When we talk about memory limitations we're still talking in terms of GDDR. HBM may be something of a game changer in that respect. We've certainly seen that it's not a limitation for Fury X so possibly you can't apply the same rules and assumptions to HBM that you do with GDDR.
You forget that AMD have 2 engineers sitting and hand manage memory for the Fury series. 4GB is 4GB no matter what memory type. As soon as those 2 people get assigned somewhere else the card house falls. Or when they hit something they cant do tricks to get around.
Well the GTX660 and GTX660TI had the same issue of the last 512MB of VRAM being addressed more slowly too.
980Ti is slower than Fury X at 4k according to TPU despite games like anno2205 in the mix. So 'very, very few games' is a figment of your imagination. Even games where Fury X is leading by double digits aren't that rare.
You forget that AMD have 2 engineers sitting and hand manage memory for the Fury series. 4GB is 4GB no matter what memory type. As soon as those 2 people get assigned somewhere else the card house falls. Or when they hit something they cant do tricks to get around.
You forget that AMD have 2 engineers sitting and hand manage memory for the Fury series. 4GB is 4GB no matter what memory type. As soon as those 2 people get assigned somewhere else the card house falls. Or when they hit something they cant do tricks to get around.
Drivers,Not really 980 Ti vs Fury X comparison
No kidding here. Even the lowly 7870 (and derivatives) puts the hurt on the 960, which aside from bandwidth, is theoretically superior on all fronts. Given Pascal is quite far off from launching, there's zero credibility behind the Nvidia gimping Maxwell conspiracy.Man, I don't like the trend. Maxwell is going to continue to tank going forward. Those of you who were able to grab a R9 290 (non-reference) for ~$200 got a great deal. As consoles developers learn to extract more power from it, I expect the GCN cards to pull away.
nvidia won't 'gimp' maxwell in favor of pascal because pascal is more or less the same architecture as maxwell. I doubt they're adding anything on the gaming front like async compute.
What can however happen is the GCN3 in Fiji gets properly utilized and scales much better over 390X than currently. Especially with dx12 where AMD would be runaway winners if Fiji scaled well compared to Hawaii.
why not?historically speaking nvidia does that its one way to push their customers to buy the new card1. We don't know enough about Pascal to say with any certainty, it's all a guess.
2. Fiji compared to Hawaii will always be front-end bottlenecked unless it's got Async Compute tasks to do. This is because they share the same front-end design for gaming purposes, but Fiji has many more shaders.
There's no way NV is gimping Maxwell now, since Pascal is still far away, that theory needs to die in a fire.
It's simply due to modernized game engines being optimized for console GCN to extract all the performance they can get out of the weaker hardware.
All the talk so far from Pascal is that it's returning to a compute focused design, with good mix-mode and high DP capability. I suspect it will be GCN-like, with multi-engines.
why not?historically speaking nvidia does that its one way to push their customers to buy the new card
http://www.overclock.net/t/1529108/are-nvidia-neglecting-kepler-optimization-since-maxwell-release
one of the countless thread about it.. who can actually safe bet that nvidia will not pull again this stunt once pascal gets released?
You can pretty-much assume that once Pascal hits, NV will start promoting Async Compute and V.R., two things that Maxwell: 1) can't do, and 2) sucks at. Then the chips will fall naturally, and the Maxwell generation of cards will be forgotten, just like the Kepler generation pretty-much already has. And of course, you aren't a true NV believer if you don't own their newest architecture. The way (customers) were meant to be played, LOL.
You can pretty-much assume that once Pascal hits, NV will start promoting Async Compute and V.R., two things that Maxwell: 1) can't do, and 2) sucks at. Then the chips will fall naturally, and the Maxwell generation of cards will be forgotten, just like the Kepler generation pretty-much already has. And of course, you aren't a true NV believer if you don't own their newest architecture. The way (customers) were meant to be played, LOL.