AMD's HD5970 faster than Nvidia Fermi chip!

[H82LUZ73]

Man!! forgot about this show , what was it call again?
i just can't remember...?

EDIT:Found it on Google it's family matters but anyhow thanks! for make me remember that funny show
BTW:Still would not believe anything from Fudz & the Inquirer ...

Was actually A spin from another show Perfect Strangers were the mom in Family Matters was the Elevator operator.

Ha found it

 

[Super XP]

AMD already sold a record 2 million HD 5000's which is not bad considering the delays in the 40nm process. From what I've read they are begining to ramp up full force less than 2 weeks time. Nvidia has a lot of catching up to do, but that's good competition.

 

[imperialreign]

Biggest drawback I see, ATM, with nVidia taking so long to get to market - they've given up a lot of "new series" sales to the red camp. At the rate these delays are going, ATI's next series will be right behind nVidia's release . . .

Regarding the article at Fud (or any other site) . . . like all tech "rumors," I take them all with a full shaker of salt. Sometimes they're right, sometimes not . . . we won't know for sure until the hardware is starting to circulate.

But . . . y'all know how much we love speculation here at TPU

 

[Super XP]

Fermi is taking long because its going to be more than just your average graphics card If NVIDIA sticks to there original Fermi plan they are going to have a monster despite what internal sources say about it being only a mild 20% boost in performance vs. there best today. But AMD already has something coming and it's called HD 5900's

 

[imperialreign]

Fermi is taking long because its going to be more than just your average graphics card If NVIDIA sticks to there original Fermi plan they are going to have a monster despite what internal sources say about it being only a mild 20% boost in performance vs. there best today. But AMD already has something coming and it's called HD 5900's

It wouldn't surprise me - TBH, I've been under the impression for a while that the delays are either due to an inability to control thermal output, or that the performance hasn't been up to their expectations. Supposedly, Fermi is completely new architecture for nVidia, and I get the defi feeling that it's the R&D that are slowing things up.

Either way, it's all speculation at this point, as nVidia themselves haven't coughed up too much info . . . only some occasional hype.

It's taking so long, though, that I'm sure ATI have something planned for whenever green gets to the shelf.

 

[El_Mayo]

nVidia needs to hurry up and release whatever they're building, so ATi cards become cheaper.
Also, what exactly is a yield?

 

[HalfAHertz]

They have been spreading themselves too thin with all that tegra and ion business, haven't they? Still the way I see thngs: Nvidia is late with this gen, but they will have a new architecture, where as Ati has reused the 3gen old architecture. Ati will be in the same position with their next release. They need a new architecture, but at least they have a head start with the Dx11 release and hopefully will not experience any delays.

 

[yogurt_21]

They have been spreading themselves too thin with all that tegra and ion business, haven't they? Still the way I see thngs: Nvidia is late with this gen, but they will have a new architecture, where as Ati has reused the 3gen old architecture. Ati will be in the same position with their next release. They need a new architecture, but at least they have a head start with the Dx11 release and hopefully will not experience any delays.

seen this alot the past few days but nothing posted to back it up. all the specs I've seen on fermi seem to indicate a beefed up gt200 with dx11 support. any sources to back up that nvidia has left the unified shader architecture?

I find it very funny when people on this forum use "designed from the ground up" without knowing anything about what that means.

if it had been designed form the ground up it would use a completely different shader design, not just a different number. in all truthfulness neither team has had a new internal architecture redesign from the ground up since the unified shader architecture was released. And I seriously doubt nvidias fermi is either based on all the reports.

so lets go out to the net and see.

http://www.brightsideofnews.com/new...ure-unveiled-512-cores2c-up-to-6gb-gddr5.aspx

unified shader architecture

http://www.nvidia.com/object/fermi_architecture.html

unified shader architecture

http://www.bit-tech.net/news/hardware/2009/09/30/huang-reveals-fermi-architecture/1

unified shader architecture


I'm sorry am i missing something? fermi like most new series has several feature bumps over the prior generation, however it doesn't seem anywhere near "designed from the ground up"
you're thinking the 8800 series there.

 

[pantherx12]

nVidia needs to hurry up and release whatever they're building, so ATi cards become cheaper.
Also, what exactly is a yield?

I shall teach you something that will help you for ever.

Google has several built in tools, one being the "define" tool

Put

Define :

Into google followed by the word you don't know and it lets you know : ]

Much faster then asking.

 

[imperialreign]

if it had been designed form the ground up it would use a completely different shader design, not just a different number. in all truthfulness neither team has had a new internal architecture redesign from the ground up since the unified shader architecture was released.

Agreed. Regarding the red camp's design, the last new architecture was the R600 . . . and, as you had mentioned, that was the start of the red camp's unified shader architecture. All newer GPU cores from ATI since have been revisions of that design.

 

[crazyeyesreaper]

true but each revision seems to dbl its performance since then at least when comparing the xx70 variants

3870-4870-5870 so while it may be the same arch its defiently proven to be very scalable in terms of performance but eventually that will reach its end because nothing can scale that way indefinetly so we will have to see what the 6k series brings

 

[Steevo]

6K is a whole new design. Multi-core design according to the rumour mills.

 

[trt740]

the boys at geforce are building a monster it appears.

 

[phanbuey]

well they built a monster... on paper... but the people who actually make the chip can't seem to build it.

 

[eidairaman1]

Rest Assured TMSC will have trouble building the part just as much as they did with ATIs stuff, thats why i think Radeon HD6 will be built by GloFo

 

[HalfAHertz]

seen this alot the past few days but nothing posted to back it up. all the specs I've seen on fermi seem to indicate a beefed up gt200 with dx11 support. any sources to back up that nvidia has left the unified shader architecture?

I find it very funny when people on this forum use "designed from the ground up" without knowing anything about what that means.

if it had been designed form the ground up it would use a completely different shader design, not just a different number. in all truthfulness neither team has had a new internal architecture redesign from the ground up since the unified shader architecture was released. And I seriously doubt nvidias fermi is either based on all the reports.

so lets go out to the net and see.

http://www.brightsideofnews.com/new...ure-unveiled-512-cores2c-up-to-6gb-gddr5.aspx

unified shader architecture

http://www.nvidia.com/object/fermi_architecture.html

unified shader architecture

http://www.bit-tech.net/news/hardware/2009/09/30/huang-reveals-fermi-architecture/1

unified shader architecture


I'm sorry am i missing something? fermi like most new series has several feature bumps over the prior generation, however it doesn't seem anywhere near "designed from the ground up"
you're thinking the 8800 series there.

True, but you have to remember that the shaders are just ~50% of the GPU. There's ALOT more happening behind the scene than just rasterization.

 

[bobzilla2009]

Rest Assured TMSC will have trouble building the part just as much as they did with ATIs stuff, thats why i think Radeon HD6 will be built by GloFo

well, AMD bought gloFo last march, so that's a huge factor too. It should mean a lot more cards streaming into the market next time round though compared to how slowly the hd5xxx's have been coming out

 

[btarunr]

http://www.youtube.com/watch?v=dlgzfwIrmIc

 

[eidairaman1]

NV launch reminds me of this

http://www.youtube.com/watch?v=WOVjZqC1AE4

 

[Benetanegia]

seen this alot the past few days but nothing posted to back it up. all the specs I've seen on fermi seem to indicate a beefed up gt200 with dx11 support. any sources to back up that nvidia has left the unified shader architecture?

I find it very funny when people on this forum use "designed from the ground up" without knowing anything about what that means.

if it had been designed form the ground up it would use a completely different shader design, not just a different number. in all truthfulness neither team has had a new internal architecture redesign from the ground up since the unified shader architecture was released. And I seriously doubt nvidias fermi is either based on all the reports.

so lets go out to the net and see.

http://www.brightsideofnews.com/new...ure-unveiled-512-cores2c-up-to-6gb-gddr5.aspx

unified shader architecture

http://www.nvidia.com/object/fermi_architecture.html

unified shader architecture

http://www.bit-tech.net/news/hardware/2009/09/30/huang-reveals-fermi-architecture/1

unified shader architecture


I'm sorry am i missing something? fermi like most new series has several feature bumps over the prior generation, however it doesn't seem anywhere near "designed from the ground up"
you're thinking the 8800 series there.

Well, if you take a look at the TPU front page and follow the correct article, you'll learn how wrong you are. I too thought that when it came to graphics, it was going to be like GT200, just a 2x GT200 + DX11 and "software" (ISA level, actually) tesselation. Boy, I was wrong! We were all so wrong, especially with the tesselation part. "It will not have a dedicated tesselator". How many times we have heard that? Now the only thing I can think about the origin of that sentence is that Nvidia said "I will not have ONE dedicated tesselator..." and whispering so low no one could hear "ahem, it has 16! Along with 4 triangle setup and raster engines too!".

Also the texture units run at the shader clocks, everything except the ROPs and memory runs at shader clock in fact, so yeah, it's really different. Everything except ROPs has been moved to what now they call GPC, which is a GPU core for all intents and purposes. Fermi is a multi-core GPU!!

 

[wolf]

Even the 448sp version will be a 5870 killer, and it looks like things will go back to how they have been time and time again;

To own THE BEST graphics subsystem, means you will use Nvidia cards. That is Nvidia's bread and butter imo, and they will have many buyers (eg the likes of TPU members) just from that fact alone.

 

[Benetanegia]

I don't think there's going to be a 448 SP version. I dont see how exactly they could disable only 64 SPs on that architecture. Unless they can disable individual SPs instead of requiring the old method of disabling clusters? Now that I think about it, maybe it's that what their vastly improved scalability claims have always been about?

 

[wolf]

Potentially so, I figure since the sp's are in clusters of 32, they would be able to laser cut up to two defective ones?

 

[Benetanegia]

Potentially so, I figure since the sp's are in clusters of 32, they would be able to laser cut up to two defective ones?

I don't know... that would leave 2 GPCs with 4 clusters and 2 with 3 clusters or 4-4-4-2. Feasible, maybe but...

 

[Binge]

The GT200 could disable cores entirely via software.