Rumored Specs of Upcoming GeForce GTX 880 Appear Online
Details have been leaked about the upcoming GTX 880 graphics card from Nvidia, though we're not sure how much of it we're supposed to believe.
A number of rumors about the upcoming GTX 880 graphics card from Nvidia are floating around the web. These rumors are about the specifications, and while some of them are very believable, others are somewhat questionable.
We'll start off with the CUDA core count, which is rumored to be 3200. This is a fairly believable number, as the GTX 780 Ti's Kepler-based GPU features 2880 CUDA cores. The Maxwell-based GTX 880 is also rumored to feature 32 ROPs as well as 200 TMUs. These are both fewer than the GTX 780 Ti, though we have to remember that this is a different architecture. The GPU is expected to run at 900 MHz base, with a boost frequency of 950 MHz.
Memory aboard is expected to be 4 GB of GDDR5 memory, running at an effective speed of 7.4 GHz. Very fast, but believable. Then there is the memory bus, which is the most questionable part of the rumored specification list; the memory bus is expected to 'only' be 256-bits wide. This will limit the memory bandwidth to 238 GB/s, while the GTX 780 Ti managed to push 336 GB/s with its 384-bits wide memory interface.
Other specifications include that it will be built on the 20 nm lithography process, which is not news, as well as that it will have 7.9 billion transistors. The board is also expected to have a TDP of 230 W, which makes sense as Maxwell is a more efficient architecture and the smaller lithography process will also cut energy consumption.
There is one last rumor going around, which is that this graphics card is not actually intended to be the successor to the GTX 780 (Ti). Â Given the reduced memory bandwidth, we're not sure exactly where this product will be positioned if the above specifications are correct.
Everything above is still very much in the rumor phase though, so as always, do be sure to take it with a heap of salt.

The memory wide is actually quite believable because Kepler seems to be reasonable well feeded even with narrower memory bandwidth.
I'd prefer a die shrink and less power consumption or more performance. When is the die shrink from 28nm coming?
For the same reason that the GTX 660 used the GK104 chip instead of the full GK110 chip. Nvidia's mid-range GK104 was performance-competetive with AMD"s high-end Tahiti chip found on the HD 7970. Nvidia was able to take their mid-range chip and sell it at high-end prices because it outperformed the competition and would sell at that price. Then while AMD evolved their GCN architecture for the Hawaii chips in the R9 290 series Nvidia was able to sell off their high-end GK110 chips for top dollar as Tesla compute cards and eventually roll those chips into GeForce cards for the 780, 780Ti, Titan, and Titan Black.
My guess for this generation is that it's the same deal. Nvidia feels that their mid-range GM104 chip will be competetive with AMD's offering so they will sell the GM104 as the GTX 880 and hold onto the larger GM110 chips for high-margin Tesla cards and roll them out later as the GTX 900 series.
I'd prefer a die shrink and less power consumption or more performance. When is the die shrink from 28nm coming?
At the end of this year...
Since it's nVidia, they'll probably have a "Titan 2" down the road to bleed fanboys out of their money later on, haha.
If the specs are actually true, they sound more like an updated revision instead of a higher tier Maxwell GPU (was it Maxwell?). The 4GB of VRAM are actually hurting them in the 4K territory thanks to what the R9-295X showed, so they might be cooking something in between to justify the stupid price tags they are asking as of late.
Cheers!
in no particular order:
1. 3dfx voodoo 2s in SLI
2. geforce 256
3. 8800GT
4. GTX 570
in no particular order:
1. 3dfx voodoo 2s in SLI
2. geforce 256
3. 8800GT
4. GTX 570
I had to log in and copy pasted my message and lost the first part.
It was supposed to start with "I hope it lasts as long as my original 8800 GT"
Sorry for the double post
that's a bold strategy cotton. let's see if it pays off for 'em.
5 GPCs
24 SMMs
10MB L2 Cache
3072 Shader Cores
192 Texture Units
80 ROPs
640-Bit Memory Bus
6GB GDDR5 RAM
If it's a GTX 780 successor, disable 1 GPC:
4 GPCs
20 SMMs
8MB L2 Cache
2560 Shader Cores
160 Texture Units
64 ROPs
512-Bit Memory Bus
3GB GDDR5 RAM
I personally would love this. But if we're throwing out rumors, then here's a rumor via logic.
They will release the 880 at $225, and it will be around 85% as fast as a 780 TI
My guess for this generation is that it's the same deal. Nvidia feels that their mid-range GM104 chip will be competetive with AMD's offering so they will sell the GM104 as the GTX 880 and hold onto the larger GM110 chips for high-margin Tesla cards and roll them out later as the GTX 900 series.
Not this stupid myth again...
The GK104 was a high-end GPU. It's almost as big as AMDs Tahiti, and much bigger than AMDs midrange GPU at the time, Pitcairn.
If you want to get into the discussion about who got the most out of each square mm of die, then it's AMD: The R9 290X is only slightly slower than the GTX 780 Ti, even though Hawaii is much smaller than GK110.
The size difference between GK110 and Hawaii is 123 square mm. The difference between Tahiti and GK104 is only 58 square mm. And the difference from what you call Nvidia's high-end GPU, the GK110, to what you call AMDs high-end GPU, Tahiti, is a whopping 209 square mm. There's no way these GPUs are in the same league. It's like comparing a humvee with a tank.