Oak Ridge National Laboratory gets Kepler K20 "Big K"

[boxleitnerb]

I wonder if GK110 will come to desktop or not.

It will. Wanna bet?

 

[BenSkywalker]

Seems 'Titan' is done, 18,688 K20s-

http://www.engadget.com/2012/10/29/cray-titan-supercomputer-nvidia-tesla-gpu-k20/

Looks like it should be the fastest in the world for at least a little while.

 

[Silverforce11]

Seems 'Titan' is done, 18,688 K20s-

http://www.engadget.com/2012/10/29/cray-titan-supercomputer-nvidia-tesla-gpu-k20/

Looks like it should be the fastest in the world for at least a little while.

"has been loaded up with the first shipping NVIDIA Telsa K20 GPUs and renamed Titan".. so NV shipped them 1,000 K20s..

Its done when they announce the first run and throughput tests, and go on the world record for #1.

NV completely dominates the HPC sector and this isnt going to change anytime soon.

 

[BenSkywalker]

"has been loaded up with the first shipping NVIDIA Telsa K20 GPUs and renamed Titan".. so NV shipped them 1,000 K20s..

According to all three(the article and both press releases) it is 18.7K, not 1K. They stated it was 20PFLOPS, which should have it number one. We won't get confirmation until November, Top500 only updates once every six months(I know they wanted to get it done for this round).

 

[3DVagabond]

:snip:

The hardware behind Titan isn't meant to power your gaming sessions, but the NVIDIA says lessons learned from supercomputer GPU development trickle back down to consumer-grade cards.

So, does this mean that we will or won't get a Geforce GK110? Looks like no from the opening line of the sentence.

 

[BenSkywalker]

So, does this mean that we will or won't get a Geforce GK110? Looks like no from the opening line of the sentence.

That is a quote from the Engadget author, I highly doubt they really understand the nuances of GeForce/Quadro/Tesla.

 

[Keysplayr]

"has been loaded up with the first shipping NVIDIA Telsa K20 GPUs and renamed Titan".. so NV shipped them 1,000 K20s..

Its done when they announce the first run and throughput tests, and go on the world record for #1.

NV completely dominates the HPC sector and this isnt going to change anytime soon.

"has been loaded up with the first shipping NVIDIA Telsa K20 GPUs and renamed Titan".. so NV shipped them 1,000 K20s.."
^Old quote from a Sept 2 article. http://blogs.knoxnews.com/munger/2012/09/the-big-computing-change-is-ta.html
It's done now. Not when you say it is. All 18,688 K20's have been installed and Titan is complete.

 

[tviceman]

"has been loaded up with the first shipping NVIDIA Telsa K20 GPUs and renamed Titan".. so NV shipped them 1,000 K20s.

There you go again spitting worst case or straight up false info to further your own subliminal agendas. Go ahead and feign innocence again.

 

[boxleitnerb]

A good 6 months after tape out and they have shipped almost 20,000 cards, possibly more to other sources (without press releases). Can we infer something about the yields from this?

 

[railven]

A good 6 months after tape out and they have shipped almost 20,000 cards, possibly more to other sources (without press releases). Can we infer something about the yields from this?

Possible harvested dies will go into the next GeForce SKU?

I wonder if this machine will be able to calculate Hurricane Sandy to the second, cuz right now - I don't know if I'll survive the next two days haha.

 

[boxleitnerb]

Possibly, yes. I doubt there will be a 15 SMX Tesla or Quadro, 14 SMX at most.

 

[tviceman]

Possibly, yes. I doubt there will be a 15 SMX Tesla or Quadro, 14 SMX at most.

I remember everyone saying full spec'd big fermi's would be impossible on 40nm....

 

[Lonbjerg]

I remember everyone saying full spec'd big fermi's would be impossible on 40nm....

Wasn't everyone....wasn't it more^like Charlie and the red camp?

You know...buisness like usual ^^

 

[boxleitnerb]

I remember everyone saying full spec'd big fermi's would be impossible on 40nm....

Quadro != Tesla != Geforce. Geforce has more TDP headroom and less reliability issues (24/7 operation and such).

 

[ShintaiDK]

Quadro != Tesla != Geforce. Geforce has more TDP headroom and less reliability issues (24/7 operation and such).

Geforce doesnt have more TDP headroom. Telsa cards are usually clocked lower or got higher TDP for the same parts. Why? Because no Radeon/Geforce card can run 100% true loads without throttle. Thats a tradeoff you make for higher regular gaming performance.

And you can just bin it as you wish. So if you can make 40nm Geforce you can make 40nm Tesla and Quadro.

 

[railven]

I remember everyone saying full spec'd big fermi's would be impossible on 40nm....

Wasn't everyone....wasn't it more^like Charlie and the red camp?

You know...buisness like usual ^^

Odd...

<checks his badge>

You guys would firmly plant me in the red camp, and I recall never making such comments nor caring enough to even look up yields for any product to date.

You guys are crazy! Everyone? Really?

 

[boxleitnerb]

Geforce doesnt have more TDP headroom. Telsa cards are usually clocked lower or got higher TDP for the same parts. Why? Because no Radeon/Geforce card can run 100% true loads without throttle. Thats a tradeoff you make for higher regular gaming performance.

And you can just bin it as you wish. So if you can make 40nm Geforce you can make 40nm Tesla and Quadro.

Of course they do. Ever seen a Tesla at 250W TDP or even higher? Not every server/workstation has the appropriate power connectors. Power efficiency is important in the HPC segment. Cooling is important too due to electricity costs. And finally, the additional 3GB memory use up power as well.

My point being, 13 and 14 SMX for Tesla and Quadro, but up to 15 SMX for Geforce with higher TDP. Business as usual.

 

[tviceman]

Odd...

<checks his badge>

You guys would firmly plant me in the red camp, and I recall never making such comments nor caring enough to even look up yields for any product to date.

You guys are crazy! Everyone? Really?

Sorry that was too general of a statement, but there were quite a few people saying a big Fermi fully enabled would be either impossible or would have too high of a power draw to be reasonably manufactured on 40nm.

 

[railven]

Sorry that was too general of a statement, but there were quite a few people saying a big Fermi fully enabled would be either impossible or would have too high of a power draw to be reasonably manufactured on 40nm.

No, I know, I remember it too. Haha, I'm just being me and dismissing myself.

Charlie sure did lay it on thick though. Before I knew that guy was a hack, I thought he new legit info then GTX 480 rolled out and later GTX 580 without issues (outside 480 being hotter than I liked, but meh that's just preference.)

 

[tviceman]

Quadro != Tesla != Geforce. Geforce has more TDP headroom and less reliability issues (24/7 operation and such).

That, and some of the transistors dedicated to hpc specific functions may be disabled as well. I don't think the initial top end gk110 GeForce will be fully enabled, but I do believe we will see it by or before the end of summer.

 

[Lonbjerg]

Odd...

<checks his badge>

You guys would firmly plant me in the red camp, and I recall never making such comments nor caring enough to even look up yields for any product to date.

You guys are crazy! Everyone? Really?

Did i mention you by name?
No, so get your fallacy *beep* away from me...

 

[ShintaiDK]

Of course they do. Ever seen a Tesla at 250W TDP or even higher? Not every server/workstation has the appropriate power connectors. Power efficiency is important in the HPC segment. Cooling is important too due to electricity costs. And finally, the additional 3GB memory use up power as well.

My point being, 13 and 14 SMX for Tesla and Quadro, but up to 15 SMX for Geforce with higher TDP. Business as usual.

Thats rubbish. The issue is simply that a well coded GPGPU load adds a higher load than gaming and doesnt need halt stages in drivers and so on. Just look back at HD69xx cards that couldnt run some games without heavy throttle. Or the famous furmark issue.

 

[railven]

Did i mention you by name?
No, so get your fallacy *beep* away from me...

Temper temper. What fallacy, I quoted you, and I'll do so again:

Wasn't everyone....wasn't it more^like Charlie and the red camp?

You know...buisness like usual ^^

Red camp includes whom? It's your fallacy, if anyone's

 

[ShintaiDK]

Anyway with GK110 real today, poor Charlie.

 

[SlowSpyder]

http://i.top500.org/system/176544

They did, Charlie was wrong(what a shocker, heh) and it is currently ranked sixth in the world. This machine they are building with BigK parts is actually an upgrade to their Fermi based machine.

What was that about Chuck, SlowSpyder?
Don't ever give that man credit for anything.
A sensationalist "journalist" that writes for a site designated as a "satire" site. Zero cred.

I completely missed these replies somehow. That is pretty interesting, then. I remember Charlie making a big deal out of it, then everything news-related about that computer seemed to die down, I figured it wasn't built. Looks like Charlie should run a new story saying how wrong he was.