it matters to Nvidia; they are more than doubling performance in their Tesla part - but i don't think it will translate to gaming

. . . and i guess we are agreeing.

We will see how it all turns out.

I'm fairly certain nvidia will have overhauled their ROP's for GT300 (supporting DX10.1 x DX11 pretty much demands this happen) and I doubt they have made them less efficient. This will be the second architecture with 8x MSAA, so performance there should improve.

nvidia is quietly confident they have ATi thoroughly beaten with IQ, they were very condescending about ATi's new AF, so I reckon the TMU's have had an overhaul also.

I believe those who think nvidia has abandoned 3D performance in favor of GPGPU are in for a rude awakening on both performance and IQ fronts when GTX380 lands.

Agreed. That is impossible. It is Nvidia's bread and butter and their pride is at stake.

Fermi will be a beast for gaming. But it's not only about graphics. It's about computing features
like physx or DirectCompute.
It will have L1 and L2 cache, 10x faster context switching (threads), multiple kernel execution - this last one is probably most revolutionary for a GPU... imagine you have 512 GPU cores, of which as many is needed for the certain moment of execution is used as dedicated physx, other used for pure graphics. Features like physx are only starting to evolve, a lot of computing power is needed for things like this. Fermi will be game changer for this.

In a graphics situation the L2 cache could see use as a super fast texture cache for extremely high quality anisotropic filtering, or it could become a high speed window onto the frame buffer.

I wonder if we will ever move beyond bilinear style texture filtering (which has been around since the Voodoo 1) and have quadratic (for example) available? That would be sweet.

Easy there, calm down.

One step at a time.

Nothing in that post is technically incorrect Dave.

And if anyone is capable of being a game changer, it's nvidia and they've proven it at least twice in the past - the original geforce and the geforce 3.

What are you talking about?

Nvidia is miles ahead of AMD who is depending on their future Fusion to do what Nvidia is doing now. ATi division is 'gaming-only' with a token given to GPU computing, so it looks like they are competitive. AMD has little support for their products; they are so broke they cannot even support the games - like Batman AA which they ignored completely.

Fermi is the real thing. It will open a new frontier in many brand new areas. Nvidia is innovating. AMD is stagnant (in GPU computing).

Stagnant is SUCH a descriptive and appropriate term when it comes to describing ATi!

Well, i am specifically referring to GPU computing (also physX and 3D).

i am certain that half of our editors would disagree with me. But AMD pays lip service to innovation when they are just following Nvidia's lead. There is no ecosystem of support at AMD because they are broke.

otoh, Nvidia has an excellent ecosystem of support for their devs and GPU computing but their support for emerging media is completely broken