The AMD Dual Fiji videocard was briefly shown during an E3 presentation and the Fiji introduction already, but after these two moments in the summer everything went silent. Aside from a few photos th... AMD Dual Fiji videocard surfaces
I smell a severely throttled card with only 2x8 pin power connectors the fury x on its own consumes 275w.
can be twin Fury (without the X) and it may be more logical. the regular Fury can fight on the ratio price/perf on the market, the X isn't enough powerfull and is too expensive
Price 125,350 INR = 1,920.45 USD. Reason I never bought GPU from India, where I live. Sometimes it even cheaper to buy from Newegg/Amazon international and get it shipped here. No warranty but much better price.
Supposed Pascal GP100 has shipping tag around 200k INR. Dual GPU cards were always bit pricier, but they have their customers. Considering it will require bit different cooling than Fury X, price may be actually good. With Nano performance and power consumption I see no problem for Dual GPU Fiji to have lower power consumption than 295x2. It can actually have like 1.8x performance of Fury X while eating only 20% more energy.
I think if both chips are Nano based this will work just fine without much throttling. Honestly, I think having 8GB of HBM is what will make this perform so well. I don't think the extra processing cores will have as much of an impact on the performance.
I think you'll find the 8GB of HBM will be split into 2 sets of 4GB and each core will only be able to access 4GB. This is the way other dual cards have worked in the past.
Is bandwidth really the bottleneck? The 980Ti seems to do fine with 60GB/s less bandwidth. Also Crossfire/SLI don't double memory bandwidth either. DX12 might change this but it's yet to be seen. Regardless I think this will be a pretty beastly card, especially if it ships with a AIO cooler like the Fury X.
Is that a serious question? In many (not all) forms of multi-GPU rendering, both GPUs work together to render the same image. They effectively share the same data in their memory. So when you have 8GB of VRAM in a dual GPU setup, you really only have access to 4GB (until DX12/Vulkan come around to change this behavior). Therefore, you get roughly double the memory bandwidth, rather than double the capacity.
And? Depending on the application or API, Crossfire and SLI double VRAM bandwidth... A dual-die GPU is a crossfire/SLI configuration.
Most multi-GPU setups use AFR, they don't work on the same image. Regardless, the textures are loaded into both cards anyway, so the bandwidth is the same across both. In compute this isn't the case, it's doubled, but for most games the same resources are loaded on both cards. Again DX12 can change this, but no developer has done it yet.