I expect them to release info on the GT 1060 on a Friday Afternoon. If it's performance match the RX 480 in DX11 games then it needs a clear distinctive feature that separates itself from that card to be priced over $200. AMD cards just perform better in DX12, and you can't reasonably assume fanboy-ism to maintain demand for the card. I expect a straight die shrink of something like the 970 where the power draw is the touted feature of the card. However, it's really tough to say if it can draw less power than the RX 480. While nVidia's architecture draws less power than GCN, its also on 16nm verse AMD on 14nm.
While nVidia's architecture draws less power than GCN, its also on 16nm verse AMD on 14nm.
Don't read too much into 16 nm versus 14 nm, as it's not at all clear whether Global Foundries 14 nm is better than TSMC 16 nm. The former is just a licensed Samsung 14 nm, and Apple felt comfortable using both Samsung 14 nm and TSMC 16 nm interchangeably, albeit an earlier generation of each. The biggest differences that people were able to find between Apple's dies were that the dies from Samsung were several percent smaller and resulted in devices with a few percent less battery life.
There are a ton of measurement numbers relevant to any process node. Foundries basically pick the smallest number and use that as shorthand to describe the entire process node. There is a strong case to be made that Intel's 14 nm process node is the best, though, on the basis of sizes for the metal layers.
If the 1080 and 1070 have taught us anything about "FE Pricing" - that really means everything will cost that until... whenever the retailers would have normally discounted the cards anyway.
Also, reviews aren't out yet (everyone got cards, but no drivers)... but going off nVidia's claim of being "Faster than a 980" (which I read to be about the same as the 980 give or take a bit)....
At a $300 price point, and even at the $250 discounted price point (if it makes it to that level anytime soon) - that makes the AMD 480 still a very relevant card. At $250, It doesn't necessarily overprice the 1060, although at $300 I think it's a tough sell and makes the 480 look that much better. There wasn't such a wide gulf between the 970 and 980 in the first place, and you can essentially replace that with the 480 and 1060 (assuming the benches from the 1060 reviews come in line with nVidia's claims).
nVidia could have killed AMD here with one swift stroke; had they priced this at ~$225 out the gate with some measure of availability and supply, then the 480 would have been made completely irrelevant. Maybe this is another case of the price point reflecting the anticipated supply...
The specs don't look so hot. It will be marketable if it can have a 15% performance advantage over the RX 480. It could reach there with the higher clock, but it has less compute power and bandwidth. So only tests will give us a clear picture on the difference.
This will likely be the generation I see red again. Nvidia has seen enough of my green. 300 is a bit much for a middling card that looks like it is going to have issues with APIs in the not to distant future. The 1070 and 1080 will power through just on raw specs, but unless DX12 performance improves dramatically the 1060 looks like it will fall behind rather quickly unless it flat trounces the RX480 straight out of the gate by a large margin. The higher price makes that even more imperative.
Pricing it where they did also means that other manufacturers are going to have a much smaller profit window once they add better coolers and shit unless it blows the RX480 away and they can gouge. Otherwise, outside of the typical Ford vs Chevy type hardcore supporters, it won't be a good pick in the largest monetary segment of the video card market.
Between the spotty driver quality over the last few years and the DX12 early results, Nvidia is now looking like the behemoth that isn't agile enough to keep up with moving targets.
So, yeah this is a little interesting, its basically half of a 1080, but with a 192bit bus, so it's got a lot more memory bandwidth in respect to that, according to Anandtech:
Diving into the specifications then, GTX 1060 and its GP106 GPU ships with 1280 CUDA cores, which is half the number found on GTX 1080/GP104. Similarly, this means we’re looking at half the texture units, and half of the polymorph geometry engines. At this point NVIDIA has not provided an architecture diagram for GP106, so I don’t know how NVIDIA has laid out the internal working of the card, but we’ll be looking at 10 Pascal SMs in some configuration.
However on the backend of the rendering pipeline, things are a bit more interesting NVIDIA deviates a bit from tradition, and a bit from making GP106 a true halving of GP104. Whereas you’d expect half of a GP104 to ship with a 128-bit memory bus, NVIDIA has defied expectations by giving GP106 a larger 192-bit memory bus, giving the chip 50% more memory bandwidth per CUDA core, all things held equal. NVIDIA’s no stranger to 192-bit memory buses – in fact GK106 had one – but after Maxwell, this comes as a bit of a surprise.
....
In any case, the 192-bit memory bus also means that we may see an increase in the number of ROPs on the card. NVIDIA has not disclosed the ROP count, but Pascal’s internal ROP/L2/memory controller partition design means that we’re going to be looking at 6 partitions, which at 8 ROPs per would work out to 48 ROPs. If this is the case, then not only would GTX 1060 be receiving a significant boost in memory bandwidth versus its predecessor, but it would be receiving a significant increase in raw pixel throughput as well.
...
In terms of performance, NVIDIA hasn’t published any internal benchmarks, but they have been rather explicit that they’re targeting GTX 980 performance. Given what we’ve seen of Pascal so far, there’s good reason to believe that a 1280 CUDA core GPU should be able to come close, though it’s going to depend on the game. My best guess is that baring more extensive power throttling, GTX 1060 should be able to come close, though less so in games that are very shader/texture heavy.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
It's got less compute than a GTX 980, less memory bandwidth, less of basically everything. So of course Nvidia is claiming it's faster than a GTX 980. Color me skeptical. It might well be faster than an RX 480 much of the time, or faster than a GTX 970 nearly all of the time, but a GTX 980? Doubtful.
It will be clocked very high on stock (just like 1070/1080) to appear to be "as fast as", but as soon as you start overclocking both.....1060=970 (just like 1070/1080)
Maxwell was clocked very low on stock and it had huge OC potential, while Pascal is clocked very high on stock with very small OC potental.
It's got less compute than a GTX 980, less memory bandwidth, less of basically everything. So of course Nvidia is claiming it's faster than a GTX 980. Color me skeptical. It might well be faster than an RX 480 much of the time, or faster than a GTX 970 nearly all of the time, but a GTX 980? Doubtful.
It's also clocked what, 70% higher?
Let's not be intellectually dishonest here. Obviously we have to wait for results, but its not outside the realm of possibility.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
It's got less compute than a GTX 980, less memory bandwidth, less of basically everything. So of course Nvidia is claiming it's faster than a GTX 980. Color me skeptical. It might well be faster than an RX 480 much of the time, or faster than a GTX 970 nearly all of the time, but a GTX 980? Doubtful.
It's also clocked what, 70% higher?
Let's not be intellectually dishonest here. Obviously we have to wait for results, but its not outside the realm of possibility.
TFLOPS is number of shaders times clock speed times 2. So that takes clock speed into account. And the GTX 980 has 14.4% more TFLOPS than the GTX 1060.
And as we all know TFLOPS has never been a particularly accurate indicator of performance for anything other than raw compute tasks like CUDA, bitcoin mining, etc.
Again, all I'm saying is it's not outside of the realm of possibility. There are other factors. Nvidia has already claimed they're 15% faster than a 980 with it, which in all likelihood means they've cherry picked one or two benchmarks that happen to show that. As to the overall results, my guess is it won't be quite as fast as a 980, but close enough that it should be a hit.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
And as we all know TFLOPS has never been a particularly accurate indicator of performance for anything other than raw compute tasks like CUDA, bitcoin mining, etc.
Again, all I'm saying is it's not outside of the realm of possibility. There are other factors. Nvidia has already claimed they're 15% faster than a 980 with it, which in all likelihood means they've cherry picked one or two benchmarks that happen to show that. As to the overall results, my guess is it won't be quite as fast as a 980, but close enough that it should be a hit.
When you're comparing across wildly different architectures, yes, TFLOPS isn't very good as a sole measure of performance. If card A has double the TFLOPS of card B, while card B has double the memory bandwidth of card A, either one could be twice as fast as the other depending on where your bottleneck is, or anywhere in between.
But when comparing different cards of the same or very similar architectures, such that if card A has double the TFLOPS, it probably also has double the global memory bandwidth, double the local memory bandwidth, double the capacity of various caches, and so forth, TFLOPS is a pretty good single proxy of performance. Of course, so would lots of other things. If card A is double card B at everything, pick any one of those things and use that as your proxy and it will predict--and probably accurately--that card A is twice as fast as card B.
And that's the problem here for the GTX 1060. Based on the specs Nvidia listed, the GTX 980 has more global memory bandwidth, local memory bandwidth, local memory capacity, register capacity, TFLOPS, raster engine performance, tessellation unit performance, and basically everything else. I could believe that the GTX 1060 is faster in some cherry-picked corner case, but that's like saying that the RX 480 is faster than a Fury X if you create a corner case where the hardware-accelerated primitive discard becomes hugely important. Yeah, the RX 480 is faster in that corner case, but that's not typical performance.
I wouldn't rule out a 1060 just out of spite yet - if 480 has supply issues and 1060 does not, we could see the prices change a good deal.
I don't expect that to happen.. but it's remotely possible.
I'm also looking for a new card for Wife's computer, it'll either be a 480 or a 1060 later this summer, it all depends on availability and how the prices/performance shakes out.
It's got less compute than a GTX 980, less memory bandwidth, less of basically everything. So of course Nvidia is claiming it's faster than a GTX 980. Color me skeptical. It might well be faster than an RX 480 much of the time, or faster than a GTX 970 nearly all of the time, but a GTX 980? Doubtful.
Well the 1060 is still half the price of a 980 so even if it is not as fast who cares unless you are running at real high resolutions. I have a 960 because I did not see enough of a need for anything faster for the games I play.
It's got less compute than a GTX 980, less memory bandwidth, less of basically everything. So of course Nvidia is claiming it's faster than a GTX 980. Color me skeptical. It might well be faster than an RX 480 much of the time, or faster than a GTX 970 nearly all of the time, but a GTX 980? Doubtful.
Well the 1060 is still half the price of a 980 so even if it is not as fast who cares unless you are running at real high resolutions. I have a 960 because I did not see enough of a need for anything faster for the games I play.
To be fair, that's the only GTX 980 I could find near that cheap. Otherwise, they're around $400--which is high enough to be irrelevant when a GTX 980 Ti is also around $400.
A GTX 960 is basically half of a GTX 980, and the 1060 never doubles the 960. I could believe that the GTX 1060 is often or even typically faster than a Radeon RX 480, though with how narrow some of those wins are and the likelihood that Nvidia cherry-picks benchmarks, it would be surprising if the RX 480 doesn't also pick off some wins over a GTX 1060.
The bottom line is it won't be better, or cheaper than the gaming versions of the RX 480 due out in a couple weeks.
AMDGPU is shaping up to be pretty good on Linux. Freesync monitors are popping up everywhere. AMD also has all 3 next gen consoles as well so performance for future titles should be very good.
I don't see any reason to buy NV at this point unless you are getting a monster card. And for that I would wait for Vega
Comments
so my question is
'do you think this will last for 90 more days?'
Please do not respond to me, even if I ask you a question, its rhetorical.
Please do not respond to me
If it's performance match the RX 480 in DX11 games then it needs a clear distinctive feature that separates itself from that card to be priced over $200. AMD cards just perform better in DX12, and you can't reasonably assume fanboy-ism to maintain demand for the card.
I expect a straight die shrink of something like the 970 where the power draw is the touted feature of the card. However, it's really tough to say if it can draw less power than the RX 480. While nVidia's architecture draws less power than GCN, its also on 16nm verse AMD on 14nm.
There are a ton of measurement numbers relevant to any process node. Foundries basically pick the smallest number and use that as shorthand to describe the entire process node. There is a strong case to be made that Intel's 14 nm process node is the best, though, on the basis of sizes for the metal layers.
http://semiaccurate.com/2016/07/06/nvidias-gp104-based-gt1060-real/
Guess we'll see tomorrow if anything gets announced.
(249)/299$ for 6 GB model confirmed.
NVidia has totally popped, 300$ for reference 1060 rofl
We'll see if 1060 is any different, but ...
Also - reported to be GP106 by nVidia.
At a $300 price point, and even at the $250 discounted price point (if it makes it to that level anytime soon) - that makes the AMD 480 still a very relevant card. At $250, It doesn't necessarily overprice the 1060, although at $300 I think it's a tough sell and makes the 480 look that much better. There wasn't such a wide gulf between the 970 and 980 in the first place, and you can essentially replace that with the 480 and 1060 (assuming the benches from the 1060 reviews come in line with nVidia's claims).
nVidia could have killed AMD here with one swift stroke; had they priced this at ~$225 out the gate with some measure of availability and supply, then the 480 would have been made completely irrelevant. Maybe this is another case of the price point reflecting the anticipated supply...
And yeah, its GP106.
If they decide to sell GP104 as "almost half chip" it will either be "1060ti" or, more likely, "1070M" laptop part
Pricing it where they did also means that other manufacturers are going to have a much smaller profit window once they add better coolers and shit unless it blows the RX480 away and they can gouge. Otherwise, outside of the typical Ford vs Chevy type hardcore supporters, it won't be a good pick in the largest monetary segment of the video card market.
Between the spotty driver quality over the last few years and the DX12 early results, Nvidia is now looking like the behemoth that isn't agile enough to keep up with moving targets.
So, yeah this is a little interesting, its basically half of a 1080, but with a 192bit bus, so it's got a lot more memory bandwidth in respect to that, according to Anandtech:
http://www.anandtech.com/show/10474/nvidia-announces-geforce-gtx-1060-july-19
Diving into the specifications then, GTX 1060 and its GP106 GPU ships with 1280 CUDA cores, which is half the number found on GTX 1080/GP104. Similarly, this means we’re looking at half the texture units, and half of the polymorph geometry engines. At this point NVIDIA has not provided an architecture diagram for GP106, so I don’t know how NVIDIA has laid out the internal working of the card, but we’ll be looking at 10 Pascal SMs in some configuration.
However on the backend of the rendering pipeline, things are a bit more interesting NVIDIA deviates a bit from tradition, and a bit from making GP106 a true halving of GP104. Whereas you’d expect half of a GP104 to ship with a 128-bit memory bus, NVIDIA has defied expectations by giving GP106 a larger 192-bit memory bus, giving the chip 50% more memory bandwidth per CUDA core, all things held equal. NVIDIA’s no stranger to 192-bit memory buses – in fact GK106 had one – but after Maxwell, this comes as a bit of a surprise.
....
In any case, the 192-bit memory bus also means that we may see an increase in the number of ROPs on the card. NVIDIA has not disclosed the ROP count, but Pascal’s internal ROP/L2/memory controller partition design means that we’re going to be looking at 6 partitions, which at 8 ROPs per would work out to 48 ROPs. If this is the case, then not only would GTX 1060 be receiving a significant boost in memory bandwidth versus its predecessor, but it would be receiving a significant increase in raw pixel throughput as well.
...
In terms of performance, NVIDIA hasn’t published any internal benchmarks, but they have been rather explicit that they’re targeting GTX 980 performance. Given what we’ve seen of Pascal so far, there’s good reason to believe that a 1280 CUDA core GPU should be able to come close, though it’s going to depend on the game. My best guess is that baring more extensive power throttling, GTX 1060 should be able to come close, though less so in games that are very shader/texture heavy.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
- Friedrich Nietzsche
Maxwell was clocked very low on stock and it had huge OC potential, while Pascal is clocked very high on stock with very small OC potental.
Let's not be intellectually dishonest here. Obviously we have to wait for results, but its not outside the realm of possibility.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
- Friedrich Nietzsche
And as we all know TFLOPS has never been a particularly accurate indicator of performance for anything other than raw compute tasks like CUDA, bitcoin mining, etc.
Again, all I'm saying is it's not outside of the realm of possibility. There are other factors. Nvidia has already claimed they're 15% faster than a 980 with it, which in all likelihood means they've cherry picked one or two benchmarks that happen to show that. As to the overall results, my guess is it won't be quite as fast as a 980, but close enough that it should be a hit.
"The surest way to corrupt a youth is to instruct him to hold in higher esteem those who think alike than those who think differently."
- Friedrich Nietzsche
But when comparing different cards of the same or very similar architectures, such that if card A has double the TFLOPS, it probably also has double the global memory bandwidth, double the local memory bandwidth, double the capacity of various caches, and so forth, TFLOPS is a pretty good single proxy of performance. Of course, so would lots of other things. If card A is double card B at everything, pick any one of those things and use that as your proxy and it will predict--and probably accurately--that card A is twice as fast as card B.
And that's the problem here for the GTX 1060. Based on the specs Nvidia listed, the GTX 980 has more global memory bandwidth, local memory bandwidth, local memory capacity, register capacity, TFLOPS, raster engine performance, tessellation unit performance, and basically everything else. I could believe that the GTX 1060 is faster in some cherry-picked corner case, but that's like saying that the RX 480 is faster than a Fury X if you create a corner case where the hardware-accelerated primitive discard becomes hugely important. Yeah, the RX 480 is faster in that corner case, but that's not typical performance.
거북이는 목을 내밀 때 안 움직입니다
I don't expect that to happen.. but it's remotely possible.
I'm also looking for a new card for Wife's computer, it'll either be a 480 or a 1060 later this summer, it all depends on availability and how the prices/performance shakes out.
http://www.newegg.com/Product/Product.aspx?Item=N82E16814127901
How much is a GTX 1060, again?
To be fair, that's the only GTX 980 I could find near that cheap. Otherwise, they're around $400--which is high enough to be irrelevant when a GTX 980 Ti is also around $400.
http://videocardz.com/62122/nvidia-geforce-gtx-1060-official-performance-leaked
Not really 980 numbers.
AMDGPU is shaping up to be pretty good on Linux. Freesync monitors are popping up everywhere. AMD also has all 3 next gen consoles as well so performance for future titles should be very good.
I don't see any reason to buy NV at this point unless you are getting a monster card. And for that I would wait for Vega