Pages: [1]
Print
Author Topic: [Rumors] Kepler GPUs and PhysX  (Read 25713 times)
Broski
Newbie
*
Posts: 31


View Profile
« on: January 29, 2012, 01:52:32 pm »

Hai, I just found some interesting news/rumors:

Quote
The first news (if true) is breakthrough new feature for PhysX and CUDA through its processing. If this thing really confirmed, it will be a big bomb.

http://www.obr-hardware.com/2012/01/some-new-info-about-kepler-and-chiphell.html

What could it be? Smiley
Logged
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #1 on: January 29, 2012, 02:17:49 pm »

Well, even if author of this blog can be trusted, I don't know what to guess.
Hardly it is going to be new SDK feature (not possible on previous gen-GPUs), probably something related to performance.
Logged

PhysXInfo.com Team
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #2 on: February 01, 2012, 12:20:40 pm »

So now there is this
http://news.techeye.net/chips/nvidias-kepler-suffers-wobbly-perturbations
Quote
Some industry watchers suggest that Nvidia gave up a lot of space on its chip, trying to buff up Kepler by bringing Ageia to the hardware.
Post-Ageia hardware, while NVIDIA has shut down the Ageia hardware engineering group soon after acquisition ?
Special blocks for "fluid dynamics", while this is exactly what runs great on current GPU architecture ?

Errr.. sounds unlikely.
Logged

PhysXInfo.com Team
Spets
Newbie
*
Posts: 43


View Profile
« Reply #3 on: February 01, 2012, 01:47:03 pm »

So now there is this
http://news.techeye.net/chips/nvidias-kepler-suffers-wobbly-perturbations
Quote
Some industry watchers suggest that Nvidia gave up a lot of space on its chip, trying to buff up Kepler by bringing Ageia to the hardware.
Post-Ageia hardware, while NVIDIA has shut down the Ageia hardware engineering group soon after acquisition ?
Special blocks for "fluid dynamics", while this is exactly what runs great on current GPU architecture ?

Errr.. sounds unlikely.

That's the first thing that came to mind, and with physx 3.x being a complete overhaul.. I just don't see it happening. The whole article is probably false.

Also since this is my first post, would like to tell you I love your site  Cheesy I always check it for more news and developer articles related to physx.
Logged
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #4 on: February 01, 2012, 01:58:31 pm »

Quote
The whole article is is probably false
Seems so. But rumors usually don't occur out of nowhere, right ? Maybe Kepler will still have "special something" for GPU PhysX, either increased  effectiveness of streaming multiprocessors or optimizations in SDK 3.x

Quote
Also since this is my first post, would like to tell you I love your site
Thanks  Grin

P.S.
Some good posts at overclockers.co.uk
Logged

PhysXInfo.com Team
Spets
Newbie
*
Posts: 43


View Profile
« Reply #5 on: February 01, 2012, 06:00:25 pm »

P.S.
Some good posts at overclockers.co.uk
Lol, was an informative read, the link back to his previous post was funny too. Just goes to show it doesn't really make sense, although the Maxwell architecture is meant to introduce project Denver within it? Might see a "special something" there Tongue I'm more interested with Maxwell than Kepler to be honest  Cheesy
Logged
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #6 on: February 01, 2012, 06:59:08 pm »


Quote
I'm more interested with Maxwell than Kepler to be honest
Denver seems to be more oriented on GPGPU, not gaming (like Tesla).

Speaking about GPU PhysX as "fixed function" - there is actually no such thing, but, for example, devices with 2.x compute capability will run it more efficient in comparison to 1.x - according to docs.

P.S. Spoke with some insider people - technical data from that TechEye article is not correct Smiley
Logged

PhysXInfo.com Team
Spets
Newbie
*
Posts: 43


View Profile
« Reply #7 on: February 01, 2012, 07:32:11 pm »


Denver seems to be more oriented on GPGPU, not gaming (like Tesla).
Was thinking that, but it'd still be nice Tongue

Quote
Speaking about GPU PhysX as "fixed function" - there is actually no such thing, but, for example, devices with 2.x compute capability will run it more efficient in comparison to 1.x - according to docs.
Didn't think it would be, or even needed to be.

Quote
P.S. Spoke with some insider people - technical data from that TechEye article is not correct
Haha, thought so. Sounded like the whole article was fabricated from the start. The sooner the cards come out though, the sooner we can put these "rumours" to rest Tongue
Logged
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #8 on: February 01, 2012, 08:23:28 pm »

Quote
The sooner the cards come out though, the sooner we can put these "rumours" to rest
Indeed.
Also, release of the new GPU architecture is a good time to reveal some new GPU PhysX games and present cool techno-demos (like Supersonic Sled or Raging Rapids Ride for GTX 480) - at least, I'm hoping for that Smiley

And some more from Demerjian (hot topic these days, really.. even becoming annoying  Undecided)
Some are calling this PhysX block a dedicated chunk of hardware, but more sources have been saying that it is simply shaders, optimisations, and likely a dedicated few new ops. In short, marketing may say it is, but under the heat spreader, it is simply shaders and optimisations.
Logged

PhysXInfo.com Team
Spets
Newbie
*
Posts: 43


View Profile
« Reply #9 on: February 02, 2012, 02:50:10 am »

And some more from Demerjian (hot topic these days, really.. even becoming annoying  Undecided)
Some are calling this PhysX block a dedicated chunk of hardware, but more sources have been saying that it is simply shaders, optimisations, and likely a dedicated few new ops. In short, marketing may say it is, but under the heat spreader, it is simply shaders and optimisations.

Well that makes more sense now, but I still don't completely understand why Nv would have a shader block dedicated for physics simulations when they can add more shaders for both gaming and physics. Guess we'll have to wait for it to be released. Reading through the article again, to see him mention it will be a $299 mid-range card, the performance seems to be quite good. Hoping that pricetag will stick too Tongue

Quote
Also, release of the new GPU architecture is a good time to reveal some new GPU PhysX games and present cool techno-demos (like Supersonic Sled or Raging Rapids Ride for GTX 480) - at least, I'm hoping for that
I'm looking forward to the tech demo's too Tongue I always have a bit of fun with those and the Cuda SDK. It's what also brought me to this site a while ago Tongue
Logged
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #10 on: February 02, 2012, 12:00:31 pm »

Quote
Well that makes more sense now
From other hand, mr. Demerjian is saying that Kepler will be able to accelerate software PhysX in regular games (like the PPU, just much more effective). So basically GK104 is slower than Tahiti, but performs faster in selected games, "most [of which] use the software API" because "Nvidia is accelerating their own software APIs on Kepler without pointing it out explicitly".
For me, it is looking as unlikely as special PhysX chip on new GPUs.

Quote
I'm looking forward to the tech demo's too
I'm hoping for a new version of Lighthouse - but not just video, downloadable demo Smiley
Logged

PhysXInfo.com Team
Spets
Newbie
*
Posts: 43


View Profile
« Reply #11 on: February 02, 2012, 01:31:54 pm »

I'm hoping for a new version of Lighthouse - but not just video, downloadable demo Smiley
Haha, I tried looking for that when I first saw the video, it would be nice to run. Hoping for something that'll be like a physx sandbox Tongue Be able to play with a mass of particles from rigid bodies to fluid simulations.

Quote
From other hand, mr. Demerjian is saying that Kepler will be able to accelerate software PhysX in regular games (like the PPU, just much more effective). So basically GK104 is slower than Tahiti, but performs faster in selected games, "most [of which] use the software API" because "Nvidia is accelerating their own software APIs on Kepler without pointing it out explicitly".
For me, it is looking as unlikely as special PhysX chip on new GPUs.
I still agree, to have a dedicated shader cluster and sacrificing other portions just seems too unlikely, especially when the architecture is made for GPGPU computing.
Logged
Zogrim
Administrator
Hero Member
*****
Posts: 564


Expert Scrounger


View Profile WWW
« Reply #12 on: February 02, 2012, 02:42:48 pm »

Quote
Haha, I tried looking for that when I first saw the video, it would be nice to run
I think "fluid dynamics" was mentioned in previous articles for a reason.
Knowing that hybrid heightfield fluid solver will be a PhysX feature.. in the future (Wink), there is high possibility that one of the Kepler techno-demos will actually showcase it. Maybe new version of Raging Rapids Ride demo ? Would be nice.

Also, it can be demo with cloth/character clothing (new one from SDK 3) or something massive with GPU Rigid Bodies.
Logged

PhysXInfo.com Team
Spets
Newbie
*
Posts: 43


View Profile
« Reply #13 on: February 03, 2012, 03:43:06 pm »

Hope so Tongue Can't wait to see the new tech demo's!
Logged
Pages: [1]
Print
Jump to: