Southern Islands in November?

[Hauk]

Looking forward to seeing a Southern Islands flagship. Can't wait to see temp, power consumption, and performance (esp DX11) vs. GTX 480. If general performance is even just 5% over 480 while maintaining decent power consumption, it'll be painful I fear, especially if this is nVidia's planned counter..

http://www.fudzilla.com/graphics/graphics/graphics/new-nvidia-gf100-pcb-shows-up

http://www.techpowerup.com/128276/GF100_512_Core_Graphics_Card_Tested_Against_GeForce_GTX_480.html

 

[evolucion8]

Enabling full processors with Fermi will result only in a minimal 5 to 10 percent increase in performance in the best case scenario which is related to shader performance. Because the current GTX 480 has its fully functional texture units and ROP's. And by adding 32 ALU's is not gonna help, they need to respin the chip or launch a tweaked GF104.

 

[HurleyBird]

If they keep running six months behind it will be the end of them most likely. I'm sure they must be endeavoring towards this goal, with the way this cycle has gone for them.

For sure, the big question is how can Nvidia catch up to ATI. What we're seeing here might be another R600 momement, where Nvidia did extra work (but did not execute very well) and simply needs to fix the flaws in its product to let the architecture shine. To be fair, Fermi does have a certain R600 vibe. GF104 has a RV670 vibe. The most challenging thing will be coming out with the Fermi version of RV770, but God help AMD if they do and NI is their GT200...

On the other hand, it's possible that ATI is simply outengineering Nvidia, in the same way that Nvidia outengineered and killed 3DFX. The fact that they got the entire Evergreen family out in such quick sucession definitely makes it seem like they've kicked it up a notch. They seem to be help back more by manufacturnig than anything else, and the fact that they had to redesign both RV870 and Northern Islands (into southern islands) thanks to TMSC, and are *still* outpacing Nvidia is simply frightening.

Truth is, it's probably a bit of both. For GF100, it seems like a lot of the groundwork for a great product has been done. On the other hand, even without thinking about Nvidia the ATI design team seems to be on a huge hot streak. Fusion is going to rob Nvidia of most of their low end, and could also put a significant dent in their CUDA business if AMD plays their cards right. Things look bad for Nvidia, and I think it's all going to come down to their execution over the next few years. This is the defining moment where they either create and cement their own niche (CUDA) where they can survive indefinitely, or they fade away into irrelevance.

 

[Hauk]

Enabling full processors with Fermi will result only in a minimal 5 to 10 percent increase in performance in the best case scenario which is related to shader performance. Because the current GTX 480 has its fully functional texture units and ROP's. And by adding 32 ALU's is not gonna help, they need to respin the chip or launch a tweaked GF104.

Agreed. GTX 485ish won't be the answer. Vaporware or a limited supply Ultra card?

 

[Idontcare]

On the other hand, it's possible that ATI is simply outengineering Nvidia, in the same way that Nvidia outengineered and killed 3DFX. The fact that they got the entire Evergreen family out in such quick sucession definitely makes it seem like they've kicked it up a notch.

If the rumors are true and the integrated GPU on both Llano and Ontario are based on Evergreen architecture then it would be pretty much required (meaning management would have had no choice but to resource the GPU teams as needed to make it so) that the Evergreen architecture development was kicked into high gear so that the fusion product timelines did not suffer a delay due to the GPU-side of the equation.

 

[Sokar]

Realisticaly AMD is betting on their graphics as their main selling point for future cpus in the low to mid range. Look at all their new AMD vision marketing. IMO it does have its merits where at some point more cpu power is useless if you don't have enough video power (at least when you consider the multimedia boxes that are what 90% of the market want) and they are branding themselves that way. Really the GPU design is worth more to AMD than just the GPU business so they are naturally going to pour more resources into it.

Also building conservative sized chips probably saved them lots of time and effort in both design and implementation. They probably have more room for error when their chip is only like 70% the size of what nvidia is making(last 2 generations).

 

[tviceman]

Enabling full processors with Fermi will result only in a minimal 5 to 10 percent increase in performance in the best case scenario which is related to shader performance. Because the current GTX 480 has its fully functional texture units and ROP's. And by adding 32 ALU's is not gonna help, they need to respin the chip or launch a tweaked GF104.

Agreed and hopefully that is what the faintly rumored GF102 will be. HOPEFULLY GF102 will be to GF100 what NV35 (fx5900 ultra) was the NV30 (fx5800 ultra). Across the board, they need to both decrease operating temperatures, lower the power draw, enable all the shaders, and perhaps slightly increase the clockspeed to squeeze out a little bit of extra performance.

Sounds like a big task.

Looking forward to seeing a Southern Islands flagship. Can't wait to see temp, power consumption, and performance (esp DX11) vs. GTX 480. If general performance is even just 5% over 480 while maintaining decent power consumption, it'll be painful I fear, especially if this is nVidia's planned counter..

http://www.fudzilla.com/graphics/graphics/graphics/new-nvidia-gf100-pcb-shows-up

http://www.techpowerup.com/128276/GF100_512_Core_Graphics_Card_Tested_Against_GeForce_GTX_480.html

There is no way that card is using a current GF100 chip. If it's real, it's more than likely A0 stepping and just goes to show how much of an uphill battle Nvidia had getting GF100 into a (somewhat) viable product.