Nvidia announces new Tesla Accelerators

[AlTech]

 

Nvidia has announced the New Tesla M4 and Tesla M40 based on the maxwell architecture.

 

 

 

 

source: http://anandtech.com/show/9776/nvidia-announces-tesla-m40-m4-server-cards-data-center-machine-learning

 

[EmeraldKiwi]

That card looks really cool. I'm not really into workstation cards though, so I don't really understand the implications of a card like this.

EDIT:

It looks to me like the M60 is a dual GPU model though, which sounds like its going to produce alot of heat.

[Imakuni]

Lok at dat FP64 number!

 

All Hail Maxwell!!! (not)

[yannickverc]

B-b-b-b- ... But can it run crisis?

 

 

 

^{sorry not sorry}

[AlTech]

That card looks really cool. I'm not really into workstation cards though, so I don't really understand the implications of a card like this.

The M40 and M4 are the first GPUs dedicated for Machine Learning basically.

[AlTech]

B-b-b-b- ... But can it run crisis?

 

 

 

^{sorry not sorry}

Triple SLI M40s, Hell yeah. Triple SLI M4s? Not so much

[EmeraldKiwi]

The M40 and M4 are the first GPUs dedicated for Machine Learning basically.

What exactly is machine learning?

[yannickverc]

Triple SLI M40s, Hell yeah. Triple SLI M4s? Not so much

Sad day in history once more

[PlayStation 2]

Fermi part 2 /jk

[zMeul]

..

those are not graphics cards at all, they are specialized compute units, aka Tesla Accelerators

to note: they don't have video outputs

[AlTech]

What exactly is machine learning?

A type of artificial intelligence that provides computers with the ability to learn without being explicitly programmed.

[Samfisher]

Your thread must include some original input to tell the reader why it is relevant to them, and what your personal opinion on the topic is.

 

Sad times when people just link stuff and go.

[KemoKa]

A 250W passive cooler?

I smell throttling...

[Diederik_Gr]

A 250W passive cooler?

I smell throttling...

These things are designed to sit inside server cases, where an extra fan isn't needed to provide airflow to these cards.

[AlTech]

A 250W passive cooler?

I smell throttling...

Really? COs I smell burning, throttling and dying

[KemoKa]

Really? COs I smell burning, throttling and dying

I mean, apparently they get all the airflow they need from their server fans, but... We know from the reference 290X that that isn't a thing. If it is, they'll have to be binned to run at 95C without throttling

[Bl00dgod]

Your thread must include some original input to tell the reader why it is relevant to them, and what your personal opinion on the topic is.

 

Sad times when people just link stuff and go.

bunshishi tawara is right

 

tesla are the "predecessors" of independend working pcs controlled by ai ..... winter*cough* skynet is coming

[KemoKa]

And, uh... exactly how much is this new flagship Tesla going to cost? Upwards of $5000 a piece? Because that compute performance is no longer that impressive. 

 

Let's see this card mine scrypt coins.

 

Ohhh, the GM206 chips! Will some of them have 3.5GB of RAM?

[AlTech]

And, uh... exactly how much is this new flagship Tesla going to cost? Upwards of $5000 a piece? Because that compute performance is no longer that impressive. 

 

Let's see this card mine scrypt coins.

 

Ohhh, the GM206 chips! Will some of them have 3.5GB of RAM?

I believe the 3.5GB are only on GM204-300. Not GM206. Only 128 Bit sadly.

[patrickjp93]

Lok at dat FP64 number!

 

All Hail Maxwell!!! (not)

And it's being marketed at non-FP64 workloads, so it's no problem. AI doesn't need FP64. 

[patrickjp93]

And, uh... exactly how much is this new flagship Tesla going to cost? Upwards of $5000 a piece? Because that compute performance is no longer that impressive. 

 

Let's see this card mine scrypt coins.

 

Ohhh, the GM206 chips! Will some of them have 3.5GB of RAM?

They're being marketed for things other than FP64, so frankly your kvetching is without a basis in reality. Nvidia said hat they're good for. If you buy them for something else, that's on you.

[Misanthrope]

I mean, apparently they get all the airflow they need from their server fans, but... We know from the reference 290X that that isn't a thing. If it is, they'll have to be binned to run at 95C without throttling

Server cards actually are binned to run not only at higher Temps but with constant work load.

Along with double precision and ecc memory it's the reason Nvidia charges 10x and more per card.

[patrickjp93]

Server cards actually are binned to run not only at higher Temps but with constant work load.

Along with double precision and ecc memory it's the reason Nvidia charges 10x and more per card.

That and up to 24GB of VRAM for a 384-bit bus.

[BuckGup]

A 250W passive cooler?

I smell throttling...

You do realize these are not your average card. They go into servers which have a large cooler system that goes through the racks. The CPUs also have a similar cooler on them. Fans on them would not do them justice. They have large air conditioners that blow through then the warm air is cooled then goes to the next row behind them

[Prysin]

And it's being marketed at non-FP64 workloads, so it's no problem. AI doesn't need FP64. 

sad fact though... even the K40 is like 0.7TFLOPS slower in FP64 then the FirePro W9100....

 

 

oh, for those who think FuryX is beast in FP64, it is not. Fiji was not designed like Hawaii, so they do not have 1/2 FP32 performance, instead they have 1/4 FP32...

 

 

i hope Pascal will give us proper compute on Nvidias side... cuz maxwell isnt that good