But I believe it doesn't have double data rate, merely it can natively address 16bit registers relieving register pressure but not.provode the full speed up
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Poll: The Vega Review Thread.
- Thread starter Kaapstad
- Start date
Because they are paid to, hardly surprising.
Like nvidia dont pay and throw devs at them to make it easier to get gameworks into it so deep it lowers amd gpus performance and to even change render engines to make gameworks actually work. Who ye think has the most cash, nvidia or amd ?
I can't remember which video I watched recently (British guy) but I thought he was absolutely correct in what he said regarding Vega was made for features that 'could' be implemented by developers which was the wrong way to do things. People want performance now, not the hope of extra performance in the future. No idea what the AIB cards will go for but unless they can reduce the power draw by 150-200 watts at a higher clock speed Vega just isn't worth it for a gamer. Only raising this due to the people on the last few pages saying *if* developers implement this and this etc it will be better but AMD doesn't have the overall marketshare to necessitate that effort. By the time some of those features are the norm we will be a 1-2 generations further and again nvidia will just be better at raw power + those features. And with the price of Vega, AMD don't have the price/performance advantage anymore, let alone TDP.
Your probably on about adorded, but there is only really one feature in vega that requires devs to impliment. That being rapid packed math. Everything else is done by the gpu itself, what jim was on about more then anything with vega is the fact that for the full potential of vega it needs vulkan api tobe implimented.
I don't think it was adored but could be wrong, but as Microsoft is now pushing DX12 (Due to XB1X) I can't help but feel there will only bew a few Vulkan titles with the vast majority being DX12+ and AMD will stag lag behind in most games.
Well the point of dx12 is that its slim line, no legacy, that means there is next to no intel or nvidia legacy stuff in there. Bill gates himself has spoken out about nvidia and intels practices with backloading direct x. Thats why in a lot of dx12 games if its implemented right then amd can do just as well as nvidia in them and you do see it in the benchmarks when people say well nvidia is better in dx11 and amd run better in dx12. And its not an and or with vulcan and dx12, games can run both api, its simply a tab option. And the true power of vulkan is that its open source, ye cant hide nish in it. The main problem amd face when it comes to gpus isnt the apis anymore unless we are looking at dx11 and below but gameworks.
And Gameworks is the thing that the new features if you exclude rapid packed math are there to combat. All that tesselation in hairworks, rayworks and shadows etc, thats what primitive shaders, primitive discard, hbcc etc are built for. To remove that excessive tesselation can be ignored, that the bulk of the effects can be counter acted before they even hit the pipeline.
HBCC seems like an utter waste of time to me gaming wise though can be useful for compute - VRAM progress is vastly outstripping gaming type application developer progress and in the main becoming cheaper to throw lots of it at a GPU and in a scenario where you truly had vast worlds you'd be streaming data from disc not system memory and you'd want all the system memory possible for the non-visual data about the world - there are already ways to efficiently page streaming assets from disc - games like RAGE are held back more by having to support legacy hardware as well as bleeding edge and the logistics of distributing 100s of GB or more than technical progress at the hardware level.
Hbcc has many advantages, it allows to offload not so quite needed data and then call it fast when its required, also an oddity is when you turn hbcc on even when its not being used it does give a small increase which makes some think that with it being attached to ram that the infinity fabric in vega works the same as ryzen as being dicated in speed via the rams speed. But going forward the key to hbcc is that as games use more than 8gbs of vram that vega has no upper limit on vram. Thats why its pointless selling the same gpu with different vram capacity.
Then if we look at some of the upcoming games, final fantasy looks like its heavy on effects and stuff like that which will eat up vram. then you have games like star citizen where its massive open world like we have never seen before so the more vram you have the better, the more efrfects you can do. And loading data from ultra fast memory like system ram and nvme drives is way way way faster the dragging it from normal storage and its a direct connect between vram and its hbcc connected memory, its got no stops between the data and vram.
And then we have to think of the future, crossfire can use mismatched cards. This means that the 7nm vega or even navi comes out, you get that, you move vega down a spot, hbcc on both those cards can use the vram on vega even if the games dont use crossfire.
Last edited:
You'd still be bottlenecked by the available PCI-e bandwidth - with any serious transaction bulk you'll get the same performance as current virtual memory techniques.
The only saving grace for it would be if you could mount an nvme card directly to the GPU but then until nVidia implemented something similar it would sit there unused.
The only saving grace for it would be if you could mount an nvme card directly to the GPU but then until nVidia implemented something similar it would sit there unused.
And thats where the rest of the ecosystem comes into play by having gpu, system ram and 1 nvme device going right through ryzen, and going through ryzen is a lot faster then going through the chipset to cpu then to gpu.
And amd already have a gpu with a nvme right on it, its only required for huge bulk workloads.
Edit, also it always makes me chuckle, when amd do something like this then its well meh, its nothin, nvidia do something and its the second coming
And amd already have a gpu with a nvme right on it, its only required for huge bulk workloads.
Edit, also it always makes me chuckle, when amd do something like this then its well meh, its nothin, nvidia do something and its the second coming
Last edited:
Didn't see that. Well it turns out that Gibbo was not lying
This part of their statement isn't very smart "While the cost of the chip is high right now, there is enough margin in place for them to sell at $499 should they choose to." , I don't know if this is a direct quote from AMD or some author liberties taken...
Because AMD can't say that the e-tailers have enough margins, AMD have no idea how much the said e-tailer needs to earn to function, just like the e-tailer doesn't know how much profits AMD have to make, this could be a situation where AMD think the profit is enough, but it really isn't, the profit might only just cover operating costs/overheads. I just find it a bit disingenuous to make such statements for obvious reasons...
Doesn't make much odds while you still have the PCI-e 3.0 bus in play - as soon as you have any significant intensity of transactions needed you'll see a massive slowdown from regular VRAM.
Edit, also it always makes me chuckle, when amd do something like this then its well meh, its nothin, nvidia do something and its the second coming
nVidia have something similarish:
No one has got excited by it as the usefulness for gaming is kinda of meh in reality and the main area it will see use is compute.
Last edited:
And ye know what, 1080ti has what, 12gb of vram, when thats full; what ye gonna do apart from buy a new gpu and its not like we are even close to saturating pciex16, there is hardly any impact going from x16 to x8. And the reason why no one has gotten excited about it from pascal is cos pascal is a big thing, it goes all the way from a 1060 all the way to a quadro gp100 hbm2 which is the one that can do that and that thing costs 7k.
Last edited:
A lot of chat about features and capabilities being unused but the real facts are people here wanted a GPU to give them a great Amd alternative to the 1+yr old Nvidia 1070/1080/ti hardware but they got something that was barely capable of matching it.
Amd could have been best bang for buck contender but Vega ended up more expensive than Nvidias aging generation. This is more compounded by the ridiculous power draw and potential costs of a Psu upgrade.
I don't think Vega is a bad product per say , I just think it's lost in the market when Nvidia have the 1080.
Amd could have been best bang for buck contender but Vega ended up more expensive than Nvidias aging generation. This is more compounded by the ridiculous power draw and potential costs of a Psu upgrade.
I don't think Vega is a bad product per say , I just think it's lost in the market when Nvidia have the 1080.
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,471
its even worse than that, as Gibbo has said his buying price is already over $499, so there is no way he could meet that as a sales price as it would be below cost
Hardware unboxed is an excellent channel. Below they have benched every Vega variant.
I agree, he is starting to become really good. And amdmatt can i ask you, what is going on with some of those dx12 results, i know gamesworks can do some damage but how much of dx12 do some of these companys have to impliment to have the dx12 stamp on them as it seems some of these dx12 tests resemble very closly to the dx11 results, as if its mainly dx11 and not so much dx12.
he was running it against cards he had, as an aussie and not the biggest channel i dont think he has access to everything, but he did do it against custom cards and not just the founders. the video has more results and so does his website.