Alleged Nvidia GeForce GTX 880 gets pictured

[Adam Elteto]

That's why the VRM section is pixelated! To hide the secret sauce connection...

Impressive! A steampunk graphics card!

 

[dividebyzero]

I find it hard to believe it will cost less than the 780 with the possibility of 8GB of VRAM.

The 3GB 780 or the 6GB 780 ? I'd say there's a pretty good chance that the 3GB version will go EOL if GM 204 meets expectation, so that would leave the field clear for high-clocking B1 silicon 6GB cards. If the GTX 880 is 256-bit then there could possibly be a market for the 780 due to its wider bus width.

Extrapolating the architectural gains of Maxwell over Kepler so far, it seems likely that GM 204 could offer 10-20% performance over the 780 Ti/Titan Black at 1080p/1600p, but that would probably evaporate at 4K/Surround resolutions and/or heavy compute scenarios. Of more interest to most people might be the salvage parts; a GTX 870 and 860 Ti could net you ~ GTX 780 performance for $300-350

 

[amstech]

With todays GPUs sucking so badly at running 4K, I wonder how this next generation will perform.

 

[Jad Chaar]

The 3GB 780 or the 6GB 780 ? I'd say there's a pretty good chance that the 3GB version will go EOL if GM 204 meets expectation, so that would leave the field clear for high-clocking B1 silicon 6GB cards. If the GTX 880 is 256-bit then there could possibly be a market for the 780 due to its wider bus width.

Extrapolating the architectural gains of Maxwell over Kepler so far, it seems likely that GM 204 could offer 10-20% performance over the 780 Ti/Titan Black at 1080p/1600p, but that would probably evaporate at 4K/Surround resolutions and/or heavy compute scenarios. Of more interest to most people might be the salvage parts; a GTX 870 and 860 Ti could net you ~ GTX 780 performance for $300-350

Interesting thoughts. I don't think they will move from 384-bit to 256-bit though. They need that extra bandwidth for 4K and in order to keep up with AMD, I wouldn't expect them to downside.

How much money would it save nVidia if they were to downsize the memory bus? Because if the rumors are true that the price will be lower, this could be one of the places they cut back... not sure what to think of it though.

 

[DDG4005]

Hopefully the Maxwell GPUs will be powerful enough to drive a 4K display with only a single connection. Otherwise...

 

[Jad Chaar]

Hopefully the Maxwell GPUs will be powerful enough to drive a 4K display with only a single connection. Otherwise...

I think that won't come till Volta or whatever the new name is. Volta brings huge memory bandwidth improvements that may help a lot.

@dividebyzero is memory bandwidth the issue? If so, would my assumption about Volta be correct?

 

[DDG4005]

Hopefully the Maxwell GPUs will be powerful enough to drive a 4K display with only a single connection. Otherwise...

I think that won't come till Volta or whatever the new name is. Volta brings huge memory bandwidth improvements that may help a lot.

@dividebyzero is memory bandwidth the issue? If so, would my assumption about Volta be correct?

Makes sense. I guess that means no 4K display for me (not that I can afford one anyway).

 

[dividebyzero]

@dividebyzero is memory bandwidth the issue? If so, would my assumption about Volta be correct?

Once 4K becomes more standard it will become a little more pressing, but the big "need" for bandwidth will come when both AMD and Nvidia need to put daylight between discrete graphics and IGP. Both vendors need to differentiate their discrete graphics from the integrated option- AMD because APUs don't deliver on high margins and high visibility the way high-end graphics does, Nvidia because the high-end big GPU is also needed for the professional Tesla and Quadro markets. Somehow I couldn't see Nvidia surviving if it had to rely on Tegra!

I think you can pretty much guarantee that both AMD and Nvidia are working on the next cycle of image quality enhancements for game engines. TressFX and GameWorks Hair/Wave/FaceWorks at the moment are trailers to the main feature. Once the graphical horsepower exists to comfortably (at the high end) run extensive physics and moves beyond pure rasterization, bandwidth will again need to make the leap as it did from 32, 64, 128, and 256-bit.

Interesting thoughts. I don't think they will move from 384-bit to 256-bit though. They need that extra bandwidth for 4K and in order to keep up with AMD, I wouldn't expect them to downside.

GM 204 is a replacement for the GK 104 (256-bit) GPU, not GK 110.

From the latest pictures GM 204 scales out at approximately 420mm². If it is to accommodate a high number of cores then the uncore needs to kept in check. Uncore typically makes up half the silicon. Some is fixed in size regardless of the segment the GPU will be sold in (PCI-E interface, display out, command processor,video transode/encode etc), and some is variable depending upon design. The largest uncore die area is devoted to memory interfaces and memory controllers. If the numbers are correct (420mm² / 3200 cores) then it packs a substantial increase over AMD's Hawaii (438mm²/ 2816 cores). Given that transistor density can't radically change on the same process node, something has to give - the logical start point would be bus width.

My last post extrapolated a GTX 870 and 860 Ti as being the sweet spot for most people with GM 204....seems today Charlie D is on board with the salvage parts also....why do I feel.......dirty?

 

[Jad Chaar]

Once 4K becomes more standard it will become a little more pressing, but the big "need" for bandwidth will come when both AMD and Nvidia need to put daylight between discrete graphics and IGP. Both vendors need to differentiate their discrete graphics from the integrated option- AMD because APUs don't deliver on high margins and high visibility the way high-end graphics does, Nvidia because the high-end big GPU is also needed for the professional Tesla and Quadro markets. Somehow I couldn't see Nvidia surviving if it had to rely on Tegra!

I think you can pretty much guarantee that both AMD and Nvidia are working on the next cycle of image quality enhancements for game engines. TressFX and GameWorks Hair/Wave/FaceWorks at the moment are trailers to the main feature. Once the graphical horsepower exists to comfortably (at the high end) run extensive physics and moves beyond pure rasterization, bandwidth will again need to make the leap as it did from 32, 64, 128, and 256-bit.

GM 204 is a replacement for the GK 104 (256-bit) GPU, not GK 110.

From the latest pictures GM 204 scales out at approximately 420mm². If it is to accommodate a high number of cores then the uncore needs to kept in check. Uncore typically makes up half the silicon. Some is fixed in size regardless of the segment the GPU will be sold in (PCI-E interface, display out, command processor,video transode/encode etc), and some is variable depending upon design. The largest uncore die area is devoted to memory interfaces and memory controllers. If the numbers are correct (420mm² / 3200 cores) then it packs a substantial increase over AMD's Hawaii (438mm²/ 2816 cores). Given that transistor density can't radically change on the same process node, something has to give - the logical start point would be bus width.

My last post extrapolated a GTX 870 and 860 Ti as being the sweet spot for most people with GM 204....seems today Charlie D is on board with the salvage parts also....why do I feel.......dirty?

Thanks DBZ. I can always rely on you .