The word is that Vega implements a High Bandwidth Cache (HBC) and Controller. According to PCper.com: "The controller, likely a part of the silicon itself, is responsible for talking to the cache and any other memory systems available to it. AMD is still being very vague about what these other options will be and different cards built for different markets will likely have different configurations. In the diagram examples AMD lists NVRAM (flash essentially), network storage and primary system memory. Of those three, all of them have very different latency, capacity and bandwidth characteristics that could be balanced to provide the best possible experience for a particular workload."
This could be AMD's design engineers showing their genius if the implementation is well executed. The latencies involved in anything across the system I/O controller are likely too large to be of much use, at least initially; with fiber though ... yeah. RAM and NVRAM (when that becomes a consumer product) however being available to the GPU is a big, BIG deal in machine learning. How much of this is relevant to gaming is difficult to say. Most games don't use more than a fraction of the VRAM allotted on the 8 GB cards because the developers have to design for the market. The market is diverse to say the least. How all this evolves on Ryzen and Naples with the Infinity Fabric will be really interesting to watch. Exciting times. I will be getting Vega certainly with every intention of skipping over NAVI. Unless I get a paying gig in deep learning.
Stand by what I say, amd is going to need much more than just being a bit cheaper and a bit better.
people with gsync will stay nvidia
people who use cuda will stay nvidia
people who buy into gameworks as an experience enhancer (these people do exist...) will stay nvidia.
If they had more success stories like doom / vulcan, where a fury-x outperforms a 1070 in both fps and frame time delivery due to async compute being implemented properly then I would say amd would have a chance.. but I can't see that becoming the norm anytime quick enough that it shows in the day one product benches done by reviewers at launch... and that counts for a lot as that influences / perpetuates the next hardware upgrade buying cycle.
I think reviewers removing GTA (due to the now known engine bugs that wendell and GN pointed out) will play in amd's favour as its a game that has always favoured nvidia.. and given its engine constraints SHOULD NEVER HAVE BEEN USED AS A HW TEST . (in their defence things only really came to light when hardware became 'too powerfull').
I am going to crystal ball gaze using the usual suspects that show up in benching.. we can revisit this at a later date and laugh over its 'accuracy' .
Odds are * Doom - can't see nvidia winning this round, if a furyx could beat a 1070... * Ashes will still be one of the main titles benched (even though hardly anyone is still playing it..) - most likely a win for amd. * Total Warhammer is another one - most likely a win for amd. * F1 - god knows why - don't know how this will fall, could go either way but in benches a 390x was only 15 fps less than a 1070 * Shudder.. FALLOUT 4 - not seeing amd win this round, not unless some performance enhancing fixes are made to the game engine inbetween now and launch. * Dues Ex - Going to say AMD * Sniper Elite 4 - again amd sponsored so will most likely favour amd * Hitman - Going to be amd I think * GOW 4 - A very good job on the pc port - cant see any shenanigans going on - going to say amd * BF1 - Not sure but most likely nvidia
My point being, if its launched this year with GTA removed from the benches (which sensible reviewers should be doing) and amd (hopefully) working on optimizations for Fallout 4 (they would be fools if they didnt as this is going to be a staple in benches for years to come) then I think things are going to be good for AMD as it was mainly these two titles which nerfed AMD in reviews.
What is more or less known is Vega is coming before August at least, will have up to 8gb hbm2, and 4096 SPs. Extrapolate from the performance of the 480 and we are looking at 1080 level of performance. That said, there is a second Vega card in the works that we don't really know anything about. IT could be something to bridge the gap between the 580 and the top end Vega card, or it could be faster than the Vega card we know about. We have no way of knowing. Regardless, AMD is always great with their pricing, so it might be the king of the hill, but it will definitely be priced to compete, regardless. IT will come when it is ready. AMD isn't holding onto it longer than they need to. That would be counter productive. It just isn't ready to ship quite yet and there is nothing anyone can do abotu it. So chill.
One thing I would ask people to do is be on the look out for games which show a brand of gpu (be it amd or nvidia) that do good in the canned game benchmarks but worse when actually benchmarked 'in game'.
Example being, Dues Ex built in benchmark vs in game actual fps / frametime bench. (just as an example - not actually claiming Deus Ex is doing this).
If you ever see that happening then this could be a sign of either amd or nvidia optimizing for the benchmark sequence... which is a bit of a no no and got BOTH of them into hot water not that long ago.
More info at computex or whatever it is on May 5th? I hope so, really would like to at least know the pricing structure. I'm going to assume there to be 3 Vega cards...
HBM2 Main Features over GDDR5 Summary * Lower Power Requirements * On die placement and reduced footprint (finally no more mega sized cards that can't fit ITX!) * Slightly better bandwidth (atm it seems to only do better then GDDR5 by 20 or so GBPS)
I don't think AMD is sweating too hard. Only a small percentage of people buy enthusiasts high level cards brand new. 480s racked in money for AMD, and the 500 series will probably sustain until whenever Vega releases. For instance I bought a new 290x back when they dropped to $240 when the 300 series launched.
Most all of the tech press fails to comprehend the most important reasons to value AMD now. AMD has changed their focus from a mostly proprietary development model to an mostly open collaborative model. It started with OpenGL and Mantle and now Vulkan, however the paradigm has grown to include OpenGPU. Open GPU is a collaborative effort between AMD and software designers to innovate the full space of opportunities of the GPU within and beyond gaming. AMD is also collaborating with Linux developers to enhance and expand Radeon drivers and application resources on the Linux desktop and Linux servers. This is far more important than the performance on any single graphics card. Buy Vega, buy Ryzen, and go open source. Tear down the walled gardens of Intel, Microsoft, and the "partners." Reward AMD financially for making business decisions that are beneficial to the consumer beyond any single product.
As it stands, a 460/560 which will be quite a bit faster are only $10-20 more. The value just isn't there. But even then the 460 is almost hard to recommend as the 1050 is similar in price and faster.
Your faulting the tech press that reviews performance for not appreciating industry things. The tech press isn't concerning Linux improvements and open source developments because that's not what concerns the audience they serve. They're looking at the performance of products, not the industry politics of the companies. The tech press isn't reviewing the company and their "tech poltiics", they're reviewing their product. Confusing the two is to fundamentally confuse and over-complicate the very simple reality of their job, to review products.
Point taken. I suppose my critique would be more fitting of some of the commentary and recommendations. To be honest, my take on AMD could be dead wrong in the next year. They are a corporation beholden to shareholder above all else. I am encouraged in the near-term when I listen to Mark Papermaster and Raja Koduri, not on a stage but in one-on-one interviews - way deep in the weeds on some of these technologies regarding present and future hardware/software capabilities related to scalability - and I see the improved, albeit baby steps, open source cooperation.
it's polaris with a 50% boost in power. the old 480's had a limit at 120 watts for the die and now they have 180. many of the old 400's cards couldnt support this voltage besides the gtr and the new ones can, and you can still flash your old 400's if they can take the power. unfortunately it's only a few % better then a 400 series, but hey they are the same price or close.
and ofc amd really needs the money. and this cost very little to do compared to a brand new architecture.
i dont see a problem. 80%~ or so of people are using cards that arent as good as the 580 by a significant margin and this is something those people can use and it's cheap. amd aimed at the low/medium end twice in a row and we arent their target audience. im ok with them ignoring me for a little bit. they'll get to the high end with a little more time and i rather give them time and get a perfect product then get another rushed one like ryzen which takes weeks of bug fixes, and maybe isnt as good as it could have been with more time.
