• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Big Polaris or Big Pascal, which do you think will win the fastest GPU crown.

JediFragger said:
I'm realllllllly hoping that DX12 gives Crossfire the shot in the arm it needs!!

I'm betting on AMD this time, mainly because they *need* the win! If not then I'll settle for the 'better value' crown ;)
Click to expand...

You can see how well explicit MGPU works using mantle in BF4. But the same with mantle, a game dev needs to code a mgpu system into their game engine using dx12.
 
Mauller said:
You can see how well explicit MGPU works using mantle in BF4. But the same with mantle, a game dev needs to code a mgpu system into their game engine using dx12.
Click to expand...

Agreed, but it really does seem that the older WDDM framework held it back somewhat too. It'll be nice to see what it can do with WDDM 2.0/DX12 :)
 
JediFragger said:
Agreed, but it really does seem that the older WDDM framework held it back somewhat too. It'll be nice to see what it can do with WDDM 2.0/DX12 :)
Click to expand...

Yeah, I read about that on Johan Anderssons twitter. Them having to make workarounds for WDDM that lowered performance. He even said their mantle implementation would have run better on WDDM 2.0 if coded for it.

Makes me wonder if the vulkan crowd will have these problems and just choose to code for win 10.
 
I think NVIDIA will be so late with volume (and probably so much more expensive) that it really won't matter even if they're miles faster, which is very unlikely in itself.

I expect AMD will be faster though ... they've done interposers, HBM and have not stripped all the compute functions out of their previous lines of GPUs, and GCN 4.0 is just a considerable refinement of existing GCN which already has an onus on parallelisation and asynchrous compute and shaders. Plus, I'd expect Samsung (and to some extent by extension GF) to have immensely better yields and lower costs than TSMC meaning they can afford to go with more transistors and push clocks much nearer to the max that the designs allow.

NVIDIA will be doing interposers, HBM, readding compute functions (which take up space), doing their first architecture that's aimed at addressing paralellisation and asynchrous compute and shaders. Also, Pascal and HBM have been a rush job. When Volta was delayed another 2 years, having already been delayed, they needed something to act as a stop gap and stop themselves getting wiped out in enterprise and supercomputer space, and losing lots of share in consumer / gaming. Volta was the real new ground up design for parallelisation, asynch and compute and was meant to use HMC not HBM (as far as we know it still does). Pascal may just be a heavily modified Maxwell. Also, the focus remains on industry (with NVLink and automotive) for Pascal ... the design may have lost out on development time for consumer space. They'll be doing all this on a much smaller, FINFET node that's not quite as dense as Samsung's and it'll likely be in short supply at a higher cost per wafer.

So much can go wrong for NVIDIA, and you have to think that if they were confident for next gen. they wouldn't be so hell bent on consumer and developer lockin with Gsync and Gameworks, and alienating so many people as a result.

I suspect AMD could be 20% faster than NVIDIA in absolute terms .... but maybe they won't launch a BIG chip soon or at all if they don't have to, as they may have such a huge lead in time to market, price and availability. I'm sure NVIDIA's top SKU will be above 450mm2. I'd expect AMD's to be less than 375mm2 initiallly, and only later in the year a big boy IF they need it.

So let's say FPS/$ lead of 30-40%.
 
I'd say both at stock the pascal chip will be around 7% faster and when both overclocked to the max I'd say pascal will be around 20% faster as Nvidia chips are prone to better overclocking. This is at 1080p of course I'm making my guess on
 
Tommy_Here said:
I'd say both at stock the pascal chip will be around 7% faster and when both overclocked to the max I'd say pascal will be around 20% faster as Nvidia chips are prone to better overclocking. This is at 1080p of course I'm making my guess on
Click to expand...

The only reason Maxwell was overclocking so well is due to the amount of stuff the stripped out of its architecture. Once added back with Pascal it won't have as much overclock headroom.
 
I think the battle will be taking place a year from now or more.

No idea how fast the fastest will be, just for a giggle I'll say Polaris 11% faster stock and 10% OC'd.
 
Last edited:
Pascal 5% faster at stock
Pascal 18% faster when overclocked

But then I don't think big Polaris will be with us in 12 months time, not if medium Polaris is coming sometime before September.
 
die size should logicaly go to 16nm, the process is more mature than 14nm, so my guess high end will be for pascal, low/mid for Polaris
 
AlamoX said:
die size should logicaly go to 16nm, the process is more mature than 14nm, so my guess high end will be for pascal, low/mid for Polaris
Click to expand...

IIRC TSMC 16nm actually results in a slightly larger core but also more power efficient when all else is equal resulting in about 2-3% net lead to 16nm for the same design.

(Does give potential for a mature 14nm to eventually take the lead).
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom