The GT300/Fermi Thread

[Sam_oslo]

Good thread! I hope all kids in this thread can behave.

I was about to order a 5790 a couple of times, because it is a really good GPU. But the supply was low and the price was high, so I decided to wait until I know more about Fermi.

[Tim]

[FUD] nVidia to showcase Fermi solutions later today

Source : Fudzilla

[LedHed]

[FUD] nVidia to showcase Fermi solutions later today

Source : Fudzilla

nice, I'm in the webinar now. Govt IT passes (my job)

[LedHed]

hell yea I got my question answered by Sumit Gupta, whoever that is, haha.

[dengyong]

hell yea I got my question answered by Sumit Gupta, whoever that is, haha.

What question and what answer ?

[LedHed]

What question and what answer ?

there is a live webinar going right now: https://www2.gotomeeting.com/register/606754650

try to get in, it's still going. All I had to do was say I work for Govt in IT.

[LedHed]

One customer asks about ATI vs NVIDIA. Answer by Sumit Gupta: "I really encourage you to try an ATI product, the programming is plain horrible. They don't even support Double Precision in their OpenCL compiler." Of course this is regarding professionals, not gamers.

[N19h7m4r3]

One customer asks about ATI vs NVIDIA. Answer by Sumit Gupta: "I really encourage you to try an ATI product, the programming is plain horrible. They don't even support Double Precision in their OpenCL compiler."

What?! As far as I know the Firestream 9170 supports it and was one of the first industry cards to do that.

Infact all the 4 series and 5 series cards do, with alot of FirePro and even Mobility cards.

[dengyong]

there is a live webinar going right now: https://www2.gotomeeting.com/register/606754650

try to get in, it's still going. All I had to do was say I work for Govt in IT.

What question did you ask and what was the answer ?

[Russian]

Sumit is one of the guys that is very high up in Nvidia, he isnt a nobody.

[LedHed]

What question did you ask and what was the answer ?

I asked if Fermi was developed around the GPGPU application in Windows 7, or if Fermi was already developed around the GPGPU mindset before GPGPU apps became known to the public.

The answer was that Fermi was designed very closely with Microsoft and most definitely Windows 7 shaped some of the design. However Fermi was in development as a GPGPU while Vista was the primary OS.


I also asked if the new "Infiniband" will reduce the memory bottleneck at enthusiast resolutions and the answer was Inifinband is a very expensive feature and most likely won't make it to the Geforce line. (Ifiniband bypasses one read/write cycle to the main system RAM reducing latency by 20%)

[N19h7m4r3]

Any chance of finding out the power requirement and physical size of the cards?

[LedHed]

They won't talk about Geforce, only Tesla. I asked anyway, we'll see if they answer.

[annihilat0r]

520-630 Gigaflops DP performance

[LedHed]

Here is one slide I found interesting



And this is an actual die shot of Fermi:

[LedHed]

aw it's over, next one in January

[Nedjo]

And this is an actual die shot of Fermi:

any word on size?

[LedHed]

any word on size?

nope, not unless I missed it in the first few slides, I came in the webinar 10 mins late.

[SimBy]

Nothing new, a bunch of slides, no talk about graphics whatsoever.

[ajaidev]

This sucks all i kept on getting was:



${Webinar.Subject}
Come Back Later
This Webinar is not yet open to attendees. Please come back at the scheduled time below.
${Webinar.Subject}
${webinarDate} ${webinarStartTime} - ${webinarEndTime}

[LedHed]

This sucks all i kept on getting was:



${Webinar.Subject}
Come Back Later
This Webinar is not yet open to attendees. Please come back at the scheduled time below.
${Webinar.Subject}
${webinarDate} ${webinarStartTime} - ${webinarEndTime}

Yea I was surprised I got in so easily, must just be based on what Dept/Job you're associated with

[Nedjo]

did we learned anything new about Fermi from this webinar??

[LedHed]

did we learned anything new about Fermi from this webinar??

depends on what you already know, there is noway I'm going to try and summarize that webinar.

The only new thing I heard about was Infiniband.

[003]

Hmm so what is Infiniband and how does it work exactly? It's nice maybe now their Quadro and Tesla cards won't simply be exactly the same as GeForce but with more memory.

[Solus Corvus]

I have seen infiniband used as an interconnect between physically separated machines in a high performance cluster. When used in that capacity it's more like an ultra-fast ethernet. I had no idea it could be scaled down to the board level but I suppose there is no reason that it couldn't.