Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Nvidia Pascal GPU has 17 billion transistors
So the 1080 is 2x as fast as the titan X in VR. It is looking to be 25% faster or more than the titan x. The 1070 will be 379 and faster than a titan x. Yet I read people on anandtech whining that the 1070 will be gimped way further than the 970 was???

It actually looks like it will end up very much the same. The 970 was around 20% slower than the 980. So much so, you could not overclock and catch up. Even at 1550mhz, it still couldn't catch up the a 980.

People are really being silly. But the ones posting so far seem to be mostly diluted fanboys. The big and loud gang seem mute so far. I think they are trying to get their jaws strapped back to their heads so they can spend the rest of the night trying furiously to come up with some kind of attack plan. I have seen a few people posting, well we got to wait to see what AMD launches.

It's pretty clear. A 290ish performance at 150 watts is a huge flop man. A massive flop. AMD has some really big shoes to fill here. Even on price, they are gonna have a hard time selling those. Many many pc gamers have 28nm cards now. And the performance polaris is looking to be at, it's not enough to make many people go out and buy. It's 2013 performance. At 300 bucks, it's a fail. At 250, it's not anything to get excited over.

Pascal really really really moves the performance per dollar mark. Especially on the 1070. Pascal seems to be moving the performance per watt mark by a whole lot too.
I just think amd has some big large shoes to fill. They may be getting left behind, far behind this time. I truly believe AMD is gonna be in a difficult position and have been holding out on their launch. But really, this is speculation. Perhaps amd has been hiding the fact that their polaris is gonna run at 2000mhz core clocks. Perhaps the sub 1000th clocks are just to mislead.

You got to think about it. Polaris 10 seems to have similar core counts to gp104. Nvidia and amd core counts vs performance hasn't been equal but they aren't extremely way off either. We are looking at polaris running at half the gp104 speed alone. Not counting the differences that there may be in performance core to core. Gp104 looks to demolish this time.

But you hear people saying, "well at least we know that amd will win in mobile and Polaris will have better performance per watt". I say, I highly highly highly doubt that. The gp104 is looking really efficient at almost 2000mhz with their flagship. I am pretty sure that this translates to very good even for pascal lower tdp solutions. It seems highly unlikely amd will run away with efficiency this round. If polaris manages to edge out pascal mobile, it simply won't be by a lot. It seems like a tall task and I wouldn't bet on polaris being more efficient at all. They will sell chips that perform slower. You can get that for less tdp.

It seems to me though, if a 2000mhz pascal can be that efficient then a 1000mhz version cut down could be significantly more power efficient. So much so that I cannot imagine where it stands.

See, amd is claiming 2.5x increase in perf per watt for a card that may perform around that of a 390. Nvidia is showing 2x increase in per per watt for a card faster than a titan x =>25% or so and running +1800mhz. While both amd and nvidia could be exaggerating with best case scenarios, we also know that nvidia had a very notable perf per watt edge with maxwell already. So, this 2.5x amd speaks of, even if it is exactly that, mathematically does not appear to me much of an edge. And since we are speaking of best case 2.5x for amd and we see nvidia promoting 2x with their ultra fast flagship.....I am led to believe that this idea that Polaris will be ultra efficient compared to pascal seems illogical.
The numbers we have from amd and nvidia don't suggest this at all. Especially since amd bases their gains on their 28nm chips and nvidia bases theirs off maxwell.

It's just not adding up. I think anyone who thinks amd will base their 2.5x off their more efficient hbm water cooled fury (which still wasn't more efficient than the 980) they are dreaming. Their polaris 10 is not a replacement for fury x. I am thinking this will be based off the 290x or 7970ghz. And if it is any other chip that fury X, well...you do the math
Reply


Messages In This Thread
RE: Nvidia Pascal GPU has 17 billion transistors - by ocre - 05-07-2016, 10:34 AM

Forum Jump:


Users browsing this thread: 1 Guest(s)