Originally Posted by sdlvx
384GB/s theoretical bandwidth in this card.
GK104 has 192GB/s and loses to 150GB/s 7870 at 1080p with 8xMSAA when you benchmark for VRAM bandwidth.
It's over, Nvidia is relegated to 1080p rig status. This is probably going to be the only thing remotely capable of pushing 1440p and better on next gen console ports. This thing is gonna have 100GB/s more theoretical over GTX 780, and as Tom's showed in their "testing for bandwidth limitations" review Nvidia theoretical bandwidth isn't close to AMD theoretical bandwidth.But I do like how the top Hawaii is being saved for FirePro. What happened guys? I thought AMD was going to focus on APUs only? You mean they're focusing the most on high performance HSA workstations?
Not to mention die size of Hawaii is like 420mm^2 (yeah, frame the bundle 420 yolo) and GK110 is around 550mm^2.
Gk110 is about 30% bigger than Hawaii. Meaning AMD can charge 30% less and if they do better than 30% behind Gk110 then they've got a great GPU on their hands in regards to die size to performance.
And look at that, 30% off of GTX 780 price of $649 on newegg is almost exactly $449. If this thing is 15% slower than Titan or GTX 780 and it comes in at $449, it's going to be a heck of a deal.
I believe you're being overly dramatic but it has been stated that AMD has been teaming up with Adobe to support Premiere.
8xx msaa isn't the limit for Nvidia....
Solidworks synthetic , anti-aliasing
Since a lot of non-fanboy Nvidia users buy Nvidia due to need of CUDA for Adobe Premiere and other such applications that don't make use of Quadro drivers this could be a big win for AMD if they get their driver issues and perception issues sorted out.
Some applications don't make use of CUDA at all, such as AutoCAD (the 3D mode), but just happen to run better on Geforces.
On the flipside compute applications are destroyed by AMD card, such as Autodesk Inventor, Monte Carlo/Biniomial Option pricing, crypto; benches such as GPU Caps Viewer PostFX (originally part of Nvidia's development kit) and BasemarkCL I'd say GCN has a lot of untapped potential. OpenCL has only been around since 2009 after all.
Also if you compare the progress AMD has made with GCN, which performs well even under 8x AA (unlike Fermi and Cayman), I'd say they made huge progress. Jacket, the hugely popular MATLAB toolbox has started to support OpenCL. Simulation programs such as SIMULIA and ABAQUS support OpenCL, but the most widely used simulation program, ANSYS , uses CUDA compute acceleration
that is only supported on Quadro K5000 /5000 and above ($2K GPUs are out of reach for most people anyhow).
We already see Dell, HP , and Lenovo implementing Firepros into their ultrabook workstations
and Apple is using Firepro in their new Mac Pro.
Cape Verde and Pitcairn are pretty power efficient , we see it used as a basis for their mobile GPUs too.
Regardless, I'm getting sick of the hyping. We need to see real performance numbers.
edit: @ Stay Puft, the Nvidia reference hardware quality is not up to par and it's heavily locked down. Their drivers are better generally (due to ingrained use by developers and the TWIMTBP titles), that's the difference. I'm amazed they let MSI/ EVGA and others use LN2 BIOS switches.Edited by AlphaC - 9/21/13 at 12:53pm