Are they announcing all specs and prices July 7th? Or are we waiting again until the week after to see everything ?
That's better since AIB 480's won't be out yet. Depending on what they price this I may be interested in this. Since I'll be getting a high end when Vega or the next iteration comes through, as long as it's reasonable.BenchLife says NDA lifts on July 7, so all info and possibly reviews will be out.
Guys, The graphs in the OP is wrong, this is the new one from Nvidia.
But honestly, it should be like this
How did you get to 1.63xIf Nvidia's graph is to be trusted, then the perf/watt of the GTX 1060 is going to be huge!
Based on my calculations it's should be 165% better than the 480. (1.15 * 1.43)
This is almost in line with the perf/watt of the GTX 1070/1080.
No, if the graph is absolute performance, as I assume, then you have to take both into account.
I figure it this way. Performance is 1.15 x at ~ 75% of the power, based on TDP. So the performance per watt is 1.15/0.75 or slightly more than 50% better.
Again, who knows, the slides could be totally fake. I'm just hazarding a guess that 163% makes more sense than 143%.]
GTX 1060 vs. GTX 1080
50% shader power
75% throughput
66% bandwidth
60% performance of a GTX 1080 seems to be right in line with those specs. It should be faster than RX 480 in everything (including AOTS), except Hitman which is very, very AMD friendly. I'm still hoping for a $249 6gb price, but knowing Nvidia it'll be $279. $299 would be absurdly close to GTX 1070.
GTX 1060 vs. GTX 1080
50% shader power
75% throughput
66% bandwidth
60% performance of a GTX 1080 seems to be right in line with those specs. It should be faster than RX 480 in everything (including AOTS), except Hitman which is very, very AMD friendly. I'm still hoping for a $249 6gb price, but knowing Nvidia it'll be $279. $299 would be absurdly close to GTX 1070.
They are measuring against 150W, I'm on mobile right now but last time I checked graph numbers, they were on spot for this figure.We don't know what 480 TDP Nvidia are measuring this against. I rather doubt if it'll have been 160w as even AMD clearly weren't expecting it to be remotely that bad.
Stupid to speculate though, we'll see
I will show you it this way.
1280 CUDA cores is exactly 66.67% of GTX 1070.
GPU will have most likely same core clock, 48 ROP and 192 Bit memory bus.
So it also will affect performance of the GPU. How it will stack?
Although the GTX 1060 will be a 1080p card (just like the RX 480 is a 1080p card, despite being marketed as a 1440p card), I think the most valid comparison should be at 1440p to eliminate some potential CPU bottlenecks. That said, at 1080p the GTX 960 is 54% as fast as a GTX 980, and at 1440p GTX 960 52% as fast as a GTX 980. So cutting GM204 in half ALMOST resulted in exactly half performance. Since the rumors and "leaked" slides are indicating this isn't cut in half (60% bandwidth, 75% ROPs) it will fair better against GTX1080 than GM206 did against GM204. 60% faster is exactly in line with GTX 980 performance which also happens to be what the "leaked" slides indicate. So I think my prediction is pretty much spot on. It will 10-15% faster than RX 480 and the one and only game/benchmark it will lose in is Hitman.
Once again, I'd like to see a $250 price point. That would be competitive with the 8gb RX 480 given the likely general performance and vram delta between the two cards, but if Nvidia supply really is as constrained as some rumors state, then I don't see that happening unless Nvidia simply sees a need to slow down RX 480 sales.
1280 CUDA cores from 2048 is exactly 62.5%. 32 ROPS, 128 Bit memory bus, high core clocks, with over 1200 MHz boost clocks.
And the GPU performance was ~53% of GTX 980.
So no. Do not expect anything else than 55-65% of GTX 1070 performance from GTX 1060. With much closer to 60% of GTX 1070 mark.
Nah you're wrong on this one, but at least you're not too far off. It'll be 70-75% of a GTX 1070. Despite Nvidia talking up Pascal's memory bandwidth compression, GTX 1080 and 1070 are both still bandwidth starved, which is why you are seeing the cards with bigger memory overclocks getting nearly 1:1 in performance gains.... like this one: https://www.techpowerup.com/reviews/MSI/GTX_1070_Gaming_X/27.html ...an overclocked GTX 1070 beating a stock GTX 1080, which people thought would be impossible given the huge disparity in specs between the 1080 and 1070.
The proportionally larger ROP and bandwidth of the GTX 1060 will help it scale noticeably beyond the 1280 cores it has vs. the 1920 cores of the GTX 1070 and 2560 cores of the GTX 1080. It'll be 70-75% of a GTX 1070.
P.S. As pointed out, you have the GTX 960 specs wrong. It was 1024 cuda cores. Everything about the 960 was exactly half of the 980.
The problem with GTX 1080 is that it is ROPs bottlenecked. It has exactly the same amount of ROPs as GTX 1070, that is why it is possible to OC GTX 1070 to its levels.
GTX 1080 will not be able ever to achieve its full potential just because of the amount of ROPs. GTX 1070 on the other hand is able to achieve its full hardware potential.
About GTX 1060, I already have spoken. And my calculations are correct. 1280 CC is 66.67% from 1920 CC.
Yes, my bad, I was constantly under the impression it has 1280 CC.
That does not change that much my calculations. It SHOULD be around GTX 970, or slightly lower in techpowerup performance summary, depending on resolution.