Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
GTX 1070 already reviewed
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
It looks like you're new here. If you want to get involved, click one of these buttons!
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
Comments
It's possible that the bottleneck is some fixed-function graphical hardware to some degree, which would explain the results if the GTX 1070 is created by disabling parts of several GPCs while still keeping all GPCs active. But I'd bet on Nvidia disabling a GPC outright in at least some GTX 1070s, if not all of them. If they do so in some but not all, then you could get substantially different performance in different GTX 1070s, which would really be two different cards entirely--and they'd surely send only the higher performing ones with all GPCs active for reviews. A CPU bottleneck is the more likely explanation, though.
The real comparison, which we don't have yet, is to the top Polaris 10 card. It's likely that the GTX 1070 will be both faster and more expensive than Polaris 10, but by how much on both counts is a huge matter. Paying 10% more for 50% more performance is a much better value than paying 50% more for 10% more performance.
We'll see when they're actually available. The nominal launch date is June 10, but that could be anywhere from a very thin soft launch (like the GTX 1080) to an actual hard launch.
1080GTX appears to not be as a big deal as PR is making it out to be
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
Instead of this there (1 whole GPC disabled) it showed 1 cluster per GPC disabled, and another cluster in the bottom right GPC disabled to make up for the 5.
The explanation was if i recall correctly that this way all a repeat of 3.5GB+0.5GB impossible (which was wierd to me because i was under the impression that it only mattered that ROPs weren't cut for that)
I found the other cores diagram :
And that this would be what you mentioned Quizzical, with Nvidia sneakily sending the non-gimped version for testing and for instance the first founder's edition batch, and then after the reviews are over, sending out the second batch to unknowingly destroy performance ?
And how does this correlate to the ROP count ?
If the ROP could would hypothetically be the same in both versions, would the gimp only be slight then ?
I think GTX 1070 might be the big deal here. If NVidia can just launch and manufacture the non-founders editions it'll be really good performance at decent price.
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
The second version of a GTX 1070 would be a better card than the first version. I'm not sure how much better, but for Nvidia to freely mix the two and not tell you what you're getting is definitely not kosher. That would be worse than the 3.5 GB fiasco with the GTX 970, in which all cards were the same and the launch reviews were representative of what you actually got.
I don't know if Nvidia will do that, though. If only one version exists, that's fine, of course. If both exist but are labeled differently (e.g., the 768 MB and 1 GB versions of the GTX 460), it's fine. I'd even be okay with it if both cards exist, but the worse one gets sent out for reviews--but of course, Nvidia would never do that. But sending out a ringer for reviews that isn't representative of what gets stuffed into retail boxes is definitely not okay.
So why do I suspect that Nvidia might do exactly that? The reason for the GTX 1070 is so that you can still sell chips that are partially defective. Nvidia and AMD have both done this for many years, and there's nothing wrong with it.
But chips can be defective in different ways. Suppose that a raster engine is defective. In that case, if the other three GPCs work, you can disable the one with the broken raster engine and you get the first graph. Suppose instead that two compute units from different GPCs are defective, but everything else works. In that case, you can disable five compute units, including the two defective ones, and you get the second graph.
Both of those are going to happen in the real world, though I have no idea how common they will be. But Nvidia surely wants to sell both types of parts. If they have a further salvage part (say, a GTX 1065) that claims, say, 13 compute units, 3 GPCs, and a 192-bit memory bus, most GP104 chips will be able to meet those requirements, so they can sell anything that doesn't fit the second graph as a GTX 1065. And Nvidia might do exactly that; that's basically what they did with the GeForce GTX 465, for example, which was completely above board.
The GTX 970 couldn't do this. It had 13 compute units enabled, while disabling an entire GPC would leave only 12. Of course, GM104 was on a much more mature process node, so it probably had much better yields than GP104 does so far. Disabling 1/4 of your compute units for your top salvage part is an awful lot, but it's what you have to do if you're going to disable an entire GPC and you only have four of them.
Hopefully someone in the tech media will ask Nvidia directly. This does matter, after all.
거북이는 목을 내밀 때 안 움직입니다
거북이는 목을 내밀 때 안 움직입니다
http://www.hardwareunboxed.com/nvidia-geforce-gtx-1070-review-22-games-tested-at-1080p-1440p/
he also has OC testing:
http://www.hardwareunboxed.com/geforce-gtx-1070-overclocking/
거북이는 목을 내밀 때 안 움직입니다
The 1070 and 1080 have vastly less overclocking room because of two reasons :
1. New immature node
2. More aggressive and higher base clock set boost 3.0
It's basically as if u would take a 20% overclocked 980 Ti from a board partner and make it into the founder's edition, just without adding better cooling, or proper power delievery.
It's absolutely an ingenious move to make your semi-paper tiger even more fierce