AMD Radeon Pro Duo Dual Fiji GPU Now Available, Built For VR

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

uglyduckling81

Distinguished
Feb 24, 2011
719
0
19,060


For a start you know nothing of what they are going to release. People always think a new release is going to change everything. The reality is it will a small improvement as always. In all likelihood most of the lower end will be refresh releases while yields are low. It's certainly not all doom and gloom for AMD like Nvidia fan boys would have you believe.
Hopefully next year both companies will have new releases on new nodes and competition will be healthy.
 

bit_user

Polypheme
Ambassador
28 nm -> 16 nm is not a small improvement.

And AMD has released roadmaps that don't show Vega until 2017. I've also seen leaked benchmarks & specs on Polaris 10 and 11. So, whatever you're claiming I didn't know, I think you're mistaken.

Anyway, the GTX 1080 specs are now public, and this card has (allegedly) been outclassed by one costing 40% as much. We won't know for sure, 'till we see the benchmarks, but it's pretty clear that AMD missed the market window with this card.

BTW, I'm no NV fanboy. I have only AMD cards, so far, because their OpenCL support is better and I don't like being locked into buying a G-Sync monitor. However, if I get a chance to do some work with VR, this year, I'm quite likely to go with a GTX 1080 ...unless AMD manages to surprise us.
 

uglyduckling81

Distinguished
Feb 24, 2011
719
0
19,060
Sandybridge to Skylake is a bigger node shrink. 5 years of extra development and its still only 10% - 20% max performance increase in best case scenario. Your dreaming if you think they are going to magically make double performance figures as they are claiming. It's the CO's job to inflate share prices and that's what all his over the top BS sales talk is.
 

bit_user

Polypheme
Ambassador
You're confusing serial vs. parallel performance improvements. With GPUs, node shrinks naturally lead to improved performance, since you can put more cores in the same die area. Don't take my word for it, go back and look at the performance curve of GPUs, over the past 10 years, vs CPUs. Moore's law only stopped for CPUs, because they already hardwired everything that made sense, could no longer keep turning up the clock speed, and more cores don't help their typical user like they do for GPUs.

There's always a certain amount of hype, but if you simply look at the relative numbers of sharder cores * clock speed ratio, then a roughly 2x performance improvement should be expected. Again, we'll have to see, but it's not going to be too far off from what he's saying.

I still don't really know what's your thesis, here.
 
Status
Not open for further replies.