ON Veeky Forums, HUH
WATCHING FOR THOSE GREEN CANDLE STICKS, OOH
ALL OF THIS SHIT BE NEW TO ME
BOUGHT ON KUCOIN, WHEN WE MOONIN, G?
FUCK YOUR MEME LINES WHEN THE WE LIVE ON HUOBI?
BITCH YOUR BREATH SMELL LIKE BOUNTY0X
I’D RATHER FUCK PAJEET FROM THE PENNY STOCKS
THESE BAGS HEAVY LIKE SOME TRON ROCKS,
CHINK COINS MAKE ME WANT TO TURN THE GAME OFF
DBC STILL SELL THAT ETH
BRAIN STACKS TO THE SKYNET AI TECH
FUCK A TAY AI MAKE HER PUCCI WET

BRAINY GANG, BRAINY GANG, BRAINY GANG, BRAINY GANG, BRAINY GANG, BRAINY GANG (BRAINY GANG!)
SPENT .00024 ETH ON BRAIN CHAIN
MY DBC LOVE DO MOON GAINZ, OOH
BOUGHT THE DIP, FUCK THE WHALES THIS IS NOT A GAME, YEAH
CAN’T BLAST THRU SELL WALLS WITH THESE BAGS MANG, AYE
BRAINY GANG, BRAINY GANG, BRAINY GANG

So should I buy this shit or not?

But it. Don’t expect moon gainz in a day it will take a few weeks of holding to actually pop.

BRAIN GANG

WHAT’S THE STATUS? DID I JUST FLUSH MY LIFE SAVINGS DOWN THE TOILET?

...

Sauce? I'm sitting comfy on a good amount, but from .25 -> $5 is kinda crazy.

Absolutely insane. Did you hear Pewdiepie was endorsing BAT? Shits going to explode next week when normies get in.

nope, but quite interesting.