It is good to see that Hilbert frags, the picture you see of the abandoned building is a building in battlefield 4, it was a picture taken during bf4 development for game art, it means that Vega will be launched Luke Hawaii, it will launch the same day BF1 launches.
That would be 64 compute units vs 36, so more like 80% Also Vega 10 will probably have at least 64 ROP's
Was thinking Polaris 10 had 40CU's with 2560 cores. Even then you are looking at I'd venture to say 60% on the high end. Vega 10 will not have as high a base clock if history is any indication. GCN does not scale to bigger chips without having to drop frequency. But at 60% faster than 480 it should trade blows with 1080 and do better in DX12. Although Vega 10 looks like Fiji XT with improved efficiency and new tricks on a smaller node. Should be interesting. Still more curious about Vega 11.
I think you guys are describing Vega 11 and not 10. Unless they manage some power magic between now and when Vega ships, how they are going to manage doubling RX480 and keeping it under 300w? Furthermore, if Vega 10 is that, how are they going to have an even larger chip and keep it under 300w? And then what happens to the massive gap between the RX480 and this GTX1080 level Vega 10? Are they just going to cut Vega 10 by like half and ship it to fill that gap? It makes more sense to me that Vega 11 is the 4096 core chip and Vega 10 is something around a 1070. If you double a RX470 you get slightly higher then GTX1080 performance. RX470 is already somewhat bottlenecked by memory performance so HBM2 will increase that gap slightly further. That + clocking it a little higher + whatever improvements architecturally that Vega brings and I think it will be 10-15% faster then a 1080 maybe even faster.
Feels the same to me too. I actually believe that Vega is the first new design in this cycle. Vega is a new IP (9.0) with different libraries used to design it, and (for the first time in years), MANUAL tracing. Polaris was essentially a tweaked Tonga (IP from 8.0 to 8.1). If Vega 10 is 4096/64 with better-than-Polaris anti-bottlenecking, a 48/3072 configuration with GDDR5 would be fine and probably save AMD a ton of money on top.
I thought the whole point of Vega was HBM2? If they don't use HBM2 then what's the point? Also wasn't Polaris ip 9?
My bad, I thought Vega 10 was the bigger chip. So were probably looking at 48-56 compute units with 64 ROP's. Vega 11 with 64 compute units, HBM2 and more async units.
"AMD Stock Could Explode If Vega GPUs Arrive In 2016" http://amigobulls.com/articles/amd-stock-could-explode-if-vega-gpus-arrive-in-2016 Did nVdia actually delayed Volta?
Nvidia hasn't given any time frame for Volta. There are slides from several years ago that put it around 2017/2018 but they're from 2014. I've mentioned before that I expect to see Volta in 2017 in HPC form -- that's going into the Oak Ridge/Sierra clusters. I don't know if we will see consumer Volta cards in 2017 though. I guess more recent rumors are saying 2H of 2017. But we all know how accurate rumors are with release dates. Honestly I wouldn't even be surprised if Nvidia completely split it's architecture for Gaming/HPC. They kind of already are with Pascal, the GP100 has a completely different CUDA core, memory architecture and communication bus. Nvidia's HPC revenue doubled since last year and they expect it to double again by next year and I think that estimate is probably low. There was an AMA on reddit today with an engineer that designs/develops prosthetic limbs and brought up how neural networks are going to be absolutely huge going forward in medical field. Eventually I think Nvidia will find it necessary to just split it entirely -- especially if the design of one market is going to come at the cost of another. Polaris is IP8 but I think people are looking too much into it. Basically anything they add that requires exposure on the driver level can potentially bump the IP level. For example, if they go Nvidia's route with FP16 doubling -- boom IP level increased and yet it won't effect gaming at all. I'm sure Vega will have some improvements but there are people who think it means that it will be a complete architecture revamp. It's possible but unlikely. Anyway stuff like this is generally how "over-hyping" happens.
Based on that article, I estimate between GTX1070 and GTX1080 with some DX12 games nearly GTX1080 performance except Tomb Raider where Pascal still wins. DX11 however might be slower than GTX1070 like Far Cry Primal, or faster like The Division. Nice enough performance. All comes down to price now. If they can do it for £450-500 it'll be very nice. If it's closer to £600, then, might be better off with GTX1080. I'm assuming the 20-30% performance boost is in comparison to FuryX. Denial; I think it's easy for us to forget how Nvidia surprised us and how quickly Pascal actually came out and the rumours were that it wouldn't be ready for another quarter or 2 (6months).
Considering what card I now have might skip Vega hoping for it to be awesome anyway tho. Unless AMD delivers Titan X Pascal level of performance with the bigger card then I could jump that. I am just gonna talk about big Vega since people seem to be confused about the numbering because it is the other way around vs polaris.
Well let's hope it's on time and not delayed , i'm on the waiting game for a new video card. I shall wait for Hilbert's review , until then we can only speculate the performance of this video card.
