I thought I'd start this post as I am waiting..extremely patiently for Vega! AMD very quiet.
So thoughts conjecture on performance, price, differences between vega 10 and 11, release date etc.
I thought I'd start this post as I am waiting..extremely patiently for Vega! AMD very quiet.
So thoughts conjecture on performance, price, differences between vega 10 and 11, release date etc.
I wouldn't say AMD have been very quiet. They've already given a lot of info about the high level tech:
http://www.anandtech.com/show/11002/...tecture-teaser
I'm keeping a close eye on it. I'm not overly sold on the concept of HBM2 - nothing against the tech, but the market and developers have to embrace it as well.
Take my current Fury X for example - the performance on the Ghost Recon: Wildlands beta was catastrophic on AMD architecture, and that's purely down to Ubisoft's lackadaisical approach to anything that's not team Green. Building to the market majority is understandable but HBM is so under-utilised that it's a terrible shame. Vega will have to showcase well to be adopted by developers, which will in turn mean strong sales (or vice versa) when the performance stats show AMD to be a strong competitor once more.
They have every capability of blowing NVidia clean out of the water with this. Ryzen showed their capabilities in one market; will be interesting to see what they make of this generation.
Isn't release date supposed to be in May? Meanwhile new RX should be out at the start of April.
HBM2 doesn't need developing for - it's literally just a different type of memory, it doesn't change the way you program at all. For developers it's no big a change than going from GDDR3 to GDDR5 - all the memory access is handled by the GPU.
The reason some games perform better on nvidia hardware (or indeed, on AMD hardware) is because they use special instructions and optmisations for the shader cores. Nothing to do with the memory.
Where things get more complex is with nvidia's Gameworks tools - they're proprietary and to a certain extent black-box, so it's very hard for AMD to optmise for Gameworks because they don't know exactly how it's going to compile and run code on their hardware. And Ghost Recon: Wildlands is indeed a Gameworks game - so the issues you were seeing were nothing to do with Fiji having HBM, and everything to do with nvidia's black-box developerment suite making it nigh-on impossible for AMD to optimise performance.
Judging by expected clockspeeds and SP's, provided that Vega CU's are at least as performant as Polaris ones at the same clock speed, Vega should in general be a little over twice as fast.
If that's true then it should trade blows with the Geforce 1080ti.
Of course, clock speed speculations are based on Public statements by AMD on performance in TFLOPS, the scaling with SP's isn't necessarily linear, nor is it with clock speed.
The again, Vega CU's could have higher IPC as well.
"In a perfect world... spammers would get caught, go to jail, and share a cell with many men who have enlarged their penises, taken Viagra and are looking for a new relationship."
I'm hoping Vega continues the AMD tradition of being good for rendering - http://www.anandtech.com/bench/GPU15/1231
There are currently 1 users browsing this thread. (0 members and 1 guests)