Well, after the AMD Radeon Pro Duo surfaced and was given a soft launch, nVidia GeForce hits out with their first new Pascal GPU that will be made available to the market in 2 weeks. The GTX 1080 is going to be a nice interesting addition, and something which I plan to replace either my 2 GTX 970s I have in my gaming system, or start my i3 4330 build with it.
The coverage and the content of this card has been absolutely promising, to say the least. Although it uses the middle ground GDDR5X as to the HBM of the Fury and FuryX, and the current GDDR5 Standard. The differences between GDDR5X, HBM and GDDR5 cannot be defined just yet because HBM's clocks are rather slow in respect to its 1024-bit bus width while the GDDR5X 1080 can support higher transfer rates at a base of a good 10 Gbit/s.
With that said, the GTX 1080 has the following specs.
2560 Pascal CUDA Cores - Now Pascal Cores and Maxwell cores can't be compared but the performance has been said (with leaked benches) to surpass the 3072 Maxwell Core Titan X. I can definitely foresee a Titan in the future because on Wikipedia, their first NVLink chip which uses the GP100 core that is said to house 3584 Pascal Cores (40% more cores so there should be more SMX units on GP100 than it does on GP104 but don't quote me on this one).
1607 Mhz Core, 1733 MHz Boost Clocks - This is a VERY (Impressively) scary clocks out of the gate, even at stocks. There have been photos that it can overclock past even the 2.1 GHz mark, but whether it was done air cooled or water cooled cannot be determined just yet.
8GB GDDR5X VRAM: A new counterpart to Fiji's High Bandwidth Memory (HBM). Double of the predecessor, made even better thanks to a double clock off the GDDR5X RAM speeds. GDDR5X is a new memory standard which integrates an updated RAM controller. It requires a new PCB with it, having 190 vs 170 pins of standard GDDR5. With a Bandwidth of 320GB/s, that will be good for High Res / VR Gaming. It'll be 256-bit, same as a 980. I won't be surprised if the Ti version in the distant future would be 384-bit.
599 USD Standard, 699 Founders: Not too bad for pricing wise. Especially if you can find future AIBs on the base 599. Because that fits just between the Launch MSRP of the 980 and the 980 Ti. And it's no contest with the Titan X either. And even with that performance at the founder's price of 699, it seems pretty damn good. It's clear that a single 10 can already flat out beat a 980 Ti without a problem by 30%. It'll be behind the Pro Duo, but not all that much, say 15-20%. And pair two 1080s for a combined price lower than a Pro Duo, it is a serious contender.
API Support: Pascal will be the first to support the Vulkan API. What Vulkan is said to do, is to balance CPU either by lower the CPU overhead and dedicate full power of the GPU to the application while the CPU is free for its usual computation workloads. Or balance out the CPU load across most if not all the cores so the cores are used at the optimal level. Vulkan will also be supported on cards all the way back to Kepler GTX 600 series with a driver update.
180W TDP: The 180W TDP lies closer to the 165W 980 than it is to the 250W Titan X, 980 Ti, R9 290X which are still below the 275W of the Fiji and Hawaii based Fury and 390X. That TDP is even lower slightly than the respectably efficient Tongas. Power efficiency like Maxwell should be quite nice, and it requires a single 8-pin.
There have been leaked Fire Strike Benchmarks, made known by Overclock3D which you can see here:
The scores look to scale well with the Core Overclock. There's also been a recent leak of AIB partners that might release a Liquid Cooled one clocked at 2.5 GHz. And if that scaling is true, this may be a serious contender with the Radeon Pro Duo once you turn up the clock speeds past the 2 GHz Ceiling.
Having said that, I'm edging to see from the bigger tech reviews whether they can pit the GTX 1080 (Stock and Overclocked speeds) against:
R9 295X2
R9 390 Crossfire
R9 390X Crossfire
R9 Fury Crossfire
R9 Pro Duo
GTX 970 SLI
GTX 980 SLI
GTX 980 Ti / Titan X SLI
Then again, only time will tell!
No comments:
Post a Comment